Authors Title Abstract

The surface defects of printed circuit boards (PCB) that occur during the manufacturing process seriously affect
product quality. So it is important to detect PCB surface defects quickly and accurately. However, existing defect
detection methods still have room to be improved for PCB surface defect detection. This paper proposes an
advanced model, MDUA-YOLO, based on YOLOv5 to increase the detection accuracy of defects on PCB surface.
MDUA-YOLO: an Firstly, we introduce the C3 Mobile Vision Transformer (C3MobileViT) module in the backbone of YOLOv5, improving
advanced deep learning the feature extraction capability of the model. Secondly, the Deformable Convolutional-Receptive Field Block (DC-
Liu X., Wang H. approach for PCB RFB) module is incorporated into the neck of YOLOv5 to dynamically expands the receptive field and more accurately
surface defect capture the location information of small defects. Additionally, we design the Union Attention Block (UAB) module in
detection the neck of YOLOv5 to optimize the fusion of low-level and high-level feature maps. Finally, an extra prediction head
and new feature fusion layer are also added to enhance the ability of the model to detect small defects. On the
benchmark PKU-PCB and DeepPCB datasets, numerous experimental results show that the MDUA-YOLO surpasses
other comparison state-of-the-art models and meets the real-time detection requirement of industrial environment.
© The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.
In the production and manufacturing of Printed Circuit Boards (PCBs), defects can arise negatively impact to both
product performance and usability. To tackle the issue of false detections stemming from a high volume of small
defects and complex backgrounds in PCBs, we propose SRN_Net, an innovative small object detection framework
specifically designed for PCB defect identification scenarios. Firstly, an advanced Separated Global Context Attention
A novel PCB surface (SGC) mechanism is introduced to enhance the model’s attention to small targets and improve the detection
defect detection accuracy of the model for small targets. Furthermore, a Residual Context Aggregation (RCA) module is seamlessly
method based on integrated into the network’s neck, effectively attenuating the disruptive influence of irrelevant background noise
Zhu L., Zhao R. separated global during the fusion of small target features. Lastly, a No Stride Convolution (NSC) technique is deployed in both the
context attention to trunk and neck of the network and meticulously designed to enhance small target detection accuracy by minimizing
guide residual context feature loss during the convolution process. Extensive experiments on the PCB dataset demonstrate that, compared
aggregation to state-of-the-art algorithms, SRN_Net achieves increases of 1.1% in Precision, 1.3% in Recall, 0.6% in mAP@0.5,
and 4.6% in mAP@0.5:0.95, highlighting its superior performance in defect detection. To demonstrate the efficacy of
SRN_Net across other industrial datasets, we conducted an additional evaluation on the NEU surface defect dataset,
achieving an mAP of 75.8%. This work contributes to advancing small object detection accuracy and robustness in
practical applications. The code will be available at [Link] © The Author(s) 2024.
In the production and manufacturing process of electronic products, surface mount chip solder joints can cause
solder defects, which can seriously affect the quality and performance of electronic products. Traditional deep
A lightweight solder learning defect detection methods have poor detection capabilities for surface mount chip solder joint defects. To
joint defect detection address this issue, this research proposes a Lightweight Solder joint defect Detection Network (LSD-Net). Firstly, a
Ang L., Hamzah R.,
network combined with novel backbone network is proposed for feature extraction, which integrates the idea of structure re-
Bakar S.A.
enhanced residual parameterization, enhancing the feature extraction ability while also considering network computing complexity.
Mamba Secondly, An Enhanced Residual Mamba block based on the Mamba architecture is proposed. An enhanced vision
state space model is developed to augment the feature fusion capability of the module for both global and local
information. Finally, A Non local Atrous Spatial Pyramid Pooling block is proposed to improve the feature
 representation capability of small-sized targets by introducing non local operations to calculate the relationship
between the background and target features of feature maps at a global scale. This article conducts ablation
experiments and comparative experiments on the SMD solder joint defect dataset, and the ablation experiments
verify that each module can effectively improve the detection performance of the model. The comparative
experimental results indicate that, the detection precision, recall, mAP and F1 score are 91.5%, 95.5%, 92.8% and
93.5%, respectively. The model parameter size is 6.9M, which is better than the comparative detection algorithm
while ensuring a low number of parameters. The experiment results on the public dataset PKU-Market-PCB show
that this method has strong generalization ability. © 2025 Elsevier Ltd
In the manufacturing process of ceramic mugs, the detection of micro-surface defects faces technical challenges of
high difficulty and low efficiency, and efficient and high-quality production lines are crucial to maintaining market
competitiveness. The goal of this study is to improve the accuracy and efficiency of detection by developing a defect
detection algorithm and equipment based on deep learning, thereby improving product quality and reducing
production costs. The research uses the visual algorithm You Only Look Once version 8 (YOLOv8) in the field of
Artificial intelligence-
artificial intelligence as the baseline model. Firstly, a slice pre-training layer is designed to reduce the memory loss of
enabled defect
Mao W., Wu H., Xie S., large images to the graphics card. Secondly, the model structure is reconstructed to adapt to small target detection.
detection method and
Li L., Yang X. In addition, a mixed local channel cross-stage feature fusion module is proposed to enhance the recognition ability of
engineering application
small targets. Finally, a detection head with shared parameters is designed to further reduce the number of
of ceramic mug
parameters. In terms of engineering application, a set of test equipment was developed and the corresponding
software was written. Experiments show that the accuracy of the algorithm is 21.3 % higher than that of YOLOv8, and
the parameter amount is reduced by 67 %. Compared with manual detection, the equipment efficiency is increased
by 47.06 %, and the detection success rate is 99.6 %. Therefore, the research in this paper provides an efficient and
reliable solution for industrial automation detection. © 2025
Printed circuit board (PCB) is a crucial component of advanced electronic devices, and its quality control cannot be
overlooked. This paper presents a comprehensive review of machine vision-based surface defect detection methods
for PCBs, addressing the transition from traditional image processing to advanced deep learning techniques. With the
increasing complexity of PCB designs and the demand for high-precision manufacturing, automated defect detection
has become critical for quality control. The study introduces nine public datasets for PCB surface inspection and
fourteen common types of PCB surface defects, and provides an overview of commonly used performance
A comprehensive
evaluation metrics in the field of PCB defect detection. This paper systematically analyzes three categories of
review of research on
He Z., Lian Y., Wang Y., detection approaches: image processing-based methods, machine learning-based classifiers, and deep learning
surface defect
Lu Z. architectures. Furthermore, the review provides a comparative evaluation of representative works for each method,
detection of PCBs based
revealing that deep learning-based methods achieve state-of-the-art performance, while image processing methods
on machine vision
and traditional machine learning approaches remain valuable in resource-constrained scenarios. We believe that the
Transformer architecture has transformative potential in handling global defect patterns, while emphasizing the
need for lightweight model architectures through techniques like knowledge distillation. This paper also mentions
hybrid methods and ensemble approaches, which leverage the advantages of different detection techniques to
achieve defect detection, improve detection accuracy, and reduce the probabilities of false positives and false
negatives. Key challenges such as class imbalance, computational complexity, and detection of micro-defects are
 discussed, along with future directions including self-supervised learning, multi-modal learning, 3D defect detection,
and edge-cloud collaborative systems. © 2025
Defect detection plays a crucial role in chip production and is essential for ensuring the quality of electronic products.
However, existing methods still face challenges when detecting small-sized targets. To address this issue, we propose
a chip solder joint defect detection model, SSF-YOLO, based on YOLOv8. A lightweight feature extraction module,
StarNet-SimAM Block, is designed to enhance feature extraction capability while reducing the complexity of the
model’s feature extraction network, thereby improving detection efficiency. We also introduce a SCAM Fusion
A real-time detection module that incorporates a separate channel attention mechanism to fuse feature information from different levels
Ang L., Hamzah R., method for small-sized of the backbone network, thereby mitigating the loss of small defect features during feature fusion. A position
Quanxing W. solder joint defect regression loss function, NWD-CIOU, is employed to calculate the distance between the bounding box of small
detection targets and the ground truth, effectively reducing the missed detection rate for small defects. Experiments
conducted on two datasets, SMT chip solder joint defects dataset and PKU-Market-PCB dataset, demonstrate that
SSF-YOLO improves the mAP metric by 1.5% and 2.3%, respectively, compared with YOLOv8. The detection speed on
SMT chip solder joint defects dataset reaches 128 FPS, representing a 6.7% improvement over YOLOv8, indicating
enhanced detection accuracy and real-time detection capability. © The Author(s), under exclusive licence to
Springer-Verlag GmbH Germany, part of Springer Nature 2025.
Printed Circuit Board (PCB) manufacturing demands accurate defect detection to ensure quality. Traditional
methods, such as manual inspection or basic automated object inspection systems, are often time-consuming and
inefficient. This work presents a deep learning architecture using Faster R-CNN with a ResNet-50 backbone to
Automated Defect automatically detect and classify PCB defects, including Missing Holes (MHs), Open Circuits (OCs), Mouse Bites (MBs),
Weerakkody K.D., Identification System in Shorts, Spurs, and Spurious Copper (SC). The designed architecture involves data acquisition, annotation, and
Balasundaram R., Printed Circuit Boards augmentation to enhance model robustness. In this study, the CNN-Resnet 50 backbone achieved a precision–recall
Osagie E., Alshehabi Using Region-Based value of 87%, denoting strong and well-balanced performance in PCB fault detection and classification. The model
Al-Ani J. Convolutional Neural effectively identified defective instances, reducing false negatives, which is critical for ensuring quality assurance in
Networks PCB manufacturing. Performance evaluation metrics indicated a mean average precision (mAP) of 88% and an
Intersection over Union (IoU) score of 72%, signifying high prediction accuracy across various defect classes. The
developed model enhances efficiency and accuracy in quality control processes, making it a promising solution for
automated PCB inspection. © 2025 by the authors.
Aiming at the problems of low accuracy and large computation in the task of PCB defect detection. This paper
proposes a lightweight PCB defect detection algorithm based on YOLO. To address the problem of large numbers of
parameters and calculations, GhostNet are used in Backbone to keep the model lightweight. Second, the ordinary
A lightweight detection convolution of the neck network is improved by depthwise separable convolution, resulting in a reduction of
Yu S., Pan F., Zhang X.,
algorithm of PCB redundant parameters within the neck network. Afterwards, the Swin-Transformer is integrated with the C3 module
Zhou L., Zhang L.,
surface defects based in the Neck to build the C3STR module, which aims to address the issue of cluttered background in defective images
Wang J.
on YOLO and the confusion caused by simple defect types. Finally, the PANet network structure is replaced with the
bidirectional feature pyramid network (BIFPN) structure to enhance the fusion of multi-scale features in the network.
The results indicated that when comparing our model with the original model, there was a 47.2% reduction in the
model’s parameter count, a 48.5% reduction in GFLOPs, a 42.4% reduction in Weight, a 2.0% reduction in FPS, and a
 2.4% rise in mAP. The model is better suited for use on low-arithmetic platforms as a result. © 2025 Yu et al. This is
an open access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Industrial surface defect detection is crucial for maintaining product quality, but it faces challenges such as complex
background interference, numerous small defects, and significant variations in defect characteristics. To address
these challenges, this paper introduces a novel lightweight hierarchical aggregation task alignment network (LHATA-
Net) designed to enhance detection accuracy, computational efficiency, and generalization. LHATA-Net includes four
innovative features: (1) a fast-efficient layer aggregation network (F-ELAN) for efficient feature extraction; (2) a
hierarchical multiscale feature enhancement path aggregation network (HMFE-PAN) to improve detection of small
A lightweight defects in complex backgrounds; (3) a lightweight task aligned head (LTA-Head) to optimize feature interaction
Lv S., Liang T., Zhang hierarchical aggregation between classification and localization; and (4) a slide loss function (Slideloss) that integrates slide weighting function
K., Jiang S., Ouyang B., task alignment network with binary cross entropy with logits loss function to tackle sample imbalance. To better validate the detector, we
Li Q., Li X. for industrial surface compile a large-scale dataset, DsPCBSD+, which includes real images of surface defects on printed circuit boards from
defect detection practical industrial production. Experimental results demonstrate that LHATA-Net, with only 3.5M parameters and
18.4G floating point operations per second, achieves an inference speed of 54.2 frames per second. It also achieves
average precision of 79.6%, 70.0%, and 85.8% at an intersection over union threshold of 0.5 on two steel surface
defect datasets and the DsPCBSD+ dataset, respectively. It ranks first, second, and third compared to state-of-the-art
(SOTA) real-time detectors. The Friedman test confirms that LHATA-Net surpasses SOTA detectors in overall
performance, highlighting its superiority in practical engineering applications. The code is available at
[Link] © 2024 Elsevier Ltd
The demand for integrated circuit (IC) chips has risen markedly across various industries in conjunction with
advancements in global technology. Prior to packaging, IC elements undergo several processes, including wafer
dicing, wire bonding, and encapsulation. Scanning Acoustic Tomography (SAT) effectively analyzes the internal
structures of integrated circuit products, thereby preventing the supply of defective components, including chip
fractures, delamination, voids, and adhesion issues. The study aims to reduce operator eye strain, enhance
productivity, and minimize employee turnover rates by proposing the use of convolutional neural networks (CNN) to
Enhancing Integrated develop a predictive model for automating defect detection in integrated circuit (IC) products within SAT images,
Jou Y.-T., Putra V.P., Circuit Quality Control: replacing traditional visual inspections. To enhance the accuracy of the CNN model, we implement the flood-fill
Silitonga R.M., A CNN-Based Approach algorithm as the primary augmentation strategy and create an image augmentation model in Python. This method
Sukwadi R., Inderawati for Defect Detection in produces a training set that accurately reflects the true characteristics of defects, thereby mitigating the problem of
M.M.W. Scanning Acoustic limited defect data and ensuring the model is trained on reliable information. The incorporation of various rates and
Tomography Images scaling factors into SAT defect images, along with the manipulation of the original defect, contributes to the
development of a robust dataset suitable for real-world testing. The CNN model is trained using various batch sizes,
resulting in customized training datasets and predictive models to improve accuracy. Key findings indicate that
employing 40× augmentation alongside a batch size of 32 enhances the model’s performance, yielding a missed
detection rate below 0.4% and a false alarm rate of 0.1%. This model offers an improved solution to the issue of
manual inspections on assemblies, thereby alleviating operator stress and establishing a robust framework for
automated integrated circuit quality management. © 2025 by the authors.
 Tackling the widespread problems of inaccuracies, slow detection speed, and poor adaptability in small object defect
detection on PCB circuits, this study suggests a lightweight printed circuit board surface defect identification
algorithm, building upon an improved YOLOv8-PCB. This algorithm first introduces the C2f_SHSA attention
mechanism in the backbone network, which unites the merits of channel attention and spatial attention, facilitating
an efficient fusion of local and global features in a lightweight manner, thereby enhancing the model’s identification
preciseness for small defects. Subsequently, in the neck network, the C2f_IdentityFormer structure, which combines
the C2f structure with the IdentityFormer structure, supplants the initial C2f structure. This enhancement improves
A Lightweight PCB
the model’s sensitivity to subtle features and further optimizes the effect of feature fusion. Eventually, the PIoU is
Wang J., Xie X., Liu G., Defect Detection
presented to enhance the model’s adaptability to small, complex PCB defects with varying sizes and shapes, while
Wu L. Algorithm Based on
also accelerating the mode’s convergence speed. Experimental outcomes reveal that the improved YOLOv8-PCB
Improved YOLOv8-PCB
algorithm displays remarkable performance in the PCB dataset, with a Recall rate of 94.0%, a mean Average Precision
(mAP) of 96.1%, and an F1 score of 94.35%. Moreover, the model’s weight size is only 5.2 MB. Compared to the
YOLOv8n baseline model, the Recall rate has a 3.6% improvement, the mAP is raised by 1.8%, and the F1 score is
enhanced by 1.9%, while the model’s weight is reduced by 17.46%. The enhancements in performance metrics
confirm that the improved algorithm not only fulfills the requirements for efficient and real-time detection in PCB
surface defect identification tasks but is also better suited for deployment and operation on edge devices. © 2025 by
the authors.
Identification of PCB board defects early in the manufacturing process is crucial, as PCB quality control plays an
important role in the electronics manufacturing industry. Defective PCBs can lead to product failures, increased
costs, and reliability issues. The idea proposes a machine learning program for the detection and classification of
PCB Defect Detection
defects in PCBs using a YOLOv5 object detection algorithm and an Inception v3 CNNbased approach. The dataset
Kumari T., Raj T., Using Machine
consists of annotated PCB images for the YOLOv5 algorithm, and the images are categorized into six defect types:
Darahas P.S., Neelima Learning: YOLOv5 and
missing holes, mouse bite, open circuit, short circuit, spurs, and spurious copper. The preprocessed defect images are
N., Bhavana V. Inception v3 CNN-Based
analyzed using the YOLOv5 model for defect location, while the Inception v3 CNN model performs precise defect
Approach
classification. The results and the output clearly show that the proposed approach achieves high accuracy in both
detection and classification tasks. Hence, the aim is to provide an efficient and automated solution for PCB
inspection, outperforming conventional manual checks. © 2025 IEEE.
Detecting defects is crucial to ensuring the quality of printed circuit board (PCB) products. Due to the diminutive
nature of surface defects on PCBs, current detection algorithms struggle to extract small defect features accurately,
leading to a propensity for missed detections. To tackle these challenges, we propose a PCB defect detection
PCB Defect Detection algorithm that builds upon the YOLOv7 algorithm with enhancements. Firstly, we integrate the proposed CREC
Algorithm Combining module into the backbone network to enhance the capture of local features about minor defects. Secondly, we
Fan W., Peng J., Lan S. Cascaded-Wise Residual propose the integration of a multi-scale feature fusion module, SPPB, within the head network to selectively activate
and Dynamic Sparse channels or positions related to minor defects in the feature map, thereby enhancing the accuracy of local feature
Attention extraction for minor defects. Subsequently, the algorithm is endowed with higher efficiency in learning small defect
features with the help of a new loss function, MPNWD. Finally, a small target detection layer P2 is added to enrich
the contextual information in order to facilitate the algorithm to understand the relationship between small defects
and their surrounding regions. Experimental results demonstrate the effectiveness of the enhanced YOLOv7
 algorithm in testing the PCB_DATASET defect dataset, achieving a detection accuracy (mAP) of 99.3%, surpass-ing
YOLOv7 by 5.4%, and outperforming other algorithms in terms of detection accuracy. © 2025, Scipedia S.L.. All rights
reserved.
With the continuous development of technology, electronic devices have appeared in all aspects of our lives. Printed
circuit boards (PCBs), as key components of electronic devices, may have various defect problems such as open
circuits, short circuits and spurs, which can affect the use of electronic devices. Therefore, it is crucial to detect
surface defects of PCBs. This paper proposes a PCB defect detection algorithm based on the improved YOLOv9. A
new Median-enhanced Spatial and Channel Attention Block (MECS) is introduced in the feature extraction of the
PCB Defect Detection YOLOv9 backbone network. MECS is used to enhance the ability of YOLOv9 to capture PCB defect features, reducing
Chen S., Xing E., Hou X. Algorithm Based on the false positive and false negative rates of PCB defect feature recognition. At the same time, deformable
Improved YOLOv9 convolution V3 (DCNv3) is used to replace the conventional convolution in YOLOv9, which can better adapt to PCB
defects with different angles and shapes during PCB defect detection. In this experiment, an experimental data set
was constructed through network sampling and on-site collection, and a large number of experimental tests were
carried out. The experimental results show that the improved YOLOv9 achieves a recognition precision of 92.77% in
PCB defect detection, which is 2.41% higher than that of the traditional YOLOv9, providing a more efficient and
accurate solution for PCB defect detection. © 2025 IEEE.
In printed circuit board (PCB) production, uncertainty in the manufacturing stage and sampling uncertainty in the
inspection stage have a great impact on product quality. This article proposes rotating angles based on the optimal
Rotation-Angle-Based
detection of PCB defects under uncertainties. First, the principal features are extracted and factors of information
Principal Feature
concentration are designed to construct the reference template from the preprocessed clean dataset. After that, a
Extraction and
Luo Z.-D., Lei L., Li H.-X. rotation-angle-based feature extraction and concentration optimization are performed to select the better
Optimization for PCB
representative angles. Then, all optimized features are probabilistically synthesized and compared with the reference
Defect Detection Under
template for defect detection. Finally, extensive experiments are conducted on the bare PCB and high-resolution
Uncertainties
integrated PCB to demonstrate the proposed method is efficient and faster and requires fewer model parameters. ©
2005-2012 IEEE.
Suksukont A., A Deep Learning-Based Traditional approaches to defect detection, relying on rule-based algorithms or human visual examination, are
Onshaunjit J., System for Detecting inadequate and ineffectual in handling intricate fault patterns. Implementing computer vision techniques in PCB
Srinonchat J. Defects in Printed defect detection can improve inspection processes' accuracy, reliability, and effectiveness. Deep learning is now used
Circuit Boards in those processes to explore important technology. This study presents a novel automated method for defect
identification using deep learning methods. The system is built utilizing AlexNet, U-Net, ResNet 101, and Inception-
v3, prominent frameworks known for their outstanding capacity to extract pertinent visual data components. The
proposed technique comprehensively composes PCB photos, including faulty and defect-free samples. This training
enhances the system's capacity to precisely differentiate between conventional PCB designs and six additional
categories of errors. The suggested architecture of the CNN models consists of double process layers represented as
ConV I, Pooling I, ConV II, and Pooling II. The validation procedure is carried out by using a range of learning rate
data. The implementation findings indicate that the ResNet-101 variant achieves the highest testing accuracy of
99.286%. Indeed, incorporating this technology into the PCB manufacturing process enables the prompt
identification of defective PCBs, leading to less waste, enhanced product quality, and heightened satisfaction among
 end-users. © 2025 IEEE.
Automated optical inspection (AOI) is widely used by manufacturers for the detection of defects in printed circuit
boards (PCBs). Recent works have proposed to apply deep learning for defect detection, which is much faster and
cheaper than manual inspection. However, AOI can only capture defects on the outmost layers of PCBs using
cameras, while modern high-speed circuit PCBs usually have multiple internal layers that need to be inspected.
Compared to optical sensors, X-ray tomography provides noninvasive imaging results of all PCB layers. Though one
can directly apply an off-the-shelf deep detection model trained on optical domains for X-ray imagery, we show that
it usually leads to much lower accuracies in practice. The degraded performance is mainly due to the relatively low
quality of X-ray imaging results and the gaps between optical and X-ray modalities. Furthermore, no X-ray PCB image
Data-Efficient Deep
dataset is publicly available for training deep defect detectors. To this end, we propose a novel dataset for X-ray PCB
Learning for Printed
Nguyen H.D., Cheng D., defect detection, dubbed XD-PCB. In XD-PCB, we provide a benchmark for training X-ray automated defect detection
Circuit Board Defect
Wang X., Shi Y., Wen B. models containing synthesized X-ray images and real X-ray images with real defects. However, in a practical
Detection Using X-Ray
environment, retraining the deep model for every unseen X-ray domain is inefficient due to the domain gaps created
Images
by different X-ray machine settings and the scarcity of defects. Thus, we propose a domain adaptation framework,
dubbed feature-based domain adaptation X-ray (FDX), to improve the efficiency of X-ray PCB defect detection
methods. By minimizing the differences between the deep features extracted from abundant training images and the
scarce unseen images, we improve the model’s performance in a practical situation, thus enhancing the
generalization ability and efficiency of deep detection algorithms when exposed to unseen domains. Our results
demonstrate that XD-PCB provides a valuable training baseline for X-ray PCB defect detection, and our proposed FDX
framework can effectively increase the popular deep learning model by achieving an increment of 10% in terms of
average precisions (APs) compared to other adaptation methods. © 1963-2012 IEEE.
With the miniaturization and increasing complexity of electronic devices, the accuracy and efficiency of printed
circuit board (PCB) defect detection are crucial to ensuring product quality. To address the issues of small defect sizes
and high missed detection rates in PCB surface inspection, this paper proposes an enhanced YOLOv8s model which
not only improves detection performance but also achieves a lightweight design. Firstly, the Nexus Attention module
is introduced, which organically integrates multiple attention mechanisms to further enhance feature extraction and
Research on Deep fusion capabilities, improving the model’s learning and generalization performance. Secondly, an improved CGFPN
Learning Model network is designed to optimize multi-scale feature fusion, significantly boosting the detection of small objects.
Yan H., Zhang H., Gao
Enhancements for PCB Additionally, the WaveletUnPool module is incorporated, leveraging wavelet transform technology to refine the
F., Wu H., Tang S.
Surface Defect upsampling process, accurately restoring detailed information and improving small-object detection in complex
Detection backgrounds. Lastly, the C2f-GDConv module replaces the traditional C2f module, reducing the number of model
parameters and computational complexity while maintaining feature extraction efficiency. Comparative experiments
on a public PCB dataset demonstrate that the enhanced model achieved a mean average precision (mAP) of 97.3% in
PCB defect detection tasks, representing a 3.0% improvement over the original model, while reducing Giga Floating
Point Operations (GFLOPs) by 26.8%. These enhancements make the model more practical and adaptable for
industrial applications, providing a solid foundation for future research. © 2024 by the authors.
Shen M., Liu Y., Chen Defect detection of Detection of printed circuit board assembly (PCBA) defects is crucial for improving the efficiency of PCBA
J., Ye K., Gao H., Che J., printed circuit board manufacturing. This paper proposes PCBA-YOLO, a YOLOv5-based method that can effectively increase the accuracy
 of PCBA defect detection. First, the spatial pyramid pooling module with cross-stage partial structure is replaced in
the neck network of YOLOv5 to capture the resolution features at multiple scales. Second, large kernel convolution is
introduced in the backbone network to obtain larger effective receptive fields while reducing computational
overhead. Finally, an SIoU loss function that considers the angular cost is utilized to enhance the convergence speed
of the model. In addition, an assembled PCBA defect detection dataset named PCBA-DET is created in this paper,
Wang Q., He H., Liu J., assembly based on
containing the corresponding defect categories and annotations of defect locations. The experimental results on the
Wang Y., Jiang Y. YOLOv5
PCB defect dataset demonstrate that the improved method has lower loss values and higher accuracy. Evaluated on
the PCBA-DET dataset, the mean average precision reaches 97.3%, achieving a real-time detection performance of
322.6 frames per second, which considers both the detection accuracy and the model size compared to the YOLO
series of detection networks. The source code and PCBA-DET dataset can be accessed at
[Link] © The Author(s) 2024.
In PCB defect detection, traditional methods often face challenges such as high manual labor costs and slow
processing speeds. This paper introduces an improved approach using the YOLOv5 model, which combines the latest
developments in target detection and deep learning. The YOLOv5 model stands out by providing a simpler
implementation framework, improved detection performance, and accelerated processing capabilities, especially
PCB Defect Detection excelling in detecting small-scale targets. The paper extensively discusses the PCB manufacturing process,
Liu X., Zhang H., Chang
Based on Evolutionary emphasizing the importance of reliable component encapsulation and external circuit connectivity. Furthermore, it
W., Xiao Z., Li Z., Song
Object Detection conducts a comprehensive analysis of the YOLOv5 model structure, highlighting its operational simplicity, compact
X., Zhu G.
Algorithm model size, and clear structural layout, which have played a crucial role in its widespread adoption in industrial
applications. Finally, through extensive experimental analysis, the paper validates that the optimized YOLOv5 model
achieves an average precision mean of 98.2% in PCB defect detection, a 0.6% improvement compared to the original
model. These results demonstrate that the improved YOLOv5 model is an effective solution for PCB defect detection,
laying the foundation for its practical implementation in modern manufacturing processes. © 2024 ACM.
SMT (Surface Mount Technology) has been the backbone of PCB (Printed Circuit Board) production for the last
couple of decades. Even though the speed and accuracy of SMT have been drastically improved in the last decade,
errors during production are still a very valid problem for the PCB industry. With the exponential rise of Artificial
Intelligence in the last decade, the SMT industry was one of the most eager industries to use this new technology to
Research on PCB defect
detect possible defects during production. Lately, traditional image processing techniques started to lag behind
detection using artificial
methods such as machine learning and deep learning when the discussion came to the need of high accuracy. In this
Ural D.I., Sezen A. intelligence: a
paper, we screen academic libraries to understand which of the latest methods and techniques are used in the
systematic mapping
domain and to deduce a general process for detecting defects in PCBs. During the research we have investigated
study
research questions related to state-of-the-art methods, highly mentioned datasets, and sought after PCB defects. All
findings and answers are mapped to be able to understand where this pursuit might point towards. From a total of
270 papers, 90 of them were addressed in detail and 78 papers were chosen for this systematic mapping. © The
Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.
PCB Defect Detection Printed circuit boards (PCBs) play a critical role in electronic products. Ensuring these products' long-term reliability
Tseng S.-H., Kuo C. Based on Improved and consistent performance requires effective PCB defect detection. Although existing deep learning models for PCB
Deep Learning Model defect detection are not highly accurate, they often neglect capability considerations. This paper introduces a
 precise, fast, and lightweight defect detection model, CCG-YOLO, based on an enhanced YOLOv5 model to address
this issue. The enhancements in CCG-YOLO can be summarized as follows: (1) Improved Backbone network: The
feature extraction ability of the Backbone network is enhanced by introducing a C3HB module, which fosters spatial
interaction capabilities. (2) Lightweight feature fusion network: A lightweight convolution structure called Ghost-
Shuffle Convolution is incorporated in the feature fusion network, remarkably reducing model parameters while
maintaining performance. (3) Efficient residual networking: To enhance model performance further, a CNeB module
is introduced based on the ConvNeXt network, which replaces the C3 module in the Neck. CNeB improves model
detection accuracy and reduces the number of model parameters. The combination of these enhancements results
in impressive performance. CCG-YOLO achieves mean average precision (mAP@0.5) of 99.5% and 88.75% in
mAP@0.5:0.95 on the TDD-Net public dataset. Compared with the original YOLOv5s algorithm, CCG-YOLO offers a
4.24% improvement in mAP@0.5:0.95, a 1MB reduction in model size, a 0.472M decrease in the number of
parameters, a 0.6G floating point operation reduction in computational complexity, and a 120 frames per second
real-time inference speed. These experimental results underscore that the proposed model excels in accuracy and
speed and has a compact size for PCB defect detection. Moreover, CCG-YOLO is easily deployable on low-end
devices, making it well-suited for meeting the real-time requirements of industrial defect detection. © 2024 World
Scientific. All rights reserved.
In response to the challenges of small-size defects and low recognition rates in Printed Circuit Boards (PCBs), as well
as the need for lightweight detection models that can be embedded in portable devices, this paper proposes an
improved defect detection method based on a lightweight shared convolutional head using YOLOv8n. Firstly, the
Spatial and Channel reconstruction Convolution (SCConv) is embedded into the Cross Stage Partial with
YOLOv8n-Enhanced PCB
Convolutional Layer Fusion (C2f) structure of the backbone network, which reduces redundant computations and
Defect Detection: A
enhances the model’s learning capacity. Secondly, an adaptive feature selection module is integrated to improve the
Lightweight Method
network’s ability to recognize small targets. Subsequently, a Shared Lightweight Convolutional Detection (SLCD) Head
An J., Shi Z. Integrating Spatial–
replaces the original Decoupled Head, reducing the model’s computational complexity while increasing detection
Channel Reconstruction
accuracy. Finally, the Weighted Intersection over Union (WIoU) loss function is introduced to provide more precise
and Adaptive Feature
evaluation results and improve generalization capability. Comparative experiments conducted on a public PCB
Selection
dataset demonstrate that the improved algorithm achieves a mean Average Precision (mAP) of 98.6% and an
accuracy of 99.8%, representing improvements of 3.8% and 3.1%, respectively, over the original model. The model
size is 4.1 M, and its FPS is 144.1, meeting the requirements for real-time and lightweight portable deployment. ©
2024 by the authors.
The production quality of printed circuit boards (PCBs) is critical to ensure that electronic products work properly.
PCB Defect Detection
Images acquired in complex environments with image uncertainties such as low contrast and multiple noises affect
Model Based on
the performance of computer vision-based surface defect detection of PCBs. For datasets with image uncertainty,
Intrinsic Feature
Yu X., Han-Xiong L., this article proposes a classification model based on intrinsic feature decomposition and multilevel fusion for PCB
Decomposition and
Yang H. defect detection. First, the modified Retinex decomposition is used to generate multilayer reflection and shading
Multilevel Fusion
maps that are more appropriate to the classification task. A grading sparse encoder is designed for the reflection
Against Image
map to characterize the image features. Next, a bilateral convolutional filter with a multidimensional transformer is
Uncertainty
used to remove the noise information present in the shading map. Finally, feature fusion is performed by a multilayer
 attention module of channel attention and pixel attention for classification to improve the utilization of defective
feature information. By comparing the proposed algorithm with other advanced algorithms in a dataset set with
different luminance ratios, it is demonstrated that the proposed algorithm can effectively improve the accuracy of
PCB defect detection. © 2001-2012 IEEE.
Printed Circuit Board (PCB) is one of the most important components of electronic products. But the traditional
defect detection methods are gradually difficult to meet the requirements of PCB defect detection. The research on
PCB defect recognition method based on convolutional neural network is the current trend. The PCB defect image
recognition based on DenseNet169 network model is studied in this paper. In order to reduce the omission of PCB
Printed Circuit Board defects in actual detection, it is necessary to further improve the sensitivity of the model. Therefore, a classification
Defect Image model based on the multimodel fusion of the DenseNet169 model and the ResNet50 model is proposed. At the same
Zhang J., Chang Z., Xu
Recognition Based on time, the network structure after multimodel fusion is improved. The improved multimodel fusion model Mix-Fusion
H., Qu D., Shi X.
the Multimodel Fusion enables the network to not only retain the recognition accuracy of the ResNet50 model for NG defects and small
Algorithm defect images but also improve the overall recognition accuracy through the feature reuse and bypass settings of the
DenseNet169 model. The experimental results show that when the threshold is 0.5, the sensitivity of the improved
multimodel fusion network can reach 99.2%, and the specificity is 99.5%. The sensitivity of Mix-Fusion is 1.2% higher
than that of DenseNet169. High sensitivity means fewer missed NG images, and high specificity means less workload
for employees. The improved model improves sensitivity and maintains high specificity. Copyright © 2024 by ASME.
The integrity of product assembly in the precision assembly industry significantly influences the quality of the final
products. During the assembly process, products may acquire assembly defects due to personnel oversight. A severe
assembly defect could impair the product’s normal function and potentially cause loss of life or property for the user.
For workpiece defect inspection, there is limited discussion on the simultaneous detection of the primary kinds of
assembly anomaly (missing parts, misplaced parts, foreign objects, and extra parts). However, these assembly
anomalies account for most customer complaints in the traditional hand tool industry. This is because no equipment
can comprehensively inspect major assembly defects, and inspections rely solely on professionals using simple tools
and their own experience. Thus, this study proposes an automated visual inspection system to achieve defect
Utilizing Deep Learning inspection in hand tool assembly. This study samples the work-in-process from three assembly stations in the ratchet
Lin H.-D., Jheng C.-K.,
for Defect Inspection in wrench assembly process; an investigation of 28 common assembly defect types is presented, covering the 4 kinds of
Lin C.-H., Chang H.-T.
Hand Tool Assembly assembly anomaly in the assembly operation; also, this study captures sample images of various assembly defects for
the experiments. First, the captured images are filtered to eliminate surface reflection noise from the workpiece;
then, a circular mask is given at the assembly position to extract the ROI area; next, the filtered ROI images are used
to create a defect-type label set using manual annotation; after this, the R-CNN series network models are applied to
object feature extraction and classification; finally, they are compared with other object detection models to identify
which inspection model has the better performance. The experimental results show that, if each station uses the
best model for defect inspection, it can effectively detect and classify defects. The average defect detection rate (1-β)
of each station is 92.64%, the average misjudgment rate (α) is 6.68%, and the average correct classification rate (CR)
is 88.03%. © 2024 by the authors.
 – Due to the ongoing miniaturization of electronic products and the use of miniature printed circuit boards (PCBs),
existing AI-based defect detection methods have exhibited poor performance in detecting tiny PCB defects. This issue
can potentially compromise safety, degrade manufacturing quality, and increase production costs. To tackle this
problem, we propose two novel techniques for PCB defect detection, namely Selective Feature Attention (SF
attention) and Pixel Shuffle Pyramid (PSPyramid). SF attention identifies important features from a pyramid feature
Improving PCB defect
map to fuse the semantic and spatial information, while PSPyramid effectively fuses semantic features to detect
Fung K.C., Xue K.-W., detection using
various types of defects on PCBs, especially tiny defects. Moreover, a customized training strategy, specifically for
Lai C.-M., Lin K.-H., selective feature
PCB defect detection, is devised. To evaluate the performance of our proposed algorithms, extensive experiments
Lam K.-M. attention and pixel
have been conducted on two well-known PCB datasets containing tiny defects: the DeepPCB and TDD datasets. Our
shuffle pyramid
proposed non-referential method achieves performance comparable to existing referential methods on the DeepPCB
dataset, making it more feasible for industrial applications. Compared to state-of-the-art methods, our method
reduces the error by 16%, in terms of AP50, on the TDD dataset. The experimental results demonstrate the
effectiveness of our proposed method in improving the quality assurance process for PCBs in the electronics industry.
© 2024
Printed Circuit Board (PCB) serve as a critical medium for the interconnection of electronic components within the
modern industrial system. However, the detection of defects in PCB is challenging due to the variety of defect types,
small sizes, and complex backgrounds. To address these challenges, this paper presents an improved YOLOv8
An Improved YOLOv8 algorithm that incorporates a lightweight mobilevit architecture and introduces the CA (Coordinate Attention)
Peng Q., Gao X., Wang
Method for PCB defect mechanism. Additionally, we replace the CIoU with PIoU in the regression loss. The final experimental results
S.
detection demonstrate a 4.8% increase in precision, a 9.8% increase in recall, and a map 50 of 93.2%, with over 20% reduction
in parameter count and computational cost. This advancement makes real-time defect detection by machines a
feasible endeavor for the future. © 2024 Technical Committee on Control Theory, Chinese Association of
Automation.
In modern electronic devices, printed circuit boards play a crucial role as core components, and their quality is
essential for device performance and reliability. Due to the complexity of PCB manufacturing processes, issues such
as open circuits, short circuits, and solder defects often occur, affecting the overall quality of products. Traditional
manual inspection methods are unable to meet high standards, leading to the gradual adoption of automated defect
detection technology as the mainstream trend in the industry. This paper presents a network model for PCB defect
Printed Circuit Board detection, aimed at addressing the challenge of detecting small targets. We propose SmallObjectEnhancePyramid,
Lv Z., Chen X., Lu Z., Liu
Defect Detection Based which utilizes the P2 feature layer processed by SPDConv to obtain features rich in small object information and
K., Xia K., Zuo H., Ding
on Feature Pyramid fuses them with P3. Then, we integrate features using CSP-OmniKernel to effectively learn feature representations
B.
Spatial Prior Network from global to local. Additionally, we introduce RMTBlock to improve RepC3, reducing the computational burden of
global information and model parameter quantity. This study not only improves the accuracy of PCB defect
detection, but also helps optimize manufacturing processes and enhance production efficiency. Experimental results
show that FPSPNet effectively improves the detection accuracy of printed circuit board defects, achieving an mAP0.5
value of 98.8%, precision of 99.3%, and recall of 97.3% on the dataset. It also outperforms existing models in terms of
performance. © 2024 IEEE.
Hua L., Li K., Cheng L., Research on PCB Defect The detection of defects in PCB components has always been a popular field, with the main challenge being the
 presence of defects that cannot be recognized by 2D machine vision. This paper investigates PCB component defect
detection using a combination of 2D and 3D techniques. We propose some image enhancement and contour
extraction methods, and employ Contrast Limited Adaptive Histogram Equalization (CLAHE) to correct the color
differences caused by inter-board illumination. By incorporating morphological adaptive anisotropic diffusion
filtering, we have improved the performance of PCB board and component contour extraction. And we find that
contour extraction based on polygon approximation outperforms diagonal circular regression. In the 2D defect
detection method, masked template matching has been used to reduce the calculation of similarity involving non-
Detection Using 2D
Chen Y., Yin D., Yu F.R. component pixels, thereby improving recognition accuracy. To address defects that cannot be identified by 2D
and 3D Segmentation
techniques, we have generated the original point cloud by fusing RGB-D data of PCB components. By utilizing
RANSAC and DBSCAN algorithms, the point cloud is segmented into reference planes of component, upper planes of
component and point clouds of pins. Based on the geometric description operators for calculating depth of point
clouds, features have be made such as difference of depth between reference and upper planes and cross-sectional
area to identify defects. Experimental results demonstrate that the combination of 2D and 3D machine vision
techniques can accurately address various types of PCB component defects. © The Author(s), under exclusive license
to Springer Nature Singapore Pte Ltd. 2024.
The quality control of printed circuit boards (PCBs) is paramount in advancing electronic device technology. While
numerous machine learning methodologies have been utilized to augment defect detection efficiency and accuracy,
previous studies have predominantly focused on optimizing individual models for specific defect types, often
overlooking the potential synergies between different approaches. This paper introduces a comprehensive
Enhancing Printed
Law K.N.C., Yu M., inspection framework leveraging an ensemble learning strategy to address this gap. Initially, we utilize four distinct
Circuit Board Defect
Zhang L., Zhang Y., Xu PCB defect detection models utilizing state-of-The-Art methods: EfficientDet, MobileNet SSDv2, Faster RCNN, and
Detection through
P., Gao J., Liu J. YOLOv5. Each method is capable of identifying PCB defects independently. Subsequently, we integrate these models
Ensemble Learning
into an ensemble learning framework to enhance detection performance. A comparative analysis reveals that our
ensemble learning framework significantly outperforms individual methods, achieving a 95% accuracy in detecting
diverse PCB defects. These findings underscore the efficacy of our proposed ensemble learning framework in
enhancing PCB quality control processes. © 2024 IEEE.
 Improving detection accuracy is a challenging task when detecting defects in PCBs. To conquer the issue of
degradation of recognition performance of existing intelligent detection algorithms in different environments, we
present an improved algorithm, IEMA-YOLOv8, based on YOLOv8. First, we design a novel Efficient Multi-Scale
Attention (EMA) combined with the Inverted Residual Mobile Block (IRMB) to form a new attention mechanism
called IEMA. The IEMA module is subsequently incorporated into the C2f module to boost the model's overall
A Detection Algorithm performance. Secondly, the Spatial Pyramid Pooling Enhanced with ELAN (SPPELAN) module is employed to improve
for Surface Defects of the original Spatial Pyramid Pooling Fast (SPPF) module, thereby bolstering the model's capacity to recognize
Yao L., Zhao B., Wang
Printed Circuit Board defective regions. Finally, the More Focused Intersection over Union (Focaler-IoU) loss function replaces the original
X., Mei S., Chi Y.
Based on Improved Complete Intersection over Union (CIoU) loss function, aiming to compensate for the limitations of the current
YOLOv8 bounding box regression methodology, thus further enhancing the efficacy of detection in the task. The experimental
results show that our proposed IEMA-YOLOv8 algorithm has precision, recall, mAP50, and mAP50:95 values of 88.8%,
96.8%, 94.6%, and 51%, respectively, provided that the model complexity is kept basically the same and the Frames
Per Second (FPS) value reaches 116.2. These values are 1.1%, 3.6%, 4.9%, and 6% higher than the original YOLOv8n
algorithm. Compared with the YOLO family of one-stage detection models, our proposed algorithm has significant
advantages in defect detection. © 2013 IEEE.
During the production process, Printed circuit boards (PCBs) may encounter many defect issues which can severely
affect the functionality of the circuit. However, existing PCB defect detection methods suffer from low detection
accuracy and slow detection speed. Considering the requirements of PCB factories for detection accuracy and real-
time performance This paper first made structural improvements to the existing YOLOx defect detection algorithm,
Rapid Detection of PCB introducing PAN+FPN, SimAM, and SIoU modules to improve the detection accuracy of the algorithm, and named it
Pan Y., Zhang L., Zhang
Defects Based on YOLOx-Plus. Then, algorithm acceleration is achieved by quantifying network parameters and designing FPGA
Y.
YOLOx-Plus and FPGA accelerators. In the experiment, the average detection accuracy of YOLOx-Plus is 93.2%, the network loss is reduced
by 1.094, the model size is compressed by 64%, detection speed is improved by 68.1%, and the FPS reaches 72.6. The
experimental results show that the proposed PCB defect detection method based on YOLOx-Plus and FPGA can
efficiently detect typical defects in PCB boards, overcome the limitations of existing methods, and have a wide range
of practical applications. © 2013 IEEE.
 The surface defects of printed circuit boards (PCBs) generated during the manufacturing process have an adverse
effect on product quality, which further directly affects the stability and reliability of equipment performance.
However, there are still great challenges in accurately recognizing tiny defects on the surface of PCB under the
complex background due to its compact layout. To address the problem, a novel YOLO-HorNet-MCBAM-CARAFE
(YOLO-HMC) network based on improved YOLOv5 framework is proposed in this article to identify the tiny-size PCB
defect more accurately and efficiently with fewer model parameters. First, the backbone part adopts the HorNet for
YOLO-HMC: An
Yuan M., Zhou Y., Ren enhancing the feature extraction ability and deepening the information interaction. Second, an improved multiple
Improved Method for
X., Zhi H., Zhang J., convolutional block attention module (MCBAM) is designed to improve the ability of the model to highlight the
PCB Surface Defect
Chen H. defect location from a highly similar PCB substrate background. Third, the content-aware reassembly of features
Detection
(CARAFE) is used to replace the up-sampling layer for fully aggregating the contextual semantic information of PCB
images in a large receptive field. Moreover, aiming at the difference between PCB defect detection and natural
detection, the original model detection head (DH) is optimized to ensure that YOLOv5 can accurately detect PCB tiny
defects. Extensive experiments on PCB defect public datasets have demonstrated a significant advantage compared
with several state-of-the-art models, whose mean average precision (mAP) can reach 98.6%, verifying the accuracy
and applicability of the proposed YOLO-HMC. © 1963-2012 IEEE.
This paper presents a defect detection technique in printed circuit boards (PCBs) using self-supervised learning of
local image patches (SLLIP). Defect detection in PCBs is often hindered by the problems like a lack of defect data, the
existence of tiny components, and the cluttered background. From the observation that some local image patches of
PCB defect detection a PCB are similar in texture and brightness distribution but are different in semantic features, the proposed self-
Yao N., Zhao Y., Kong with self-supervised supervised learning method utilizes the relative position estimation, spatially adjacent similarity, and k-means
S.G., Guo Y. learning of local image clustering of patches to learn finely classified semantic features. Then, the feature consistency between the local
patches image patches and the background is learned by a local image patch completion network. The feature differences
between the estimated and the original image patches are used to detect anomaly areas in PCBs. Experiment results
on the PCB defect dataset demonstrate that the proposed SLLIP outperforms the state-of-the-art methods. © 2023
Elsevier Ltd
The detection of PCB defect quality plays an important role in PCB fabrication. However, the size of the PCB defects is
too small to identify. In order to improve the detection efficiency of existing algorithms, a joint multiscale PCB defect
target detection and attention mechanism, which named RAR-SSD, was proposed. By using lightweight receptive field
block module (RFB-s) with an attention mechanism module, we built a wider range of effective focused features,
which exploited the importance of different features in different channels without increasing the computing power
PCB defects target
of the network. In addition, we built a feature fusion module to efficiently fuse low-level feature information with
Jiang W., Li T., Zhang detection combining
high-level feature information to produce a more complete feature map and improve the accuracy of fault
S., Chen W., Yang J. multi-scale and
recognition. The proposed network improved the fault recognition accuracy of PCBs by 2.23% over the original SSD
attention mechanism
algorithm, with a recall rate of 6.51% and an F1 value of 4.85%, the model has greatly improved in terms of detection
performance. The optimized algorithm has significant speed and accuracy advantages over the algorithms YOLOv3
and YOLOv5. Experimental results show that the proposed RAR-SSD model has good performance in detecting small
and medium size targets for defects in the PCB manufacturing process and is of some guidance for the subsequent
detection of PCB defects. © 2023
 Printed circuit boards (PCBs) are a critical component of modern electronic equipment, performing a crucial role in
the electronic information industry chain. However, accurate detection of PCB defects can be challenging. To address
this problem, this paper proposes an enhanced detection method based on an improved YOLOv7 network. First, the
An Enhanced Detection SwinV2_TDD module is proposed, which adds a convolutional layer to extract the local features of the PCB. Then, the
Method of PCB Defect Magnification Factor Shuffle Attention (MFSA) mechanism is introduced, which adds a convolutional layer to each
Yang Y., Kang H.
Based on Improved branch of the Shuffle Attention (SA) to expand its depth and enhance the adaptability of the attention mechanism.
YOLOv7 The SwinV2_TDD module and MFSA mechanism are integrated into the YOLOv7 network, replacing some ELAN
modules and changing the activation function to Mish. The evaluation indexes used are Precision (P), Recall (R), and
mean Average Precision (mAP). Experimental results show that the enhanced method achieves an AP of 98.74%,
indicating a significant improvement in PCB defect detection performance. © 2023 by the authors.
Printed circuit boards are an important part of electronic products. Ensuring the quality of printed circuit boards is an
important task. Traditional target detection methods require cropping of the image, which leads to poor real-time
detection and poor accuracy. To address these issues, we propose an improved defect detection method for
YOLOv5s. We use the single-stage target detection network model YOLOv5s. Firstly, we design a set of small anchor
frames with small target sizes. Second, we add a small target detection head to the network and fuse strong
PCB defect target
semantic information from the second layer of the network feature map. Third, we embed the CBAM module and
Chen X., Zhou Y. detection based on
the SPP module into the Neck layer of the network. the CBAM module helps the network to find regions of interest in
improved YOLOv5s
the images. the SPP module fuses the multi-scale features of the targets. The purpose of the above three
components is to solve the problem that small target features are easily lost. The experimental results show that the
improved network has a precisi rate of 97.1%, a recall rate of 89.1%, an mAP0.5 of 94.5% and an mAP0.5:0.95 of
48.5%, all four metrics are higher than the YOLOv5s base model. © 2023 Copyright held by the owner/author(s).
Publication rights licensed to ACM.
Printed circuit boards (PCBs) are primarily used to connect electronic components to each other. It is one of the most
important stages in the manufacturing of electronic products. A small defect in the PCB can make the final product
inoperable. Therefore, careful and meticulous defect detection steps are necessary and indispensable in the PCB
manufacturing process. The detection methods can generally be divided into manual inspection and automatic
optical inspection (AOI). The main disadvantage of manual detection is that the detection speed is too slow, resulting
in a waste of human resources and costs. Thus, in order to speed up the production speed, AOI techniques have been
adopted by many PCB manufacturers. Most current AOI mechanisms use traditional optical algorithms. These
PCB Defect Detection
Chen I.-C., Hwang R.- algorithms can easily lead to misjudgments due to different light and shadow changes caused by slight differences in
Based on Deep Learning
C., Huang H.-C. PCB placement or solder amount so that qualified PCBs are judged as defective products, which is also the main
Algorithm
reason for the high misjudgment rate of AOI detection. In order to effectively solve the problem of AOI misjudgment,
manual re-judgment is currently the reinspection method adopted by most PCB manufacturers for defective
products judged by AOI. Undoubtedly, the need for inspectors is another kind of labor cost. To reduce the labor cost.
of manual re-judgement, an accurate and efficient PCB defect reinspection mechanism based on deep learning
algorithm is proposed. This mechanism mainly establishes two detection models, which can classify the defects of
the product. When both models have basic recognition capabilities, the two models are then combined into a main
model to improve the accuracy of defect detection. In the study, the data provided by Lite-On Technology Co., Ltd.
 were implemented. To achieve the practical application value in the industry, this research not only considers the
problem of detection accuracy, but also considers the problem of detection execution speed. Therefore, fewer
parameters are used in the construction of the model. The research results show that the accuracy rate of defect
detection is about 95%, and the recall rate is 94%. Compared with other detection modules, the execution speed is
greatly improved. The detection time of each image is only 0.027 s, which fully meets the purpose of industrial
practical application. © 2023 by the authors.
Defect detection is a critical component in ensuring the highest quality of printed circuit board manufacture. The
Defect Recognition on work investigates the application of transfer learning to identify YOLOv8 design flaws. The YOLOv8 model performs
Single Layer Bare exceptionally well, with mean Average Precision (mAP) of 98.3%, Precision of 98.6%, and Recall of 96.0%. Numerous
Printed Circuit Boards datasets provide high detection rates across a range of fault types following thorough testing and inference. This
Austria Y.D., Fajardo for Quality Control and extensive evaluation demonstrates the robustness of the model by reliably identifying Fissure, Mousebite, EC-Short,
A.C. Visual Inspection: A MC-Spur, Breakout, Pinhole, and Copper defects. The study emphasizes how the YOLOv8 design can identify flaws,
Low-Sample-Size Deep providing a dependable and effective way to guarantee the dependability and caliber of printed circuit boards
Transfer Learning throughout the production process. The foundation for utilizing YOLOv8 in industrial quality control is established by
Approach this study, which holds significant potential for improving defect detection and quality control in the manufacturing
industry. © 2023 IEEE.
With the growing competition between the various manufacturers of electronic products, the quality of the products
developed and the consequent confidence in the brand are fundamental factors for the survival of companies. To
guarantee the quality of the products in the manufacturing process, it is crucial to identify defects during the
production stage of an electronic device. This study presents a system based on traditional visual computing and new
de Oliveira G.G., System for PCB Defect deep learning methods to detect defects in electronic devices during the manufacturing process. A prototype of the
Caumo Vaz G., Antonio Detection Using Visual proposed system was developed and manufactured for direct use in the production line of electronic devices. Tests
Andrade M., Iano Y., Computing and Deep were performed using a particular smartphone model that had 22 critical components to inspect and the results
Ronchini Ximenes L., Learning for Production showed that the proposed system achieved an average accuracy of more than 90% in defect detection when it was
Arthur R. Optimization directly used in the operational production line. Other studies in this field perform measurements in controlled
laboratory environments and identify fewer critical components. Therefore, the proposed method is a real-time high-
performance system. Furthermore, the proposed system conforms with the Industry 4.0 goal that process system
digitization is essential to improve indicators and optimize production. Copyright © 2023 Gabriel Gomes de Oliveira
et al.
Accurate and efficient detection of printed circuit board (PCB) defects is essential for the reliability and yield of
electronic products. However, the PCB defects are generally too tiny to be effectively identified by existing object
detection models. In this article, a novel detection network for PCB defect detection is proposed based on the
A PCB Defect Detector coordinate feature refinement (CFR) method. The CFR structure is designed to suppress the conflicting information
Yang J., Liu Z., Du W.,
Based on Coordinate from different levels in order to highlight the PCB defect features. Then, four CFR modules are combined with the
Zhang S.
Feature Refinement YOLOv5s baseline framework whose network structure is further optimized by utilizing content-aware reassembly of
features (CARAFE) upsampler to aggregate contextual semantic information in large receptive fields, and by
integrating an additional lager detection layer to strengthen the detection for small targets. Compared with several
state-of-the-art detection models, the proposed detector exhibits significant advantage in detection accuracy of PCB
 defects with fairly compact model size, and provides a feasible solution to fulfill the industrial requirement of real-
time PCB defect detection. © 1963-2012 IEEE.
Defects in the printed circuit board assembly significantly impact product functionality and quality. Automated
optical inspection (AOI) systems, employed by manufacturing quality control teams, are designed to accurately
detect these defects in a timely manner, thereby reducing the underkill (false negatives) and overkill (false positives)
rates. An AOI system requires optimal settings for resolution, brightness, camera angle, and data variety to ensure
effective defect detection. However, consistently achieving these ideal conditions in a manufacturing environment
Toward Optimal Defect presents challenges. Our proposed framework enhances defect detection through data preparation and detection
Noroozi M., Detection in Assembled modules, effectively addressing these manufacturing challenges. We developed one-and two-stage object detectors
Ghadermazi J., Shah Printed Circuit Boards and assessed their performance using precision, recall, and intersection over union metrics. Our framework employs
A., Zayas-Castro J.L. under Adverse a diverse range of augmentation techniques to effectively train the defect detectors, enabling the expansion of a
Conditions limited data set. The trained detectors are evaluated using real-world data. We assessed quality control plans across
various confidence thresholds. At a 65% confidence threshold, one-stage detector models did not exhibit any false
negatives and had minimal false positives. The You Only Learn One Representation (YOLOR) model outperformed
both one-stage and two-stage detectors, achieving 100% precision and recall, a 96% mIoU, and an impressive
inference time of 11 ms, making it an ideal choice for high-production printed circuit board assembly lines. © 2013
IEEE.
Aiming at the problems of the high cost of PCB circuit board defect detection, many types of defects, and complex
shapes, this paper proposes a PCB defect detection method that combines image processing and deep learning. We
use the pruned YOLOv5 algorithm and image processing method to realize the identification and location of defects
PCB Defect Detection
on the PCB board, which improves the accuracy of defect detection. The combination algorithm first puts the original
Chen S., Liang X., Jiang Based on Image
image through image enhancement methods such as straightening, denoising, sharpening, and contrast
W. Processing and
enhancement, and uses the improved YOLOv5 convolutional neural network to identify defects. The experimental
Improved YOLOv5
results show that, without reducing the detection accuracy of the original model, the algorithm compresses the
model to 2.64 MB and shortens the inference time to 20.53 ms, which greatly improves the model deployment
efficiency and detection speed. © Published under licence by IOP Publishing Ltd.
Swathi G., Rohan Automated Printed In the realm of modern electronics, Printed Circuit Boards (PCBs) serve as fundamental components, facilitating the
Reddy B., Ashish Reddy Circuit Board seamless integration of circuits across a wide spectrum of applications, from smartphones to space exploration.
M., Kumar R.P. Inspection: However, addressing the formidable challenge of efficiently detecting concealed threats within complex
Incorporating YOLOv8 environments, where they frequently camouflage themselves within the background, has become a serious threat.
for Efficient Defect To tackle this intricate challenge, the implementation of an Automatic Visual Inspection (AVI) system emerges as an
Detection indispensable necessity in today's world. The objective of this research is to create a proficient system for detecting
defects in printed circuit boards (PCBs) by employing YOLOv8, the most recent advancement in the YOLO series of
object detection models. Our model has achieved a mean Average Precision (mAP50) of 0.955 and (mAP50-95) of
0.516, ensuring the robust detection of PCB defects. Experimental results have revealed that this model exhibits
enhanced detection accuracy in comparison to conventional and deep learning object detection models. To assess
the performance of this model a comparison with the previous iteration, YOLOv7 with the same experimentation
scenarios was conducted. The results demonstrated that the Yolov8 had 17% increment in the mAP50 metric. ©
 2023 IEEE.
Almost every commonly used electronic device relies on a Printed Circuit Board (PCB) produced through Surface-
Mount Technology (SMT). Thanks to advancements that have significantly reduced the size of electronic
components, PCBs have dramatically decreased in size. Consequently, it is crucial to conduct the application and
Subbulakshmi T.,
examination of solder paste during SMT with a strong emphasis on speed and precision. This research survey delves
Ramachandran V., Machine Learning
into an analysis of Solder Paste Inspection (SPI) methods, drawing insights from more than 30 research papers
Anand P., Madhavan Models for PCB Defect
spanning several decades. Initially, the survey identifies various PCB defect detection methods and assembles
R., Revathi M., Detection: A Survey
relevant papers for each approach. Subsequently, it provides a brief assessment of these papers, comparing their
Subramanian G.H.
algorithms, performance, advantages, and challenges. This comprehensive overview aims to offer fresh perspectives
to researchers and readers interested in gaining a deeper understanding of the latest developments in solder paste
inspection and the detection of defects in PCBs. © 2023 IEEE.
In the SMT process, after component placement, checking the quality of component placement on the PCB board is a
basic requirement for quality control of the motherboard. In this paper, we propose a deep learning-based
classification method to identify the quality of component placement. This is a comparison method and the novelty is
that the siamese network is trained to extract the features of the standard placement component map and the
SMT Component
placement component map to be inspected and output the probability of similarity between the two to determine
Defection
Yu C., Zhao Y., Xu Z. the goodness of the image to be inspected. Compared to traditional hand-crafted features, features extracted using
Reassessment Based
convolutional neural networks are more abstract and robust. In addition, during training, the concatenated network
on Siamese Network
pairs the sample images to expand the amount of training data, increasing the robustness of the network and
reducing the risk of overfitting. The experimental results show that this method has better results than the general
model for the classification of placement component images. © 2022, The Author(s), under exclusive license to
Springer Nature Singapore Pte Ltd.
PCB (printed circuit board) is an extremely important component of all electronic products, which has greatly
facilitated human life. Meanwhile, tons of PCBs in the waste streams become a waste of resources, which puts the
recycling and reuse of PCBs in urgent need. In the manufacturing and recycling of electronic products, the
classification of PCBs, recognition of sub-components, and defect detection have been the key technology.
Traditional manual detection and classification are subjective and rely on individuals' experience. With the
LPViT: A Transformer development of artificial intelligence, lots of research efforts have been dedicated to the automated detection and
Based Model for PCB recognition of PCBs. In this paper, we propose a transformer-based model, LPViT, for defect detection and
An K., Zhang Y.
Image Classification and classification of PCBs. We conduct the defect detection task on the dataset DeepPCB, which consists of six different
Defect Detection types of PCB defects. Defect detection benefits both manufacturing and recycling of PCBs. Among many electronic
products, a group of affordable, general-purpose, and small-size PCBs is very popular, which are referred to as micro-
PCBs. The classification and recognition of those PCBs will greatly facilitate the recycling and reuse process. We
conduct the classification task on a dataset called micro-PCB, which includes 12 types of popular, general-purpose,
affordable, and small PCBs. Through comparative experiments, our system demonstrates its advantage in both
classification and defect detection tasks. © 2013 IEEE.
Bhanumathy Y.R., Defect detection in Printed circuit boards(PCBs) are used in a majority of electronic products in Indian Space Research Org
James M.P., Jha S., PCBs using anisation(ISRO. It can have manufacturing defects, timely detection of which in early stages can prove to be a major
 advantage during the manufacturing process of a PCB, which can speed up the time in which delivery of electronic
products takes place at ISRO. Defect detection in PCBs is majorly a manual process. This process can be automated to
convolutional neural save time. A convolution neural network is proposed for the same. It consists of four layers. First layer is the
Balan S.
network convolution layer, followed by the ReLU activation function in the second layer. Third layer consists of the pooling
function and finally the fourth layer is the fully connected layer. This model provides a fair accuracy for detecting
defective PCB images, thus proving as a major time saver for ISRO. © 2021 IEEE.
A deep ensemble convolutional neural network (CNN) model to inspect printed circuit board (PCB) board dual in-line
package (DIP) soldering defects with Hybrid-YOLOv2 (YOLOv2 as a foreground detector and ResNet-101 as a
classifier) and Faster RCNN with ResNet-101 and Feature Pyramid Network (FPN) (FRRF) achieved a detection rate of
97.45% and a false alarm rate (FAR) of 20%-30% in the previous study [34]. However, applying the method to other
production lines, environmental variations, such as lighting, orientations of the sample feeds, and mechanical
deviations, led to the degradation in detection performance. This article proposes an effective self-adaption method
Automatic Industry PCB that collects 'exception data' like the samples with which the Artificial Intelligent (AI) model made mistakes from the
Board DIP Process automated optical inspection inference edge to the training server, retraining with exceptions on the server and
Li Y.-T., Kuo P., Guo J.- Defect Detection deploying back to the edge. The proposed defect detection system has been verified with real tests that achieved a
I. System Based on Deep detection rate of 99.99% with an FAR 20%-30% and less than 15 s of inspection time on a resolution 7296× 6000 PCB
Ensemble Self-Adaption image. The proposed system has proven capable of shortening inspection and repair time for online operators,
Method where a 33% efficiency boost from the three production lines of the collaborated factory has been reported [6]. The
contribution of the proposed retraining mechanism is threefold: 1) because the retraining process directly learns
from the exceptions, the model can quickly adapt to the characteristic of each production line, leading to a fast and
reliable mass deployment; 2) the proposed retraining mechanism is a necessary self-service for conventional users as
it incrementally improves the detection performance without professional guidance or fine-tuning; and 3) the
semiautomatic exception data collection method helps to reduce the time-consuming manual labeling during the
retraining process. © 2011-2012 IEEE.
In this paper, a new model known as YOLO-v5 is initiated to detect defects in PCB. In the past many models and
different approaches have been implemented in the quality inspection for detection of defect in PCBs. This algorithm
is specifically selected due to its efficiency, accuracy and speed. It is well known that the traditional YOLO models
(YOLO, YOLO-v2, YOLO-v3, YOLO-v4 and Tiny-YOLO-v2) are the state-of-The-Art in artificial intelligence industry. In
Applying deep learning
Adibhatla V.A., Chih electronics industry, the PCB is the core and the most basic component of any electronic product. PCB is almost used
to defect detection in
H.-C., Hsu C.-C., Cheng in each and every electronic product that we use in our daily life not only for commercial purposes, but also used in
printed circuit boards
J., Abbod M.F., Shieh sensitive applications such defense and space exploration. These PCB should be inspected and quality checked to
via a newest model of
J.-S. detect any kind of defects during the manufacturing process. Most of the electronic industries are focused on the
you-only-look-once
quality of their product, a small error during manufacture or quality inspection of the electronic products such as PCB
leads to a catastrophic end. Therefore, there is a huge revolution going on in the manufacturing industry where the
object detection method like YOLO-v5 is a game changer for many industries such as electronic industries. © 2021
American Institute of Mathematical Sciences. All rights reserved.
Defect detection of With the rapid development of the electronic industry, the defect detection of printed circuit board (PCB)
Shen J., Liu N., Sun H.
printed circuit board components is becoming more and more important. The types of PCB components are diverse and accompanied by
 complex character information, which is difficult to identify. The traditional detection method is inefficient, and it is
unable to effectively perform the diversified category detection of PCB components and character recognition in
complex scenes. The deep convolutional neural network has obvious advantages in object detection and character
recognition, which can be used to implement a PCB component defect detection system. In this study, the authors
have established a lightweight PCB type detection model called LD-PCB, which can perform real-time detection while
based on lightweight
improving detection accuracy. In addition, in the character detection of PCB, they have established a fast and robust
deep convolution
character recognition model, called CR-PCB. This model can effectively improve the accuracy of irregular character
network
recognition. Finally, they established and published a dataset of PCB components, and combined with LD-PCB and
CR-PCB to realise the PCB defect detection system. This system can realise the functions of defect detection, wrong
insertion, missing insertion, and character recognition in industrial PCB production. The results show that the method
proposed in this study can effectively detect defects on PCB components. © The Institution of Engineering and
Technology 2020.
In the printing industry, defect detection is of crucial importance for ensuring the quality of printed matter. However,
rarely has research been conducted for web offset printing. In this paper, we propose an automatic defect detection
method for web offset printing, which consists of determining first row of captured images, image registration and
defect detection. Determining the first row of captured images is a particular problem of web offset printing, which
Automatic defect
has not been studied before. To solve this problem, a fast computational algorithm based on image projection is
Zhang E., Chen Y., Gao detection for web offset
given, which can convert 2D image searching into 1D feature matching. For image registration, a shape context
M., Duan J., Jing C. printing based on
descriptor is constructed by considering the shape concave-convex feature, which can effectively reduce the
machine vision
dimension of features compared with the traditional image registration method. To tolerate the position difference
and brightness deviation between the detected image and the reference image, a modified image subtraction is
proposed for defect detection. The experimental results demonstrate the effectiveness of the proposed method. ©
2019 by the authors.
This paper proposes a method for automatic image-based classification of solder joint defects in the context of
Automatic Optical Inspection (AOI) of Printed Circuit Boards (PCBs). Machine learning-based approaches are
frequently used for image-based inspection. However, a main challenge is to manually create sufficiently large
labeled training databases to allow for high accuracy of defect detection. Creating such large training databases is
time-consuming, expensive, and often unfeasible in industrial production settings. In order to address this problem,
Towards automatic
Dai W., Mujeeb A., an active learning framework is proposed which starts with only a small labeled subset of training data. The labeled
optical inspection of
Erdt M., Sourin A. dataset is then enlarged step-by-step by combining Kmeans clustering with active user input to provide
soldering defects
representative samples for the training of an SVM classifier. Evaluations on two databases with insufficient and
shifting solder joints samples have shown that the proposed method achieved high accuracy while requiring only
minimal user input. The results also demonstrated that the proposed method outperforms random and
representative sampling by ~ 3.2% and ~ 2.7%, respectively, and it outperforms the uncertainty sampling method by
~ 0.5%. ©2018 IEEE.
The application of deep learning to defect detection has become a central research theme in manufacturing quality control, particularly in the context of printed circuit
boards (PCBs) and related electronic assemblies. Existing studies report substantial improvements in detection accuracy, robustness and real-time performance; however,
the literature remains highly fragmented, with contributions dispersed across diverse architectures, optimisation strategies and application contexts. As a result, extracting
transferable knowledge that can inform related tasks—such as PCB component detection—requires a structured and systematic synthesis of prior research.

To address this issue, the present study adopts a PRISMA-guided systematic literature review within the literature review phase. Rather than using PRISMA solely as a
methodological reporting framework, it is employed as a knowledge organisation and synthesis tool to consolidate empirical evidence from deep learning–based defect
detection studies. This approach enables the identification of recurring detection components and design patterns that have demonstrated effectiveness across multiple
industrial inspection scenarios.

Following PRISMA-based identification, screening and eligibility assessment, the reviewed studies were synthesised using an evidence-category approach. Instead of
summarising individual papers, studies employing similar methodological principles were grouped into approach-level evidence categories, such as YOLO-based one-stage
detectors, attention- and transformer-enhanced models, lightweight and edge-oriented architectures, two-stage detectors and multimodal inspection methods. The
resulting PRISMA-style evidence synthesis table (Table X) summarises these categories, the number of studies included, the datasets considered and the synthesised
performance outcomes.

Table X demonstrates that YOLO-based one-stage detectors dominate the defect detection literature due to their ability to achieve high accuracy while maintaining real-time
inference speeds. Attention mechanisms and transformer-based components consistently enhance performance under complex backgrounds and small-defect conditions,
while lightweight design strategies improve deployability in manufacturing environments. Importantly, the synthesis reveals a clear shift in recent research from purely
accuracy-driven optimisation toward efficiency-aware and deployment-conscious detection system design.

Building on the insights derived from the PRISMA evidence synthesis, the literature review moves beyond algorithm-level comparison to focus on the functional detection
components embedded within successful defect detection systems. Across the reviewed studies, several components recur consistently: feature extraction backbones,
multi-scale feature representation and fusion mechanisms, attention and context modelling modules, small-object enhancement strategies, optimisation-oriented loss
functions and lightweight deployment components. These components are synthesised in the preceding subsection on detection component analysis and form the
conceptual basis for subsequent model development.

The use of defect detection literature as a foundation for PCB component detection is theoretically and practically justified. Although defect detection and component
detection differ in objectives—identifying anomalies versus recognising functional elements—both tasks rely on shared capabilities such as precise localisation,
discrimination under cluttered backgrounds and robustness to scale variation. The PRISMA-guided synthesis therefore treats defect detection studies as a knowledge base
from which transferable detection components can be abstracted and adapted to the PCB component detection task.

By explicitly integrating the PRISMA evidence synthesis table within the literature review, this study ensures that the transition from prior research to proposed
methodology is evidence-driven rather than ad hoc. The detection components identified through the systematic review directly inform the design choices adopted for PCB
component detection, thereby strengthening the theoretical coherence and practical relevance of the proposed approach.

This synthesis is derived from the systematically reviewed studies provided in the uploaded document

This study contributes to the manufacturing and information systems literature by theoretically consolidating fragmented research on AI-enabled defect detection into
coherent evidence categories. By moving beyond algorithm-specific comparisons, the review highlights the evolution of defect detection theory from rule-based inspection
systems to data-driven, learning-oriented quality management frameworks. The findings demonstrate that detection performance improvements are increasingly driven by
architectural integration strategies (e.g. attention mechanisms, multi-scale feature learning and transformer-based representations) rather than isolated model
enhancements.
Furthermore, the evidence synthesis reveals a theoretical shift from accuracy-centric evaluation toward efficiency-aware modelling, where real-time performance,
computational cost and deployability are treated as integral dimensions of detection effectiveness. This supports emerging perspectives in manufacturing systems theory
that position AI models as socio-technical components embedded within production environments rather than standalone analytical tools.

From a managerial and industrial perspective, the findings provide actionable insights for manufacturing decision-makers and system integrators. The dominance of YOLO-
based one-stage detectors indicates that high-accuracy, real-time inspection is now feasible for deployment on production lines without excessive computational overhead.
The growing body of lightweight and edge-oriented models further suggests that manufacturers can implement AI-based inspection solutions at the shop-floor level,
reducing dependence on centralised computing infrastructure.

Additionally, the review highlights the practical value of multimodal inspection approaches, such as X-ray and acoustic imaging, for applications where traditional optical
inspection is insufficient. For practitioners, this underscores the importance of aligning inspection technology selection with product complexity and defect characteristics.
Collectively, the evidence supports a transition toward intelligent, adaptive quality control systems that enhance defect detection reliability, reduce manual inspection costs
and improve overall manufacturing performance.

Included
Models / Methods Main Outcomes Strength of Key Limitations
Evidence Category Studies Datasets Considered
Synthesized Synthesized Evidence Identified
(n)
High
YOLOv5–YOLOv9, YOLOx mAP consistently 96– (consistent Performance often
YOLO-based One-Stage DeepPCB, PKU-PCB,
~18 variants with architectural 99.5%, real-time results dataset-specific; limited
Detectors TDD, PCBA-DET
modifications speeds >100 FPS across cross-domain validation
datasets)
2–5% mAP
Attention modules, Increased architectural
YOLO + Attention / Public PCB datasets + improvement over
~10 MobileViT, Swin, HorNet, High complexity; higher
Transformer Hybrids custom industrial sets YOLO baselines; better
Mamba blocks training cost
small-defect recall
Model size 3–7 MB, Occasional accuracy
Lightweight & Edge- GhostNet, depthwise PKU-Market-PCB, Moderate–
~7 FLOPs ↓ 40–70%, drop on dense defect
Deployable Models separable conv, shared heads DsPCBSD+ High
competitive accuracy layouts
Slower inference;
Two-Stage (Region-Based) Faster R-CNN + ResNet mAP 85–90%,
~3 Custom PCB datasets Moderate unsuitable for high-
Detectors backbones improved localization
speed lines
Hybrid Detection– ~3 YOLO + CNN classifiers Annotated PCB Improved defect Moderate Increased pipeline
 categorization
Classification Pipelines (Inception, ResNet) datasets complexity
accuracy
Ensemble Learning YOLO, Faster R-CNN, SSD, Accuracy ≈ 95%, Low– High computational
~2 PCB defect datasets
Frameworks EfficientDet ensembles improved robustness Moderate cost; limited scalability
Multimodal / Non-Optical X-ray DL, SAT-image CNNs, AP improvement up to Limited public datasets;
~2 XD-PCB, SAT datasets Moderate
Methods domain adaptation 10%, low false alarms expensive acquisition
CNN classifiers, ViT-based
Non-YOLO CNN / DeepPCB, SAT High accuracy in Low– Often lack real-time
~3 models, self-supervised
Transformer Models datasets controlled settings Moderate performance
learning
Systematic reviews, mapping Identify DL as SOTA; High
Survey & Mapping Studies ~2 Multiple datasets No empirical validation
studies highlight gaps (conceptual)

Table notes (Emerald style)

1. Evidence strength reflects consistency of findings across datasets and experimental settings rather than statistical meta-analysis.

2. Public datasets such as DeepPCB, PKU-PCB and TDD dominate evaluation, while industrial datasets are largely proprietary.

3. Performance metrics primarily include mAP, precision, recall and FPS, consistent with industrial inspection requirements.

Table X synthesises the evidence from the reviewed studies following a PRISMA-aligned approach. The findings indicate a strong dominance of YOLO-based one-stage
detectors, particularly when augmented with attention mechanisms or transformer modules. While accuracy levels have largely converged near saturation, recent research
increasingly prioritises lightweight architectures and deployability on edge devices. Evidence gaps persist in cross-dataset generalisation and long-term industrial validation,
suggesting directions for future research

 Evidence concentration: Most high-quality evidence supports YOLO-based one-stage detectors, especially when enhanced with attention or transformer modules.

 Consistency: Accuracy gains are consistent across multiple datasets, indicating methodological robustness rather than isolated improvements.

 Shift in research focus: As accuracy saturates (~99% mAP), studies increasingly prioritize lightweight design, real-time speed, and edge deployment.

 Evidence gaps:

o Limited cross-dataset generalization studies

o Few standardized industrial benchmarks

o Sparse evidence on long-term deployment reliability

Based on the PRISMA-guided synthesis of defect detection literature and the identification of transferable detection components, the following section develops a
framework for PCB component detection using deep learning.
Detection Component Identified from Role in Defect Detection (Synthesised from Expected Benefit in PCB Component
Adaptation for PCB Component Detection
Defect Detection Literature PRISMA Review) Detection

Extracts low-level textures and high-level Learns visual characteristics of PCB Robust representation of component
Feature extraction backbone
semantic representations of defects components (pads, chips, resistors, ICs) shapes and textures

Enhances detection of small and variably Captures PCB components of different physical Improved detection of both large ICs
Multi-scale feature representation
sized defects sizes and aspect ratios and small passive components

Feature fusion mechanisms (FPN, PAN, Integrates spatial detail with semantic Combines fine-grained layout details with Better localisation of closely packed
BIFPN) context board-level structural context components

Attention mechanisms (channel, Suppresses background noise and highlights Focuses on component regions while Improved discrimination between
spatial, hybrid) defect-relevant regions suppressing PCB traces and silkscreen noise components and background

Context modelling (transformers, non- Captures long-range dependencies in Learns spatial relationships among Enhanced understanding of component
local blocks) cluttered backgrounds neighbouring PCB components layout patterns

Enhances detection of miniature PCB Reduced missed detections for small

Small-object enhancement strategies Improves localisation of micro-defects
components (e.g. SMD resistors, capacitors) components

Anchor and detection head Tailors detection heads for small or dense Customised heads for dense PCB component Higher recall in dense and high-
optimisation defects layouts resolution boards

Loss function design (IoU variants, Improves bounding box regression and Stabilises training for overlapping or visually More accurate component bounding
weighted losses) convergence similar components boxes

Lightweight and efficiency-oriented Enables real-time defect inspection on Supports deployment of component detection Practical applicability in manufacturing
design production lines models on edge or shop-floor systems environments

Addresses class imbalance and defect Improved generalisation and

Data augmentation strategies Augments limited PCB component datasets
scarcity robustness

Conceptual Framework for PCB Component Detection

Building on the PRISMA-guided synthesis of deep learning–based defect detection literature and the mapping of transferable detection components, this study proposes a
conceptual framework for PCB component detection. The framework is designed to translate empirically validated detection principles into a structured model for
identifying and localising PCB components under manufacturing-relevant conditions, such as dense layouts, scale variation and background clutter.
Framework rationale

While PCB component detection differs from defect detection in its primary objective, both tasks share fundamental visual and computational challenges, including precise
localisation, discrimination in complex backgrounds and robustness to scale variation. The proposed framework therefore treats PCB component detection as a component-
aware extension of defect detection paradigms, leveraging detection components that have demonstrated consistent effectiveness across multiple inspection contexts.

Core framework layers

The conceptual framework is organised into five interrelated layers, each corresponding to detection components synthesised from the literature.

Input and preprocessing layer

This layer processes raw PCB images and applies standardisation and augmentation strategies to mitigate variations in illumination, resolution and board orientation. Data
augmentation techniques are incorporated to address class imbalance and limited component samples, thereby enhancing generalisation during training.

Feature extraction layer

The feature extraction layer employs a deep neural backbone to learn hierarchical representations of PCB components. Informed by the literature, the backbone is designed
to capture both fine-grained texture details and higher-level structural patterns. Lightweight or hybrid architectures may be adopted to balance representational power with
computational efficiency.

Multi-scale representation and fusion layer

Given the heterogeneous sizes of PCB components, the framework integrates multi-scale feature representation mechanisms to preserve spatial detail across resolution
levels. Feature fusion modules combine low-level spatial information with high-level semantic context, enabling accurate localisation of both large integrated circuits and
small surface-mounted devices.

Attention and context modelling layer

To enhance discrimination between components and surrounding PCB structures (e.g. traces, vias and silkscreen), the framework incorporates attention and context
modelling components. These mechanisms guide the network to focus on component-relevant regions while learning spatial relationships among neighbouring elements,
supporting robust detection in densely populated layouts.

Detection and optimisation layer

The final layer performs component localisation and classification using optimised detection heads and task-specific loss functions. Loss design is informed by defect
detection literature to improve bounding box regression and convergence stability, particularly in scenarios involving overlapping or visually similar components. Efficiency-
oriented design choices are integrated to support real-time inference and practical deployment.

Framework integration and knowledge transfer

The proposed conceptual framework explicitly integrates detection components identified through the PRISMA-guided literature review, ensuring that PCB component
detection design choices are grounded in accumulated empirical evidence. By structuring the detection task around reusable components rather than isolated algorithms,
the framework facilitates systematic model development and supports adaptation to diverse PCB inspection scenarios.

Positioning within the research

This conceptual framework serves as the foundation for the proposed PCB component detection methodology presented in the subsequent section. It provides a coherent
structure that links literature-derived insights to implementation-level decisions, thereby strengthening the theoretical justification and practical relevance of the proposed
approach

Table X maps the core detection components synthesised from the PRISMA-guided defect detection literature to their corresponding roles in PCB component detection. This
mapping demonstrates how empirically validated defect detection strategies can be systematically adapted to support component-level recognition tasks.