Scribd
Upload
5 views2 pages

Deep Learning Paper Preprocessing and Models

The document summarizes two research papers focused on deep learning models for medical image classification. Paper 1 utilizes EfficientNet for brain tumor classification with specific preprocessing steps and hyperparameters, while Paper 2 combines EfficientNet and ResNet for skin disease classification, detailing its unique preprocessing and model fusion techniques. Both papers emphasize the importance of data normalization and dropout for improving model performance.

Uploaded by

saharamr34
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
5 views2 pages

Deep Learning Paper Preprocessing and Models

The document summarizes two research papers focused on deep learning models for medical image classification. Paper 1 utilizes EfficientNet for brain tumor classification with specific preprocessing steps and hyperparameters, while Paper 2 combines EfficientNet and ResNet for skin disease classification, detailing its unique preprocessing and model fusion techniques. Both papers emphasize the importance of data normalization and dropout for improving model performance.

Uploaded by

saharamr34
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

Here are the details from the two papers regarding their preprocessing steps, model

properties, and hyperparameters.

Paper 1
● Paper Name: Classification of Brain Tumors on MRI Images Using Convolutional Neural
Network Model EfficientNet.

● Paper Link: [Link]

● Preprocessing Steps:
○ Data Augmentation: Added to the dataset to manipulate data and improve results.

○ Normalization: Implemented min-max normalization.

○ Input Resizing: Images were resized to a constant value of 128 x 128 pixels.

● Model Name and Properties:


○ Models used: EfficientNetB0 and EfficientNetB7.

○ Architecture Layers: Input layer, GlobalAverage Pooling2D layer, Dropout layers,


and Dense layers.

● Hyperparameters:
○ Epochs: 100.

○ Learning Rate: 0.00146.

○ Optimizer: Adam (Scenario 1) and RMSprop (Scenario 2).

○ Dropout Value: 0.2.

○ Dense Layer Units: 1024 and 128.

○ Activation Functions: Relu (for hidden dense layers) and Softmax (for the output
layer).

○ Loss Function: Categorical Crossentropy.

Paper 2
● Paper Name: An Integrated Deep Learning Model with EfficientNet and ResNet for
Accurate Multi-Class Skin Disease Classification.

● Paper Link: [Link]

● Preprocessing Steps:
○ Dataset Split: Data was distributed into Training (80%), Validation (10%), and
Testing (10%).

○ Normalization: Features were normalized using Batch Normalization to stabilize


training.

○ Zero Padding: Applied at the initial architecture stage of ResNet50 for spatial
dimension protection.

○ Resizing: Standardized to fit the requirements of the pre-trained models (typically


224x224 for ResNet/EfficientNet, though specific target values for this paper are
managed by the "compound scaling" of EfficientNet).

● Model Name and Properties:


○ Model: A fusion-based deep learning model (ensemble).

○ Merged Architectures: EfficientNet-B0, EfficientNet-B2, and ResNet50.

○ Fusion Mechanism: Features from the three branches are concatenated and
transmitted to dense and dropout layers.

● Hyperparameters:
○ Regularization: Uses Dropout and Batch Normalization to reduce overfitting.

○ Activation Functions: SWISH (used in EfficientNet blocks) and ReLU (used in


ResNet stages).

○ Pooling: Average Pooling and Max Pooling are utilized to decrease dimensions.

○ Classification Layer: Final layer uses a fully connected layer for multi-class
predictions.

You might also like