Shodhshauryam, International Scientific Refereed Research Journal

Available online at : [Link]

© 2024 SHISRRJ | Volume 7 | Issue 2
ISSN : 2581-6306 doi : [Link]

Medical Herbs Identification Using Machine Learning

K Naresh¹, M Teja²
1Assistant Professor, Department of MCA, Annamacharya Institute of Technology & Sciences, Tirupati, Andhra Pradesh,
India
2Post Graduate, Department of MCA, Annamacharya Institute of Technology & Sciences, Tirupati, Andhra Pradesh,

India

Article Info ABSTRACT

In Ayurveda, plants have been utilised as a source of medicine since the
Article History Vedic era. The most crucial step in making ayurveda medicine is
Received : 02 April 2024 correctly identifying the plants. step, which was completed by hand.
Published : 13 April 2024 Because of the need for mass production, it's critical to automatically
identify these plants. In this research, we implement an ensemble
Publication Issue :
supervise machine learning system based on colour, texture, and
geometrical aspects called the random forest algorithm for medicinal
March-April-2024
plant identification.
Volume 7, Issue 2
Keywords : Medicinal Plant, Supervise learning, Image Processing,
Identification, Automatic, Ayurveda.
Page Number : 519-525

I. INTRODUCTION agricultural [Link] Ayurvedic doctors of old

selected the therapeutic plants and made the
Approximately 5000 years ago, during the Vedic era, remedies for their patients themselves. The majority
India developed the ancient medical system known of plants are recognised bytheir foliage Common
as Ayurveda. The primary ingredients found in procedures for classifying a plant's leaf include
ayurvedic medications are plant leaves as well as taking a picture, adjusting the size and noise level,
additional plant elements like the bark, roots, etc. It extracting features, applying the suggested
has been discovered that over 8000 plants with methodology, and ultimately identifying or
Indian origins have therapeutic properties. Herbal recognising the plant.
remedies from various Indian systems use
combinations of a tiny subset of these plants, II. EXISITING SYSTEM AND PROPOSED SYSTEM
totaling 1500. Specifically, 500 of these herbs are
used in commercial Ayurvedic treatments. The A. Existing system
majority of plants used in ayurvedic formulations—
more than 80%—are gathered from forests and The Artificial Neural Network (ANN) model was
wastelands, with the remainder being grown on the foundation of the current medicinal plant

Copyright © 2024 The Author(s): This is an open-access article distributed under the terms of the Creative 519
Commons Attribution 4.0 International License (CC BY-NC 4.0)
 K Naresh et al Sh Int S Ref Res J, March-April-2024, 7 (2) : 519-525

encourages significant contributions to the study by utilising smartphone photos of plant leaves.
and practice of herbal medicine. Ayurvedic medicinal plant identification using
processing leaf sample images The authors of this
Advantages: work are P. M. Krishna, C. M. Surya, and V. P. Gopi.
In the ayurvedic medical field, Finding the right
1. High accuracy botanical medicines for utilization in the production
2. Larger and Diverse dataset of drugs is essential. The shape, hue, & feel of a
3. Enhanced Robustness leaves
4. Potential for scalibility are the primary characteristics needed to identify a
5. Automation and time efficiency medicinal plant. Deterministic characteristics for
species identification are present in the colour and
III. LITERATURE SURVEY texture of the leaf on both sides. This work
investigates morphological characteristics and
1. An Android-based computerized plant feature vectors from the front and rear of a green
identification system H. A. Chathura, Priyankara, leaf to determine the optimal feature combination
and D. K. Withanage are authors of the paper. that optimises the identification [Link]
photos of the front and rear surfaces of leaves from
The characteristics seen on plant leaves are popular ayurvedic medicinal plants are used to
sufficient to set them apart from other species. One construct a database of medicinal plant leaves. The
of the fundamental problems in digital image distinctive feature combination is used to categorise
processing is the identification of plants from leaf the leaves. Experiments spanning a broad range of
photographs. classifiers have yielded identification rates as high
These image processing methods frequently use as 99%. By including identification by dried leaves
shape-based digital morphological cues to detect into the previously mentioned study, a combination
leaves. Extremely a few studies has focused on of feature vectors is created that allows for
handheld devices, despite the fact that several identification rates to surpass 94%.
studies were done on finding plants via leaves. In
this study, we present a plant identification system [Link] neural network identification of a leaf
that uses leaf images and blends the classifier from a Philippine herbal medicine plant CO-
developed by the Support Vector Machine (SVM) AUTHORS: R. G. de Luna and others
with the Pack of The user (BOW) model.
The work described in the above article employs a
with SIFT characteristics. After 20 species were method which identifies and recognizes specific
trained to be classified, the accuracy level of the Philippine herbs.
system was 96.48%. Based on the findings, we
created an Android app that connects to the server combines the use of artificial neural networks with
and makes it possible for users to recognize a variety image processing techniques to extract pertinent
of plants leaf attributes. Twelve distinct herbal medicine
plant leaves are sampled in real life, with each leaf

Volume 7, Issue 2, March-April-2024 | [Link] 520

 K Naresh et al Sh Int S Ref Res J, March-April-2024, 7 (2) : 519-525

captured in a single photograph. Several image It is easy to extract these features from photos.
processing techniques are used to extract various Experimental results show that when these qualities
aspects. The technique can identify the type of leaf are combined with additional well-known
used in medicinal products plants. geometric properties, The effectiveness of
traditional categorization techniques has
under examination by using an artificial neural significantly increased.
network that functions as an independent Neural
network design. Furthermore, the device can We evaluate seven classifiers using publicly
provide information concerning the ailments that a accessible leaf photos from the Internet to
demonstrate the efficacy of the approach.
herbal plant can treat. A 600-image features dataset,
with 50 photos from each plant utilized as a herbal IV. METHODLOGY
remedy for training. With Python's assistance, a
neural network framework with tailored settings is DESCRIPTION OF MODULES:
developed, resulting in 98.16 % id for the whole Dataset: We created the system to obtain the input
dataset. To evaluate the true effectiveness of the dataset in the first module of the Medicinal Herbs
system, a neural network simulation created in Identification course. The process of gathering data
MATLAB is applied to a distinct set of 72 example is the initial step in the process of actually creating
images of herbs. The an automated learning model. The quality of the
simulation is going to impacted during this essential
findings of the experiment show that the accuracy phase within an increasing approach; your more
of herbal plant identification is 96.61%. and better information I discover, the more
[Link] plants with new geometric traits using powerful it will be. There are other ways to collect
conventional data mining techniques the data, including as manual interventions and
AUTHORS: Asdrúbal López-Chau and Rafael online scraping. Our dataset may be found in the
Rojas-Hernández model folder of the project. The dataset is sourced
Plant identification is inside a certain data mining from Kaggle, a widely used standard dataset
application domain. Plant leaves are typically the repository that is used by academics worldwide.
primary feature that set one plant apart from There are 17,973 photos of medicinal plants in the
another. Feature extraction is required for accurate dataset. The dataset referenced from Kaggle may be
identification. The majority of plant recognition found at this This link will take you to a dataset for
systems reported in the literature combine medicinal plants created by Jaya Prakash Pondy on
characteristics with a classification algorithm that [Link]. importing the necessary libraries; you
has been adjusted or modified for usage in this kind will require Python. The necessary libraries, such as
of situation. Three novel geometric properties that the pandas, numpy, matplotlib, and a tool such as
explain In this paper, the skyline and diagonal will first be loaded. Keras, also will be used to build
harmony of results in is suggested. a primary hypothesis, sklearnn will divide up the
test and training information, so PIL is going to
transform the photos into an array of numbers.

Volume 7, Issue 2, March-April-2024 | [Link] 521

 K Naresh et al Sh Int S Ref Res J, March-April-2024, 7 (2) : 519-525

Getting the photos: In this module, we will take the

images out of the dataset and transform them into a
format that will allow us to use them for both
model testing and training. Reading the photos,
scaling them, and normalising the pixel values are
all necessary for this. The pictures and their labels
will be [Link], resize each image to (180,180)
so that they are all the same size for identification.
Next, turn the pictures into a numpy array.

Dataset division: The picture dataset will be split

into training and testing sets for this module. Divide Fig. 1 : Elodie, Alovera, Coriander, Drumstick,
the dataset into test and train sets. There are 20% Hibiscus, Mint, Neem, Papaya, Palem, Rui, and
test and 80% train data. This will be carried out in Tulsi are medicinal plants._Basil Model of training
order to test the model on untested data in order to
assess its correctness, validate the model's Flow Chart
performance, and train the model on a subset of the
data. Divide the dataset into test and train sets.
There are 20% test and 80% train data.

Accuracy on test set: The model's accuracy will be

evaluated on the test set subsequent to its training
and assessment on the validation set. The model's
performance will be assessed in part by measuring
the accuracy on the test set. On the test set, we
achieved 96.79% accuracy.

Storing the Trained Model: Save your tested and

created model as an.h5 [Link] file via a framework
such as pickle.

as soon as you are comfortable enough to introduce

it into a production-ready [Link] that
pickle is installed in your [Link] next step is to

Save the design to an.h5 document after importing

it into the module.

Volume 7, Issue 2, March-April-2024 | [Link] 522

 K Naresh et al Sh Int S Ref Res J, March-April-2024, 7 (2) : 519-525

V. EXPERIMENTAL SETUP 5.2 Image understanding with deep convolutional

networks
5.1 Algorithm used
ConvNets and recurrent net modules are combined
Multiple array data, neural networks are able to in a recent amazing presentation to generate image
handle challenges like processing a color image captions. Current ConvNet topologies consist of
composed of three two-dimensional arrays billions of connections between units, tens to 20
containing pixel intensities in every single one of layers of ReLUs also and tens of billions of weights.
the three color channels. Numerous arrays can Training such a multitude a few years ago could've
represent a wide range of data modalities, needed fortnight; with to advances in software,
technology, and technique speed, it can now only
including: 3-dimensional video or holographic take a few [Link]-based imaging algorithms
photographic; two-dimensional (D) pictures or execute sufficiently successfully that most major
music spectral images; & one pixel messages and tech companies, which includes Facebook, Google,
scenes, such as speech and others. Microsoft, Oracle, IBM, Yahoo!, Twitter,
. Convolutional neural networks (ConvNets) or Adobe, along with an increasing quantity of
leverage the characteristics of natural signals start-ups, have initiated R&D projects and deployed
through four main concepts: multiple layer usage, ConvNet-based visual understanding solutions and
Geographic links, combining, and sharing weights. services. ConvNets are easily and effectively
A structure of a standard ConvNet is broken up into implemented in field-programmable gates or
multiple stages. Layers made up of pooling and circuits.
layers with convolution are the layers that comprise
the first phases. The units of the convolutional layer, VI IMPLEMENTATION
as the name suggests, are split into map patterns,
with all of them connected to specific patches in the An output that meets the demands of the end user
data maps of the layer above by an adjacency bank, and effectively conveys the material is considered
which is a set of weights. Subsequently, a non- high quality. The way that processing results are
linearity such as a Reynolds Linear Option (ReLU) communicated to users and other systems is
applies on the localised calculated result. The units through any system's outputs. The output design
of a feature map share a common filter bank. determines how data is moved for instant usage and
the physical output. It is the most important and
Within a layer, distinct feature maps employ accessible source of information for the user. Well-
various filter banks. This architecture serves two considered and efficient output design improves
purposes. First, local clusters of values in array system integration and streamlines user decision-
data—like images—are frequently highly making.
connected, resulting in recognisable, easily
identifiable local themes. 1. The process of designing computer output should
be methodical and well-planned; appropriate
output should be created while making sure that

Volume 7, Issue 2, March-April-2024 | [Link] 523

 K Naresh et al Sh Int S Ref Res J, March-April-2024, 7 (2) : 519-525

every output component is made in a way that

makes the system simple and efficient for users to
utilise. When analysing computer output, designers
should pinpoint the precise output required to
satisfy specifications.

2. Decide how information will be presented.

3. Create documents, reports, or other formats with
data that the system has created. The output form
of an information system should accomplish any or Fig. 3 : Recognizing an unidentified image as Mint
all of the following objectives. Share information
about what was done in the past, the current state
of affairs, or what is anticipated in the future. 1. List
important events, opportunities, problems, etc

Utilize the knowledge as the system collected to

develop written materials, reports, or other types of
information. An data system's final type should
achieve every one of a few things goals. Talk about
what has been performed using the past, how things
are going now, or what's planned for the future.

1. Enumerate significant occurrences, chances,

Picture -4: the System’s Context Matrix Testing
issues, etc.
1. cautions.
Outcome Table -1
2. Set off an event.
3. Verify a course of action. Parameters Performance-testing
Outcome parameters
(percentage)
Percent accuracy 94.54
Indicates the sensitivity 96.23
Specificity 50.00
F-score 97.70

CONCLUSION

Fig. 2 : Understanding an unidentified image as Tulsi The To determine the proper variety of medicinal
plant, the random forest algorithm is an ensemble
managing machine learning process based on hue,
texture, and geometrical properties.

Volume 7, Issue 2, March-April-2024 | [Link] 524

 K Naresh et al Sh Int S Ref Res J, March-April-2024, 7 (2) : 519-525

method for medicinal plant identification that we and Management (HNICEM), Manila, 2017,
have applied in this study. With the help of shape, pp
colour, and texture traits, leaves may be identified [5]. Rafael Rojas-Hernández and Asdrúbal
with 94.54 percent [Link] technique's highly López-Chau, "Plant identification using new
encouraging findings highlight how suitable this geometric features with standard data mining
algorithm is for systems that identify therapeutic methods", Networking Sensing and Control
plants. In the future, this approach can be expanded (ICNSC) 2016 IEEE 13th International
to include more plant species with higher precision. Conference on, pp. 1-4, 2016.
[6]. Wang, X., D. Huang, J. Du, H. Xu, & L.
FUTURE ENHANCEMENT Heutte, 2008. Classification of plant leaf
The suggested method can develop further by images with complicated background.
investigating these directions for future research,
increasing its precision, resilience, and usefulness
for the goal of finding medicinal herbs. These
developments will support research on herbal
medicine, conservation initiatives, and the
identification of novel uses and beneficial qualities
of medicinal plants.

REFERENCE

[1]. "Computer assisted plant identification

system for Android," 2015 Moratuwa
Engineering Research Conference
(MERCon), Moratuwa, 2015, pp. 148-153.
[2]. P. M. Kumar, C. M. Surya and V. P. Gopi,
"Identification of ayurvedic medicinal plants
by image processing of leaf samples," 2017
Third International Conference on Research
in Computational Intelligence and
Communication Networks (ICRCICN),
Kolkata, 2017, pp. 231-238.
[3]. R. G. de Luna et al., "Identification of
philippine herbal medicine plant leaf using
artificial neural network," 2017IEEE
[4]. 9th International Conference on Humanoid,
Nanotechnology, Information Technology,
Communication and Control, Environment

Volume 7, Issue 2, March-April-2024 | [Link] 525