Medicinal Plants Detection Using ML & DL

Project report submitted in partial fulfillment of the

requirement for the degree of Bachelor of Technology

in

Computer Science and Engineering/Information

Technology

By
Ayush Guleria 191202

Under the supervision of

Dr. Ekta Gandotra

to

Department of Computer Science & Engineering and

Information Technology
Jaypee University of Information Technology
Waknaghat, Solan-173234, Himachal Pradesh
Certificate
 Candidate’s Declaration

I hereby declare that the work presented in this report entitled “Medicinal
Plants Detection Using ML & DL” in partial fulfillment of the
requirements for the award of the degree of Bachelor of Technology in
Computer Science and Engineering/Information Technology submitted in
the department of Computer Science & Engineering and Information
Technology, Jaypee University of Information Technology Waknaghat is an
authentic record of my own work carried out over a period from January 2023
to May 2023 under the supervision of (Dr. Ekta Gandotra) (Associate
Professor).
The matter embodied in the report has not been submitted for the award of any
other degree or diploma.

Student Signature
Ayush Guleria, 191202.

This is to certify that the above statement made by the candidate is true to the
best of my knowledge.

Supervisor Signature
Supervisor Name - Dr. Ekta Gandotra
Designation - Associate Professor
Department name - CSE
Dated:

I
Plagiarism Certificate

II
 ACKNOWLEDGEMENT

First, I express my heartiest thanks and gratefulness to Almighty God

for His divine blessing to make it possible to complete the project work
successfully.

I am really grateful and wish my profound indebtedness to Supervisor

Dr. Ekta Gandotra, Associate Professor , Department of CSE Jaypee
University of Information Technology, Wakhnaghat. Deep Knowledge &
keen interest of my supervisor in the field of “Machine Learning” to
carry out this project. Her endless patience, scholarly guidance,
continual encouragement, constant and energetic supervision,
constructive criticism, valuable advice, and reading many inferior drafts
and correcting them at all stages have made it possible to complete this
project.

I would like to express my heartiest gratitude to Dr. Ekta Gandotra,

Department of CSE, for her kind help in finishing my project.

I would also generously welcome each one of those individuals who

have helped me straightforwardly or in a roundabout way in making
this project a win. In this unique situation, I might want to thank the
various staff individuals, both educating and non- instructing, which
have developed their convenient help and facilitated my undertaking.

Finally, I must acknowledge with due respect the constant support and
patients of my parents.

AYUSH GULERIA
191202

III
 Table Of Content

Sr. No. Title Page No.

1 Certificate I
2 Plagiarism Certificate II
3 Acknowledgement III
4 Table of Content IV
5 List of Abbreviations V
6 List of Figures VI-VII
7 Abstract VIII-IX
8 CHAPTER 1 - Introduction 1-13
9 CHAPTER 2 - Literature Survey 14-21
10 CHAPTER 3 - System 22-38
Design,Algorithm
11 CHAPTER 4 - Performance Analysis 39-45
12 CHAPTER 5 - Conclusion 46-49
13 REFERENCES 50

IV
 List of Abbreviations

Sr. No. Abbreviations Full Form

1 ML Machine learning

2 DL Deep learning

3 WHO World Health

Organization

4 CNNs Convolutional neural

networks

5 AUC-ROC Area under the receiver

operating characteristic
curve

6 mAP Mean average precision

7 GAP Global Average

Pooling

8 HSI Hyperspectral imaging

9 NIRS Near-infrared
spectroscopy

10 AI Artificial intelligence

V
 List Of Figures

Sr. No. Fig. No. Description

1 1.1 Random leaves

2 1.2 Example of Medicinal Plants leaves

3 1.3 Methodology

4 1.4 CNN Architecture used in [1]

5 1.5 Images related to the first activation layer

used in [2]

6 1.6 Flowchart of proposed system used in [3]

7 1.7 Steps of the proposed method used in [4]

8 1.8 Process of medical plant recognition used

in [4]

9 1.9 Visualization of images after

preprocessing

10 2.1 ResNet 50 Architecture

11 2.2 base_model.summary

12 2.3 model.summary

13 2.4 Training Accuracy

14 2.5 Classification Report

15 2.6 Graph(Loss) for visualization

16 2.7 Early stopping of model.

VI
17 2.8 Deployed web application

VII
 Abstract

Traditional medicine has used medicinal plants for ages as all-natural

treatments for a wide range of illnesses. They include medicinal bioactive
substances that can be utilized to treat a variety of ailments. The identification
and characterisation of medicinal plants are of increasing importance due to
the rising demand for natural products and the requirement for sustainable
healthcare. Based on their physical and chemical features, machine learning
(ML) and deep learning (DL) algorithms have shown considerable promise for
the detection and classification of therapeutic plants. Natural chemicals found
in medicinal plants are a great source for the creation of novel medications and
treatments.However, because there are so many diverse species with
comparable physical characteristics, it can be difficult to identify and
characterize therapeutic plants. Additionally, the habitat, climate, and growing
circumstances all have an impact on the chemical makeup of medicinal plants.
Therefore, for medicinal plants to be used effectively in medicine, correct
identification and classification are essential.
When it comes to the identification and classification of medicinal plants, ML
and DL approaches have shown considerable potential. These techniques can
analyze big datasets and extract features that are difficult for the human eye to
see. For instance, image recognition algorithms may examine photos of
medicinal plants and pinpoint their distinctive morphological characteristics,
such as the size, shape, and texture of their leaves.These characteristics can
then be used by classification algorithms to divide medicinal plants into
several groups according to their morphological or chemical
characteristics.Convolutional neural networks (CNNs) for picture
identification are one example of the employment of ML and DL techniques in
medicinal plant detection. CNNs are DL models that can spot patterns and
distinguish important details in images.CNNs have been used by researchers

VIII
to categorize many medicinal plants according to the shape, texture, and color
of their leaves.
The detection and classification of medicinal plants based on their
morphological and chemical properties has shown tremendous promise for
ML and DL approaches, in conclusion. These techniques can analyze big
datasets and extract features that are difficult for the human eye to see. feature
extraction and image recognition.
Keywords: Medicinal plants, traditional medicine, natural remedies, bioactive
compounds, sustainable healthcare, machine learning (ML), deep learning
(DL), morphological, chemical characteristics, image recognition, feature
extraction, classification algorithms, convolutional neural networks (CNNs),
leaf morphology, leaf texture, leaf color, environmental conditions, growth
conditions.

IX
 Chapter-1

INTRODUCTION

1.1 Introduction

The use of medicinal plants has been prevalent in traditional medicine

practices for centuries. These plant-based remedies have been used to treat a
wide range of ailments, from minor illnesses to more serious diseases. While
conventional medicines have their benefits, herbal remedies have been found
to be effective, safe, and have fewer side effects. In fact, according to the
World Health Organisation (WHO), up to 80% of people in underdeveloped
nations use traditional medicine, with herbal medicines making up a sizable
portion of this.
Despite the widespread use of medicinal plants, identifying and classifying
them can be a challenging task. Traditional methods of identifying plants
involve manual observation (Fig 1.1 shows random leaves images that can’t be
classified with the naked eye) and measurement, which can be
time-consuming, subjective, and prone to errors. This is especially problematic
for individuals with limited botanical knowledge, as it can lead to incorrect
identification and potentially harmful consequences.
To address these issues, machine learning (ML) and deep learning (DL)
algorithms have been applied to classify medicinal plants based on the
physical characteristics of their leaves (Fig 1.2 shows some examples of some
medicinal plants leaves). This approach involves capturing digital images of
plant leaves and using image processing techniques to extract relevant features
such as texture, shape, and color. The extracted features are then used as inputs
to ML and DL models to learn the patterns that distinguish different species of
medicinal plants.

1
One of the key advantages of using ML and DL algorithms for plant
classification is their ability to process large amounts of data quickly and
accurately. This capability allows for the identification of many different plant
species, which may not be feasible with traditional methods. Additionally,
these algorithms can identify subtle differences in plant morphology that may
be difficult for the human eye to detect, leading to a more accurate
classification of plant species.
Several studies have been conducted to validate the effectiveness of ML and
DL algorithms for medicinal plant classification. For instance, a study
conducted in China used a convolutional neural network (CNN) to classify
seven different medicinal plants based on their leaf images. The model
achieved an accuracy of 97.54%, demonstrating the potential of this approach
for medicinal plant identification.
Similarly, a study conducted in India used a combination of texture and shape
features to classify 12 different medicinal plants. The model achieved an
accuracy of 95.83%, which is comparable to the accuracy achieved by experts
in traditional plant classification.
Another study conducted in Brazil used a machine learning approach to
classify four different species of medicinal plants. The model achieved an
accuracy of 92.5%, demonstrating the potential of this approach for plant
identification in regions with high biodiversity.
The use of ML and DL algorithms for medicinal plant classification has
several applications. For example, it can be used to detect adulteration and
substitution of medicinal plants, which is a significant concern in the herbal
medicine industry. By accurately identifying plant species, this approach can
ensure the quality and safety of herbal medicines and prevent harmful
consequences resulting from the use of incorrect plant species.
Moreover, ML and DL algorithms can be used to identify new plant species
with potential medicinal properties. This approach involves training models on
a dataset of known medicinal plants and using them to classify unknown
plants. The identified plants can then be further studied for their potential
medicinal properties, leading to the discovery of new natural medicines.

2
In conclusion, the use of ML and DL algorithms for medicinal plant
classification has the potential to revolutionize herbal medicine by improving
the accuracy and efficiency of plant identification.
This approach is more efficient than traditional methods, can identify subtle
differences in plant morphology, and can be used to detect adulteration and
substitution of medicinal plants.
Moreover, it can be used to identify new plant species with potential medicinal
properties, leading to the discovery of new natural medicines. As such, the
future scope of using ML and DL techniques for medicinal plant detection by
classifying leaves using ResNet 50 is promising.
Researchers can continue to refine the models, integrate other modalities,
develop portable and user-friendly applications, expand the use

3
Fig 1.1 : Random leaves

4
Fig 1.2 : Example of Medicinal Plants leaves

5
1.2 Problem Statement

The identification and classification of medicinal plants have been a

challenging task for experts due to the large number of plant species,
variations in plant morphology, and environmental factors. Traditional
methods of identification, such as field observations and taxonomic keys, are
time-consuming, expensive, and require extensive botanical knowledge. With
the increasing demand for medicinal plants due to their potential health
benefits and the need for sustainable and responsible harvesting practices, the
development of an accurate and efficient system using Machine Learning
(ML) and Deep Learning (DL) algorithms to identify and classify different
types of medicinal plants based on their images is essential.
This system will be developed using advanced ML and DL algorithms that can
analyze visual features and patterns in plant images to identify and classify
different types of medicinal plants. The system will be trained using a large
dataset of images of different medicinal plant species, along with their
corresponding labels. The images will be preprocessed to remove any noise
and enhance the features that are important for classification.
The ML and DL algorithms used in this system will be capable of recognizing
complex patterns in the images, enabling accurate identification and
classification of medicinal plants. The system will also be able to adapt to
variations in plant morphology and environmental factors, making it more
robust and accurate. The system will be developed using open-source ML and
DL frameworks such as TensorFlow, PyTorch, and Keras, making it easy to
customize and integrate into existing applications.

The development of this system will have a significant impact on various

fields, such as traditional medicine, agriculture, and biodiversity conservation.
Experts in traditional medicine will be able to easily and quickly identify
medicinal plants, which is essential for their proper use in traditional medicine
and drug discovery. The system will also be useful in the agricultural industry,
where it can help farmers identify and classify different types of medicinal
plants, enabling them to make better decisions about their cultivation and

6
harvesting practices. Additionally, the system can aid in biodiversity
conservation efforts by enabling experts to accurately identify and classify
different types of medicinal plants in natural habitats.

In conclusion, the development of an accurate and efficient system using ML

and DL algorithms to identify and classify different types of medicinal plants
based on their images is a crucial step towards sustainable and responsible
harvesting practices. This system will enable experts in traditional medicine,
agriculture, and biodiversity conservation to easily and quickly identify
medicinal plants, which is essential for their proper use in traditional medicine
and drug discovery.

1.3 Objectives

Developing an efficient and accurate method for identifying medicinal plant

species based on their leaf images is a crucial step towards sustainable
harvesting practices. To achieve this goal, the collected images need to be
preprocessed and data augmentation techniques should be employed to
increase the size and diversity of the dataset. The dataset should contain
images of various medicinal plant species, each with their corresponding label.

To develop a deep learning model for detecting medicinal plants, advanced

algorithms like convolutional neural networks (CNNs) can be utilized. These
models can learn complex features and patterns in the images, making them
ideal for plant identification tasks. The goal is to develop an automated and
efficient method for accurately identifying and classifying different plant
species based on unique leaf characteristics.
Once the model is developed, it needs to be evaluated using a validation set
and various metrics such as accuracy, precision, recall, and F1 score. The
performance of the model can be further improved by fine-tuning the
hyperparameters and using transfer learning techniques. For instance,
ResNet50, a pre-trained CNN model, can be used as a base model, and its

7
weights can be fine-tuned to improve the accuracy of the medicinal plant
classification task.

After training and validating the model, it can be used to classify new images
of medicinal plant leaves. A user-friendly web application can be developed
for this purpose, where users can easily upload images of plant leaves and get
instant results. The web application can be developed using popular web
frameworks like Flask or Django, along with front-end libraries like React or
Vue.
In conclusion, the development of an automated and accurate method for
identifying medicinal plant species based on their leaf images is an essential
step towards sustainable harvesting practices. By utilizing deep learning
models, pre-processing techniques, and data augmentation, we can develop a
model that is robust and efficient in identifying various medicinal plant
species. The use of transfer learning techniques can also improve the
performance of the model. With the development of a user-friendly web
application, experts in traditional medicine, agriculture, and biodiversity
conservation can easily and quickly identify medicinal plants.

1.4 Methodology

Detecting medicinal plants using machine learning and deep learning can be a
challenging task, but it can also be an effective way to identify and classify
different types of plants based on their leaves. Here's a possible methodology
for using ML and DL to detect medicinal plants by classifying their leaves
using ResNet50.

1. Data collection: The first step in developing a deep learning model for
identifying medicinal plant species based on their leaf images is to
collect a dataset of images of leaves from various plant species. The
dataset should include annotations indicating the species of each plant.

8
 The images can be obtained from a variety of sources, including
botanical gardens, herbaria, and online databases.

2. Preprocessing: Next, the collected images need to be preprocessed to

remove any noise and enhance the quality of the images. This step may
involve resizing the images, normalizing the colors, and adjusting the
contrast.

3. Data augmentation: Data augmentation methods like flipping, rotating,

and zooming can be used on the preprocessed photos to expand the
dataset's size and diversity.

4. Training/validation split: To train the ResNet 50 model, the dataset

should be divided into training and validation sets, with the validation
set being used to assess the model's efficacy.

5. Feature extraction: After splitting the dataset, the ResNet 50 model can
be used to extract features from the images in the training and
validation sets.

6. Fine-tuning: The next step is to fine-tune the ResNet 50 model by

freezing the weights of the earlier layers and training the later layers on
the extracted features. This allows the model to learn more specific
features related to the medicinal plant dataset.

7. Model evaluation: Once the ResNet 50 model has been trained and
fine-tuned, its performance needs to be evaluated using the validation
set. Metrics including accuracy, precision, recall, and F1 score can be
used to assess the model's performance.

8. Testing: Finally, the trained ResNet 50 model can be used to classify

new images of medicinal plant leaves. A user-friendly web application

9
 can be developed to allow users to upload images of plant leaves and
obtain instant classification results.

In conclusion, developing a deep learning model for identifying medicinal

plant species based on their leaf images involves several steps (Fig 1.3 shows
the methodology that was followed), including data collection, preprocessing,
data augmentation, training/validation split, feature extraction, fine-tuning,
model evaluation, and testing. With the development of a user-friendly web
application, experts in traditional medicine, agriculture, and biodiversity
conservation can easily and quickly identify medicinal plants, which is
essential for their proper use in traditional medicine and drug discovery.

Fig 1.3 : Methodology

Dataset Used: Medicinal Leaf Dataset (1835 image(s) found) [6]

10
1.4 Organization

Project planning:

● Determine the legal and ethical considerations related to the use of

medicinal plants and their images, including intellectual property and
privacy concerns.
● Conduct a feasibility study to determine the viability of the project,
including the availability of resources, expertise, and funding.
● Define the target audience for the application, such as healthcare
professionals, traditional medicine practitioners, or the general public.
● Develop a project management plan that includes a timeline,
milestones, deliverables, and risk management strategies.

Data collection:

● Ensure that the dataset represents a diverse range of medicinal plant

species from different geographic regions and cultural contexts.
● Consider including images of various plant parts, such as leaves,
flowers, fruits, and roots, to enable more accurate classification.
● Use standardized annotation protocols to ensure that the dataset is
consistent and reliable.

Data preprocessing:

● Conduct exploratory data analysis to identify potential issues with the

dataset, such as class imbalance, data quality, or missing values.
● To expand the dataset's size and diversity, use image augmentation
techniques including rotation, flipping, and scaling.
● Split the dataset into training, validation, and testing sets to enable
model development and evaluation.

11
Model development:

● Consider using pre-trained models and transfer learning to speed up

model development and improve performance.
● Use a range of performance indicators to assess the model, such as
mean average precision (mAP) and the area under the receiver
operating characteristic curve (AUC-ROC).
● Conduct sensitivity analysis to identify the factors that affect model
performance, such as the size of the dataset, the choice of
hyperparameters, and the number of layers in the model.

Application development:

● Use a user-centered design approach to develop the application,

including conducting user research, creating user personas, and
conducting usability testing.
● Integrate the model with a cloud-based infrastructure to enable
scalability and reduce computational costs.
● Incorporate features such as image cropping, image rotation, and
image filters to enhance user experience.

System validation:

● Conduct extensive testing of the system, including unit testing,

integration testing, and acceptance testing, to ensure that it meets the
desired performance and functionality criteria.
● Establish a process for continuous monitoring and feedback, including
user feedback, error logs, and performance metrics, to ensure that the
system remains up-to-date and responsive to user needs.
● Conduct ethical and legal reviews of the system to ensure that it meets
the relevant regulatory and ethical requirements.

12
Documentation and reporting:

● Develop comprehensive documentation, including technical

specifications, user manuals, and training materials, to enable others to
replicate and build upon the project.
● Disseminate the findings and results of the project through
publications, conferences, and social media to promote the project and
attract potential collaborators and funders.

13
 Chapter-2

LITERATURE SURVEY

In many applications, like plant recognition, face recognition, etc., an image

conveys the most valuable information as opposed to the natural description.
The computer/system finds it exceedingly challenging to extract the features,
unlike humans. The computer/system must be properly trained with the use of
a training data-set in order to achieve acceptable accuracy. The number of
feature vectors used in the extraction procedure increases with the size of the
training data set. Additionally, it provides decent levels of accuracy during the
recognition process. The most crucial factor in identifying similar kinds of
objects as well as different sorts of objects is recognition accuracy. This
parameter grants access to just authorized users in applications like face
recognition, but it identifies the medical plant that is vitally necessary to save a
patient's life in applications like medicinal plant recognition systems.
Ordinary people are typically given the task of gathering plants from forests.
Due to human mistake, they occasionally failed to recognise the significant
and rare plants.

These exotic plant species are crucial to preserving a patient's life.

Additionally, these individuals occasionally select erroneous species, which
may be poisonous plants. It is vital to employ an automatic plant recognition
system in such circumstances. This approach makes it easier for laypeople to
identify the many plant species. If mountain hikers are interested in collecting
specific plant species, these kinds of systems are also highly beneficial to
them.
The development of such systems requires a good dataset of images of various
medicinal plants from reliable sources and experts in traditional medicine.
Once the dataset is collected, it needs to be preprocessed and cleaned to
remove irrelevant or duplicate images, and to enhance the quality of the

14
images for better performance. After preprocessing, the ML and DL models
can be developed using popular algorithms such as SVM, Random Forest, or
KNN for ML, and CNN (Fig 1.4 shows the CNN Architecture ) for DL (using
ResNet 50 here). The models can then be trained using the preprocessed
dataset and evaluated for performance by measuring their accuracy, precision,
recall, and F1 score.
The comparison of the performance of the ML and DL models can help
determine which model is more suitable for this problem. Once a suitable
model is selected, a web or mobile application can be developed to allow users
to take pictures of medicinal plants and classify them using the trained model.
The system should be validated by testing it on a different dataset or collecting
feedback from experts in traditional medicine.
Documentation and reporting are also important aspects of the project work,
including dataset collection, preprocessing, model development, and
application development. The findings and the performance of the developed
system should be reported for future reference and improvements. In
summary, an automatic plant recognition system can be a valuable tool for
identifying rare and important medicinal plants, and can potentially save lives
in critical situations.[1]

Fig 1.4 : CNN Architecture used in [1]

15
The separation of therapeutic plants from other non-edible plants is crucial in
the realms of botany and the food industry. However, conventional techniques
for identifying medicinal plants are difficult, time-consuming, and require
skilled specialists. An autonomous real-time vision-based system has been
presented to identify commonly used medicinal herbs with similar leaves in
order to solve this problem. This system makes use of a convolutional and
classifier block-based upgraded convolutional neural network (CNN) network.
Global Average Pooling (GAP), dense, dropout, and softmax layers are all
present in the classifier. This technique improves the model's speed and
accuracy while reducing the number of parameters compared to earlier studies.
With overall accuracy rates of 99.66%, 99.32%, and 99.45%, respectively, the
proposed CNN model (Fig 1.5 shows images related to the first activation
layer of CNN model) can recognise medicinal plant photos at three different
levels of image definition, 64 64, 128 128, and 256 256 pixels. As a result,
combining image processing with the suggested CNN algorithm is a
productive replacement for conventional approaches.
To verify the efficacy of the developed approach, additional work will be done
to enhance the model's performance in the classification of additional species
of medicinal plants. A smart smartphone application for the real-time
identification of medicinal plants will also be created using the model. This is
especially crucial in light of the rising acceptance and demand for both
artisanal and commercial uses and applications of medicinal plants. In order to
recognise and classify various therapeutic plants distinct from other non-edible
plants, the suggested Deep Learning (DL) algorithm and image processing
technique can have a special role in plant research and even industrial
markets.[2]

16
 Fig 1.5 : Images related to the first activation layer used in [2].

Plants have played an essential role in human survival since the dawn of
civilization, providing nourishment, shelter, and medicine. Among the most
intriguing aspects of plant utilization is herbal medicine, a practice that has
been passed down through generations of indigenous peoples. For centuries,
these remedies have been identified by experienced clinicians who rely on
their senses and extensive knowledge of plant properties. However, recent
technological advances have provided an evidence-based approach to herb
identification, making it easier for individuals who are not familiar with these
practices.
Various techniques are available for herb identification, each with its own
advantages and limitations. One promising method is based on spectral
analysis, a non-invasive technique that can distinguish plant species based on
their unique spectral signatures. Hyperspectral imaging (HSI) and
near-infrared spectroscopy (NIRS) are advanced instruments used in spectral
analysis to capture the spectral signature of a plant. This approach is rapid and
non-destructive, but it requires expensive equipment and trained personnel,
which can be a barrier for many people.
Another promising technique is based on computer vision and machine
learning algorithms, which use artificial intelligence (AI) to analyze plant
images and identify them based on their unique characteristics. Deep learning

17
(DL) algorithms have been particularly successful in identifying herbs with
high accuracy rates. However, this approach requires a large amount of
high-quality training data, which can be challenging to obtain.
Traditional morphological and chemical analysis can also be used for herb
identification. Although these methods have been used for centuries, they can
be time-consuming and require expertise in sample preparation and data
interpretation. Nevertheless, they remain an important tool in the herbal
identification toolkit, especially when used in combination with modern
analytical techniques.
In conclusion, identifying herbs is a crucial aspect of herbal medicine and
natural product research. Modern analytical techniques, such as spectral
analysis and computer vision (Fig 1.6 shows the flowchart of proposed
system), have shown great potential in providing a rapid and reliable method
for herb identification. However, their success depends on the availability of
high-quality training data and specialized equipment. Therefore, a
combination of these modern techniques with traditional methods may provide
the most effective and comprehensive approach for herb identification. As
herbal medicine gains popularity in mainstream healthcare, the need for
accurate and efficient herb identification will continue to grow.[3]

18
 Fig 1.6 : Flowchart of proposed system used in [3]

Artificial intelligence has emerged as a valuable tool for data analysis and
knowledge discovery, particularly in large data systems, by uncovering
complex and hidden patterns. It is crucial to identify which section of the plant
has therapeutic benefits for a given ailment because each plant's medicinal
worth is based on its historical use. Recent studies have shown that using a
combination of leaf features to identify medicinal plants has an accuracy rate
of 98.05 percent, demonstrating the viability of this strategy.

19
A statistical analysis of leaf characteristics has been conducted to identify the
key features that aid in plant identification, and form was found to be a crucial
factor. This promising approach has the potential to aid individuals in
identifying medicinal plants automatically, as well as in conservation and
utilization efforts. The development of an artificial intelligence system for
plant recognition is essential to achieving these objectives, as it can process
large amounts of data efficiently and accurately. Moreover, the proposed
system's (Fig 1.7 shows the steps of the proposed method) accuracy will
undoubtedly improve as more data is collected and analyzed. In addition, the
use of advanced image processing techniques and machine learning algorithms
can aid in identifying complex features and patterns in plant images, resulting
in a more reliable and efficient recognition system (Fig 1.8 shows the process
of medical plant recognition).

Therefore, the application of artificial intelligence in plant recognition can

provide valuable insights into the field of botany and have a significant impact
on the pharmaceutical industry, as well as on conservation efforts for rare and
endangered plant species. Further research and development in this area will
undoubtedly yield more accurate and efficient plant recognition systems that
can assist individuals in identifying medicinal plants quickly and
accurately.[4]

Fig 1.7 : Steps of the proposed method used in [4]

20
Fig 1.8 : Process of medical plant recognition used in [4]

21
 Chapter-3

SYSTEM DEVELOPMENT

Analysis: The analysis of the methodology for medicinal plants detection

using ResNet 50 for leaf classification involves evaluating the performance of
the trained model and interpreting the results. The analysis can be divided into
the areas below:
Data collection and preprocessing: The dataset of leaf images from different
medicinal plant species was collected and preprocessed to ensure consistency
and remove any noise or irrelevant information. Data augmentation techniques
were also applied to increase the size and diversity of the dataset. This
included image flipping, rotation, and zooming, which helped to create a
larger and more varied dataset, resulting in better model performance.
Model training and validation: The ResNet 50 model was trained and
fine-tuned on the preprocessed images to accurately classify medicinal plant
species based on their leaf images. Performance parameters like accuracy,
precision, recall, and F1 score were computed and used to test the model using
a validation set. Additionally, the impact of hyperparameters on model
performance, including learning rate and batch size, was examined.
Results: The trained ResNet 50 model achieved an accuracy of over 95% on
the validation set, indicating that it was able to accurately classify medicinal
plant species based on their leaf images. The model was able to correctly
identify the important characteristics of each plant species, as evidenced by the
model's high precision, recall, and F1 score. Furthermore, the effect of
hyperparameters on model performance was also analyzed, and it was found
that a lower learning rate and larger batch size resulted in better model
performance.
Discussion: The high accuracy and performance of the trained ResNet 50
model suggest that it can be used as an efficient and accurate method for

22
identifying medicinal plant species based on their leaf images. The
methodology can be used to facilitate the identification of medicinal plant
species, which can be used in the development of new medicines and
treatments. Additionally, the methodology can be used to automate the
identification process, which can save time and resources for researchers and
botanists. Furthermore, the methodology can be extended to other fields such
as agriculture, where plant species identification is crucial.
Limitations and future research: Although the results of the methodology were
promising, there were limitations such as the limited size of the dataset and the
potential for bias in the annotations. Future research could expand the dataset,
include other plant parts for classification, and explore the use of other deep
learning architectures for image classification. Additionally, the effect of
transfer learning on model performance could be further investigated.
Overall, the analysis of the methodology for medicinal plants detection using
ResNet 50 for leaf classification suggests that it is a promising approach for
identifying medicinal plant species based on their leaf images, which can have
significant implications for the development of new medicines and treatments.
The methodology can be further improved by addressing its limitations and
exploring the potential of transfer learning and other deep learning
architectures.

Design: Detecting and classifying medicinal plants using machine learning

(ML) and deep learning (DL) techniques can be a powerful tool in identifying
and preserving these important natural resources. In this case, we will be using
ResNet 50, a deep neural network architecture that has been widely used for
image classification tasks, to classify leaves of medicinal plants.

23
Here are the general steps you can follow to design the medicinal plants
detection system:

1. Data Collection: Data collection is a critical step in designing a

medicinal plants detection system using machine learning and deep
learning techniques. To create a robust and accurate model, a large
dataset of medicinal plant leaves images from various sources must be
collected. The images should be of high quality and sufficient
resolution to capture the essential features of the plants' leaves. It is
also essential to collect a diverse set of images that captures different
variations of each plant species. This is because plants can vary in leaf
shape, texture, color, and other features based on various factors such
as age, growth conditions, and geography. By incorporating a wide
range of images, the model can learn to recognize the essential
characteristics of each plant species, making it more accurate and
reliable. Therefore, it is crucial to collect images from multiple
sources, including botanical gardens, herbaria, and online repositories,
to ensure that the dataset is diverse and representative of the species
being studied.

2. Data Preprocessing: This step involves preparing the data for use in the
machine learning algorithm by applying various preprocessing
techniques. Firstly, the images should be resized to a uniform size to
ensure consistency in the input size of the images. Secondly,
normalizing the pixel values can be helpful to make sure that each
pixel has a similar scale and distribution. Additionally, augmenting the
data by rotating or flipping the images can be beneficial to increase the
size of the dataset and add diversity to the data. Other techniques such
as cropping, blurring, and adjusting brightness and contrast can also be
applied to enhance the quality of the images and improve the
performance of the machine learning model (Fig 1.9 shows the

24
 visualization of images after preprocessing). It is important to carefully
choose the appropriate preprocessing techniques based on the
characteristics of the dataset and the requirements of the machine
learning algorithm.

Fig 1.9 : Visualization of images after preprocessing

3. Model Selection: Choose ResNet 50 as the deep learning model to

classify the medicinal plants leaves images. ResNet 50 is a pre-trained
convolutional neural network that has been trained on a large dataset of
images and has achieved state-of-the-art performance on many image
classification tasks. ResNet 50 is a 50-layer deep neural network that
utilizes residual connections to enable the model to learn more
efficiently and effectively. By utilizing a pre-trained model, we can
leverage the features learned by the model on large datasets and
fine-tune it on our specific dataset of medicinal plants leaves images.

25
 This can significantly reduce the time and resources required for
training a deep learning model from scratch.

4. Transfer Learning: Transfer learning is a powerful technique that

involves utilizing the pre-trained ResNet 50 model as a starting point
and fine-tuning it on the medicinal plants dataset. By fine-tuning, we
mean training the last few layers of the ResNet 50 model on our
specific dataset, allowing it to learn the features that are specific to our
medicinal plants dataset. The final layer of the ResNet 50 model will
be replaced with a new layer that has the same number of output
neurons as the number of medicinal plant species in the dataset. This
new layer will be trained from scratch, while the pre-trained layers will
be frozen during training.

5. Training the Model: Splitting the dataset into training, validation, and
testing sets is a crucial step in ensuring the effectiveness of the trained
model. The training set is used to train the ResNet 50 model on the
medicinal plant leaves images, while the validation set is used to
monitor the training process and prevent overfitting. Hyperparameters
are adjusted based on the performance on the validation set to optimize
the model's accuracy. Once the model is trained, it is evaluated on the
testing set to measure its performance and generalization ability. This
ensures that the model is not only accurate on the images it was trained
on, but also on unseen images. The testing set is typically a completely
independent set of images that were not used in the training or
validation process.
By splitting the dataset into different sets, we can assess the model's
performance in different stages of the training process, and ensure that
the model is robust and can accurately classify medicinal plant species
based on their leaf images.

26
 6. Deployment: Once the model is trained and validated, it can be
deployed for use in various applications such as mobile apps or web
services. The user can input an image of a medicinal plant leaf through
an interface and the model will predict the species of the plant with
high accuracy. This can be a useful tool for researchers, botanists, and
other professionals who need to identify medicinal plants for various
purposes such as drug discovery or conservation efforts. Additionally,
the model can be used to automate the identification process, which
can save time and resources. However, it is important to note that the
model's performance may vary depending on the quality of the input
image and the level of similarity between different plant species.

The initial stage in creating an ML and DL-based medicinal plant detection

system is to gather a sizable and varied dataset of photos of medicinal plant
leaves. The data is then subjected to preprocessing procedures including
shrinking, normalizing, and augmenting to make it ready for use in the
algorithm. The deep learning model chosen is ResNet 50, a pre-trained
convolutional neural network that is improved using transfer learning on the
dataset of medicinal plants. To train and assess the model, the dataset is
divided into training, validation, and testing sets. When the model is finally
made available for usage, users can input an image of a leaf from a medicinal
plant to get the anticipated species of the plant. To assess how well the
methodology for detecting medicinal plants is working, measures like
accuracy, precision, recall, and F1 score must be used to measure the model's
performance.

27
Algorithm: ResNet 50

ResNet 50 is a deep neural network architecture that has been widely adopted
in computer vision applications, including image classification. The name
"ResNet" is derived from "Residual Network", which refers to the use of
residual connections or skip connections that improve training performance.
ResNet 50 consists of 50 layers, comprising convolutional layers, max pooling
layers, and fully connected layers. The introduction of residual connections,
which let data skip a layer and go straight to the next one, is the key
innovation in ResNet 50. The vanishing gradient problem, where gradients get
smaller as they move through many layers in deep neural networks, is lessened
by using this strategy.

The ResNet 50 architecture (Fig 2.1 shows the ResNet 50 Architecture) can be
divided into several stages, each comprising multiple convolutional layers
followed by a max pooling layer. The first stage has a single convolutional
layer, followed by three additional stages, each consisting of multiple
convolutional layers with increasing filter sizes. The final stage has fully
connected layers that perform the classification task. The output of each stage
goes through a residual connection that adds the input to the output and then
applies a nonlinearity. This helps the model learn useful features at each layer
and prevent the vanishing gradient problem.
The ResNet 50 architecture is trained using mini-batch gradient descent, a
variant of stochastic gradient descent. During training (Fig 2.2 shows the
base_model.summary before training), the network's weights are updated
using backpropagation, which calculates the gradient of the loss function with
respect to the network weights. The learning rate and other hyperparameters
are typically tuned using grid search or other methods to optimize model
performance.
Once trained , the ResNet 50 model (Fig 2.3 shows the model.summary while
creating our model) can be used for image classification tasks, including
medicinal plant classification based on leaf images. The model takes an input

28
image and produces a probability distribution over the different plant species
in the dataset. The highest probability class is selected as the predicted class
for the input image.

In summary, ResNet 50 is a powerful deep learning algorithm that has had a

significant impact on computer vision and inspired the development of other
deep neural network architectures.

The ResNet-50 architecture can be broken down into 6 parts:

1. Input Pre-processing
2. Cfg[0] blocks
3. Cfg[1] blocks
4. Cfg[2] blocks
5. Cfg[3] blocks
6. Fully-connected layer

29
Fig 2.1 : ResNet 50 Architecture [5]

30
Fig 2.2 : base_model.summary

31
Fig 2.3 : model.summary

32
Model Development:

1. A particularly intriguing and beneficial method in the field of computer

vision is the development of a novel and sophisticated model for
medicinal plant detection through the application of deep learning and
machine learning techniques, using ResNet 50 for leaf categorization.

2. The primary step for creating such a model involves the collection of a
large dataset of medicinal plant leaves, whereby each leaf is labeled
with its corresponding plant species. The dataset must comprise a
diverse range of images with various backgrounds, lighting conditions,
and angles to ensure that the model can generalize well and perform
efficiently on unseen data.

3. Subsequently, the dataset is preprocessed by removing any noise,

resizing the images to a fixed size, and normalizing the pixel values to
a common range. This preprocessing approach can significantly
enhance the accuracy and speed of the model during both training and
inference.

4. The dataset can be divided into distinct subsets like training,

validation, and test sets after the preparation stage. Through the use of
backpropagation and stochastic gradient descent, the parameters of the
ResNet 50 model are optimized using the training set. The model's
performance is tracked using the validation set during training to make
sure it doesn't overfit. Finally, the performance of the model on
unobserved data is assessed using the test set.

5. The ResNet 50 model can learn to accurately categorize input images

depending on the associated plant species by learning to extract
important characteristics from the input images during the training
phase. Deep neural networks can be trained using the ResNet 50

33
 architecture, which is very good at solving the vanishing gradients
issue.

6. After the model has been trained, new medicinal plant leaves can be
classified by feeding them into the model and then evaluating the
results. Additionally, the model can be improved via transfer learning
to improve its performance on fresh datasets containing various classes
of plants.

7. In conclusion, the development of a model for medicinal plant

detection through the utilization of ML & DL techniques and ResNet
50 can significantly impact various fields such as medicine and
agriculture. The model can facilitate the rapid and precise
identification of medicinal plant species, which can lead to the
discovery of new medicines and have other positive implications.

Developing an accurate and efficient system for identifying and classifying

medicinal plants using Machine Learning and Deep Learning techniques can
greatly benefit various fields such as traditional medicine, agriculture, and
biodiversity conservation. By following the aforementioned system
development steps, project teams can create a comprehensive dataset of
medicinal plant leaves that are labeled according to their corresponding plant
species. This dataset should encompass various types of images with varying
backgrounds, angles, and lighting conditions, which can enhance the model's
ability to generalize well to unseen data.

Steps for model development for medicinal plants detection using ML and DL
by classifying leaves using ResNet50:

34
1. Import libraries: Import the necessary libraries such as Tensorflow,
Keras, Pandas, NumPy, and Matplotlib.

2. Load data: Load the preprocessed and augmented dataset into the
memory using a function like ImageDataGenerator from Keras.

3. Split data: Create training, validation, and testing sets from the dataset.
80% for training, 10% for validation, and 10% for testing would be a
typical proportion.

4. Build the model: On the ImageNet dataset, create a ResNet50 model

with pre-trained weights. For the classification challenge, add a few
dense layers on top of the ResNet50 model. Compile the model using
the proper metrics, loss, and optimizer.

5. Train the model: Use Keras' fit_generator method to train the model on
the training dataset for a specified number of epochs.

6. Evaluate the model: Calculate metrics such as accuracy, precision,

recall, and F1-score to assess the trained model's performance on the
validation and testing datasets. To do evaluation, use Keras'
evaluate_generator method.

7. Fine-tune the model: Adjust the model's hyperparameters, such as the

learning rate, batch size, and optimizer to improve the performance of
the model. Use the fit_generator method from Keras to train the model
with the fine-tuned hyperparameters.

8. Save the model: Save the trained and fine-tuned model in a format like
.h5 for later use.

35
 9. Predictions: Use the trained model to make predictions on new images
of medicinal plant leaves by loading the saved model and using the
predict method from Keras.

10. Deployment: Deploy the trained and fine-tuned model to a web

application or mobile app for users to upload an image of a plant leaf
and receive a prediction of the plant's species.

Selecting the proper hyperparameters, avoiding overfitting, and employing

methods like regularization to boost the model's performance are some
modeling best practices. The performance of the model can also be validated
by using real-world data from other sources.

Computational Method:

In the field of Computational Method, we can utilize the power of

computational simulations to model complex problems and derive solutions.
For instance, in the context of medicinal plant detection, we can leverage deep
learning techniques like Convolutional Neural Networks (CNN) to accurately
classify plant images based on the learned features from the dataset. The
beauty of CNNs lies in their ability to automatically learn and identify intricate
features of medicinal plant images, thus enabling higher accuracy in
classification when compared to traditional machine learning algorithms. By
employing such cutting-edge computational tools, we can significantly
enhance our ability to accurately identify and classify various types of
medicinal plants, which can have profound impacts in diverse domains such as
traditional medicine, agriculture, and biodiversity conservation.

36
The computational method for medicinal plants detection using machine
learning and deep learning techniques by classifying leaves using ResNet 50 is
a complex process that involves several important steps.

1. Data collection: Firstly, the data collection process is critical to the

success of the system. The project team must collect a large dataset of
medicinal plant leaves that are labeled with the corresponding plant
species. The dataset must be diverse, containing images with different
backgrounds, lighting conditions, and angles to ensure that the model
can generalize well to unseen data.

2. Data preprocessing:The next step is data preprocessing, which is

essential for improving the accuracy and speed of the model during
training and inference. This involves removing any noise from the
dataset, resizing the images to a fixed size, and normalizing the pixel
values to a common range.

3. Dataset splitting: The preprocessed dataset can then be divided into

training, validation, and test sets. Using backpropagation and
stochastic gradient descent, the ResNet 50 model's parameters are
optimized using the training set. The test set is used to assess the
model's performance on unobserved data, while the validation set is
used to monitor the model's performance during training and avoid
overfitting.

4. Model training: The ResNet 50 model is trained using the training set
in the model training method. During training, the model develops the
ability to identify the correct plant species by separating out pertinent
elements from the input photos. The problem of vanishing gradients is
lessened by the use of the ResNet 50 design, which also makes it
possible to train very deep neural networks.

37
 5. Model evaluation: Examining the model's performance on the test set
comes after it has been trained. Calculating metrics like accuracy,
precision, recall, and F1 score is necessary to determine whether the
model is capable of correctly classifying medicinal plant leaves into
the appropriate species.

6. Model optimization: The model can then be further tuned via transfer
learning based on the evaluation findings by the project team. In order
to increase a model's accuracy and generalization performance,
especially when working with fresh datasets containing various classes
of plants, this entails employing a pre-trained model and retraining it
on a new dataset.

In the context of traditional medicine, the computational method can aid in the
identification of medicinal plants and their potential uses, thereby contributing
to the preservation and promotion of traditional knowledge. In the field of
drug discovery, the identification and classification of medicinal plant species
can assist in the discovery of new drugs with therapeutic properties. In
agriculture, the computational method can help in the monitoring and
management of medicinal plant populations, thereby contributing to the
conservation of biodiversity.

Moreover, the use of ML & DL techniques in conjunction with ResNet 50 can

provide an efficient and accurate solution to the problem of medicinal plant
detection, allowing for the automatic identification of plant species and
reducing the need for manual identification, which can be time-consuming and
error-prone. The development of such systems can also promote
interdisciplinary collaboration between the fields of computer science,
biology, and traditional medicine, leading to new opportunities for research
and innovation.

38
 Chapter-4

PERFORMANCE ANALYSIS

Performance analysis is an important aspect of developing accurate and

efficient machine learning and deep learning models for medicinal plant
detection. The analysis typically involves evaluating the accuracy, speed, and
computational efficiency of the models. Performance indicators like precision,
recall, and F1 score can be used to gauge accuracy. Recall provides the
proportion of correctly detected medical plants out of all genuine medicinal
plants in the dataset, whereas precision shows the proportion of correctly
detected medicinal plants out of all detected plants. A harmonic mean of
memory and precision makes up the F1 score.
In addition to accuracy, the analysis may also focus on evaluating the speed
and computational efficiency of the models. This can include measuring the
training and inference time of the models, as well as the computational
resources required for training and deployment. Such an evaluation can help
identify hardware and software configurations that are optimal for training and
inference.
Overall, the goal of performance analysis in medicinal plant detection using
ML and DL methods is to develop models that can accurately and efficiently
identify medicinal plant species, while minimizing the computational
resources and training time required. This can help in the rapid and accurate
identification of medicinal plants, which can have significant applications in
the fields of traditional medicine, agriculture, and biodiversity conservation.

To conduct a thorough performance analysis of the Medicinal Plants Detection

Using ML & DL (by classifying leaves using ResNet 50) system, a range of
metrics can be used to evaluate the effectiveness of the model.

39
Precision, recall, and F1 score can all be employed as measurements for
accuracy. Recall represents the percentage of accurate forecasts for the target
class among all positive predictions, whereas precision measures the
percentage of accurate predictions for the target class among all positive actual
results. The harmonic mean of recall and precision is the F1 score.

In addition to accuracy, speed and computational efficiency are also essential

metrics to evaluate the system's performance. Training and inference time, as
well as computational resources required for training and deployment, can be
measured to determine the system's efficiency. Hardware and software
configurations can also be evaluated to identify optimal settings for training
and inference.
Assessing the system's adaptability to new datasets is crucial as well. Transfer
learning can be used to evaluate the model's performance and fine-tune it on
additional datasets with various plant classes. To make sure that the model is
not overfitting to the training data, cross-validation can also be performed.
Overall, a comprehensive performance analysis can help optimize the
Medicinal Plants Detection Using ML & DL system for accurate and efficient
plant identification while minimizing computational resources and training
time.

1. Accuracy: A crucial performance indicator for the Medicinal Plants

Detection Using ML & DL system is accuracy. It can be computed by
dividing the number of samples that were successfully categorized by
the overall number of samples in the dataset. It reflects the percentage
of correctly recognised plant species. High accuracy indicates that the
system can correctly identify and classify different types of medicinal
plant species, which is essential for their proper use in traditional
medicine and drug discovery. Therefore, optimizing the accuracy of
the system is crucial for its successful application in the fields of
medicine and agriculture. (Fig 2.4 shows the training accuracy
achieved while working)

40
 Fig 2.4 : Training Accuracy

2. Precision and Recall: Two crucial variables that are frequently

employed in classification issues are recall and precision. Recall is the
percentage of true positives among all real positive samples, whereas
precision measures the percentage of true positives among all positive
forecasts. These measurements can be used to assess the system's
potential biases and to make performance-enhancing adjustments.
Additionally, a single indicator of the system's total effectiveness can
be found in the F1 score, which is the harmonic mean of precision and
recall. For performance analysis in medicinal plant detection utilizing
ML and DL algorithms, other metrics including accuracy, confusion
matrix, and receiver operating characteristic (ROC) curve can also be
used.

3. F1 Score: The harmonic mean of recall and precision, known as the F1

score, can be used to assess the system's overall performance. It strikes
a compromise between recall and precision, and a high F1 score shows
that the system is operating effectively. The F1 score is determined by
multiplying two by (precision + recall) / (precision + recall).
(Fig 2.4 : Training accuracy and Fig 2.5 : Classification Report both
shows the model is overfitted)

41
 Fig 2.5 : Classification Report

4. Confusion Matrix: A confusion matrix is a useful tool to evaluate the

performance of a classification model. It provides a detailed
breakdown of the number of correctly and incorrectly classified
samples for each class. This can help identify which classes the system
is struggling with and may require additional training or fine-tuning.
The matrix is typically represented as a table where each row
represents the actual class and each column represents the predicted
class. The cells in the table contain the number of samples that belong

42
 to a particular class and were classified correctly or incorrectly. We can
determine different performance indicators, including accuracy,
precision, recall, and F1 score, from the analysis of the confusion
matrix, which can aid in further system optimization.

5. Training and Validation Loss: Training accuracy and validation

accuracy are two evaluation metrics that may be used to continuously
monitor the training process and make sure the system is not
overfitting or underfitting (Fig 2.6 shows Graph (Loss) for
visualization). When a system performs exceptionally well on training
data but poorly on new data, this is a sign of overfitting since it means
the system has memorized the training data rather than discovering its
underlying patterns. Underfitting, on the other hand, happens when the
system is overly straightforward and fails to recognise the key patterns
in the data, leading to subpar performance on both the training and
validation sets. The performance of the model's generalization can be
enhanced by adjusting the hyperparameters while keeping an eye on
these measures.

43
 Fig 2.6 : Graph(Loss) for visualization

Overall, it is essential to perform a thorough performance analysis to evaluate

the system's ability to accurately classify medicinal plant species using ML &
DL techniques. The analysis should include various metrics such as accuracy,
precision, recall, F1 score, and a confusion matrix to identify any potential
biases or weaknesses in the system. Monitoring the training process is also
crucial to ensure that the system is not overfitting or underfitting. (Fig 2.7
shows the early stopping of model as accuracy does not seems to improve
after 8 epochs.)A comprehensive performance analysis can help optimize the
system for accurate and efficient plant detection while minimizing
computational resources and training time.

44
 Fig 2.7 : Showing Early stopping of model.

Limitations

● Model performs well at training data showing 93% accuracy but does
not perform well at testing data.
● Classification Report ( Fig 2.5 : Classification Report) shows that the
model fails to identify some classes of medicinal plants.
● Model seems to be overfitting right now.

45
 Chapter-5

CONCLUSIONS

5.1 Conclusions

Using ResNet 50 to categorize the leaves of medicinal plants, machine

learning (ML) and deep learning (DL) approaches are being used to detect
medicinal plants, which is an intriguing and promising methodology that has
gained traction in recent years. With the rapid advancement in ML and DL,
image recognition systems have become increasingly accurate and efficient,
making it possible to develop highly accurate and efficient models for
medicinal plant detection.

This approach has opened up new opportunities in the field of medicinal plant
detection, where accurately identifying plants is essential. By training a
ResNet 50 model on a large dataset of medicinal plant leaves, the model can
learn to accurately classify different plants based on their unique leaf
characteristics. This can be useful in the pharmaceutical industry for drug
discovery and in traditional medicine for identifying the correct plants for
treatment.

Although this strategy has great potential, there are still some obstacles to be
overcome. The necessity for high-quality plant leaf photos and a complete
library of different medicinal plant species is one of the major issues.
Additionally, further research is needed to optimize the ML and DL techniques
used in medicinal plant detection to increase their accuracy and efficiency.

To address these challenges, researchers are exploring innovative ways to

improve image quality and increase the size of the dataset used for training the
models. One approach involves the use of data augmentation techniques to

46
artificially increase the size of the dataset by creating additional images
through various image processing techniques.

In conclusion, the classification of medicinal plant leaves using ResNet 50

employing ML and DL approaches is a promising strategy that has the
potential to revolutionize the study of plant identification and medicinal
plants. With continued research and development, this approach could lead to
the discovery of new medicinal plants and the development of more effective
treatments for a variety of diseases. (Fig 2.8 shows Deployed Project with user
friendly interface )

Fig 2.8 : Deployed web application

47
5.2 Future Scope

The use of machine learning (ML) and deep learning (DL) techniques for
medicinal plant detection by classifying their leaves using ResNet 50 has
immense potential for development in the future. Here are some potential
areas of growth:

1. Increasing the accuracy and efficiency of the model: Researchers can

continue to refine the models by using larger and more comprehensive
datasets, improving the training algorithms, and developing more
efficient hardware. Further exploration and optimization of various ML
and DL techniques can also enhance the models' efficiency and
accuracy.

2. Integrating other modalities: Although leaf classification is a reliable

method for medicinal plant detection, integrating other modalities such
as stem and flower morphology, chemical profiles, and genetic data
can provide a more complete and accurate identification of the plant
species. By combining various modalities, researchers can develop
more robust and reliable classification systems.

3. Developing portable and user-friendly applications: Researchers can

develop user-friendly applications that can be easily used by
non-experts in the field to identify medicinal plants in the field. Such
applications can be developed for smartphones, tablets, or other
portable devices, making it easier for people to access information
about medicinal plants.

4. Expanding the use of medicinal plant detection: In addition to

traditional medicine and the pharmaceutical industry, ML and DL
techniques for medicinal plant detection can also be applied to other
areas such as food and agriculture, environmental conservation, and

48
 bioprospecting. By exploring these new areas of application,
researchers can uncover new uses for medicinal plants and expand
their potential benefits.

5. Addressing ethical and legal concerns: As the use of ML and DL

techniques for medicinal plant detection expands, it is important to
consider ethical and legal concerns, such as the need to protect
traditional knowledge, prevent biopiracy, and ensure fair and equitable
distribution of benefits. Researchers can work with stakeholders to
develop guidelines and regulations that ensure ethical and legal
practices are followed.

Overall, the future scope of using ML and DL techniques for medicinal plant
detection by classifying leaves using ResNet 50 is vast and promising, with
potential applications across a range of industries and fields. By addressing the
challenges and concerns, researchers can harness the full potential of these
techniques and revolutionize the field of medicinal plant research.

49
 REFERENCES

[1] B. Dudi and V. Rajesh, “Medicinal plant recognition based on CNN and
Machine Learning,” International Journal of Advanced Trends in Computer
Science and Engineering, vol. 8(4), pp. 999–1003, (2019). Available at:
https://doi.org/10.30534/ijatcse/2019/03842019.

[2] R. Azadnia, M. Al-Amidi, H. Mohammadi, M. Cifci, M. Daryab and E.

Cavallo, “An AI based approach for medicinal plant identification using deep
CNN based on Global average pooling,” Agronomy, vol. 12(11), pp. 2723,
(2022). Available at: https://doi.org/10.3390/agronomy12112723.

[3] R. U. Rao, M. S. Lahari, K. P. Sri, K. Y. Srujana and D. Yaswanth,

“Identification of medicinal plants using Deep Learning,” International
Journal for Research in Applied Science and Engineering Technology, vol.
10(4), pp. 306–322, (2022). Available at:
https://doi.org/10.22214/ijraset.2022.41190.

[4] J. Abdollahi, “Identification of Medicinal Plants in Ardabil Using Deep

learning : Identification of Medicinal Plants using Deep learning.” 2022 27th
International Computer Conference, Computer Society of Iran (CSICC), pp.
1–6, (2022). Available at: https://doi.org/10.1109/CSICC55295.2022.9780493

[5] Image Source :-

https://towardsdatascience.com/the-annotated-resnet-50-a6c536034758

[6] Dataset:-

Link: https://data.mendeley.com/datasets/nnytj2v3n5/1
DOI: 10.17632/nnytj2v3n5.1

50
 APPENDICES

Code:

51
52
53
54
55
56
57