MEDICINAL PLANT IDENTIFICATION

USING MACHINE LEARNING

DOMAIN : IMAGE PROCESSING

Dr.A.Bhavani Sankar M.Mugesh, K.Mamannan,

S.G.Mohan, D.Dhasarathan
Professor and Head,
Department of Electronics and
Department of Electronics
Communication Engineering,
andCommunication Engineering.
Anjalai Ammal Mahalingam
Anjalai Ammal Mahalingam
Engineering College,
Engineering College, Kovilvenni,
Kovilvenni,Thiruvarur.
Thiruvarur

ABSTRACT : transfer learning to leverage pre-

trained CNN models for improved
Medicinal plants have been used classification performance. Our results
for centuries in traditional medicine demonstrate the effectiveness of the
systems for their therapeutic proposed approach in accurately
properties. Accurate identification of identifying medicinal plants, with
these plants is essential for ensuring potential applications in healthcare,
their safe and effective use. In this biodiversity conservation, and
project, we propose a machine traditional knowledge preservation.
learning-based approach for the This project highlights the synergy
automated identification of medicinal between machine learning and
plants from images. We utilize a botanical sciences, showcasing how
dataset of images containing various advanced technologies can enhance
medicinal plant species and employ our understanding and utilization of
convolutional neural networks (CNNs) medicinal plants for the benefit of
to extract features and classify plants society.
into their respective species.
Additionally, we explore the use of
I INTRODUCTION : medicinal plant identification, paving
the way for more efficient and accurate
Medicinal plants have been used use of these important natural
for centuries in various traditional resources.
medicine systems for their therapeutic
properties. The identification of these Medicinal plants play a crucial
plants is crucial for ensuring their safe role in traditional medicine systems
and effective use in healthcare. and have been used for millennia to
However, manual identification treat various ailments and promote
methods can be time-consuming and health. However, the accurate
require expertise in botany and identification of medicinal plants is
pharmacology. essential to ensure their safe and
effective use. Misidentification can
Machine learning, particularly lead to serious consequences,
deep learning techniques such as including ineffective treatments,
convolutional neural networks adverse reactions, and even poisoning.
(CNNs), has shown great promise in
automating the identification of Traditional methods of
medicinal plants. By analyzing images identifying medicinal plants rely on
of plant specimens, machine learning the expertise of botanists and
models can learn to recognize patterns herbalists, who use morphological
and features that distinguish different characteristics such as leaf shape,
plant species. This can significantly flower color, and fruit structure to
speed up the identification process and classify plants. While effective, these
make it more accessible to non- methods can be time-consuming and
experts. require specialized knowledge.
In this project, we explore the Advances in technology,
application of machine learning for the particularly in the field of machine
automated identification of medicinal learning, offer new opportunities for
plants. We aim to develop a system that automating the identification of
can accurately classify medicinal plant medicinal plants. Machine learning
species based on images, providing a algorithms, especially convolutional
valuable tool for healthcare neural networks (CNNs), can analyze
professionals, researchers, and images of plants and learn to recognize
conservationists. This project patterns and features that distinguish
highlights the potential of machine one species from another. This
learning to revolutionize the field of approach has the potential to
significantly accelerate the species. However, the manual
identification process and make it process is laborious and time-
more accessible to a wider range of consuming. Recently,
users.
technological development has
In this project, we propose to called for more efficient methods
develop a machine learning-based to meet species’ identification
system for the automated identification
requirements, such as developing
of medicinal plants. By training a CNN
on a dataset of images of medicinal digital-image-processing and
plants, we aim to create a model that pattern-recognition techniques.
can accurately classify plants into their Despite several existing studies,
respective species. This system has the there are still challenges in
potential to revolutionize the field of automating the identification of
medicinal plant identification, plant species accurately. The
providing a fast, reliable, and
proposed model achieved 87%
accessible tool for healthcare
professionals, researchers, and and 84% Top-1 accuracies on a
conservationists. test set for the private and public
datasets, respectively, which is
II RELATED WORKS :
more than a 10% accuracy
Automated Real-Time improvement compared to the
Identification of Medicinal baseline model. During real-time
Plants Species in Natural system testing on the actual
Environment Using Deep samples, using our mobile
Learning Models - The application, the accuracy slightly
identification of plant species is dropped to 78.5% (Top-1) and
fundamental for the effective 82.6% (Top-5), which may be
study and management of related to training data and testing
biodiversity. In a manual conditions variability.
identification process, different
Identification of Medicinal
characteristics of plants are
Plant Using Machine Learning
measured as identification keys
Approach From Vedic times
which are examined sequentially
plants have been used as a source
and adaptively to identify plant
of medicine in ayurveda. In the
preparation of ayurvedic cultivated on large scale owing to
medicine, identification of correct their high market demand,
plant is the most important step, conservation value and medicinal
which have been done manually. properties. Identification of these
Due to demand of mass plants in the field requires
production, Identification of these taxonomic skills, which is one of
plants automatically is important. the major bottlenecks in the
In this paper we have been conservation and management of
implement a technique for these plants. In this background, a
medicinal plant identification hyperspectral library of the above
using random forest algorithm, an three medicinal plants has been
ensemble supervise machine prepared by collecting its spectral
learning algorithm based on color data from Himachal Pradesh and
,texture and geometrical features Uttarakhand states of Indian
Himalaya. The Random forest
Identification and classification
(RF) model was implied on the
of medicinal plants of the
spectral data for the classification
Indian Himalayan region using
Hyperspectral remote sensing of these medicinal plants which
resulted in training accuracy of
and random forest techniques
84.39 % (kappa coefficient =
The Indian Himalaya region 0.72) and testing accuracy of
harbours approximately 1748 85.29% (kappa coefficient =
plants, which have been 0.77). This RF classifier has
prioritized for medicinal usage. identified Green (555-598 nm),
To ease the pressure on these red (605 nm), and NIR (725–840
plants, the government of India is nm) wavelength regions suitable
encouraging the in-situ for discrimination of the above
cultivation of medicinal plants. species.
As a consequence, Saussurea
costus, Valeriana jatamansi, and
Picrorhiza kurroa are some of the
important crops which are being
III EXISTING METHOD : Disadvantages of Existing
method
Existing methods for identifying
medicinal plants typically rely on Subjectivity: Traditional
a combination of traditional and methods such as taxonomic keys
modern techniques. Traditional and visual comparison rely
methods involve the use of heavily on the expertise and
taxonomic keys, herbarium subjective judgment of the person
specimens, and field guides, identifying the plant. This can
where experts compare physical lead to inconsistencies and errors
characteristics of plants to known in identification, especially
species. While effective, these among non-experts.
methods can be time-consuming Time-Consuming: Traditional
and require specialized
methods are often time-
knowledge. consuming, particularly when
Modern approaches leverage dealing with a large number of
technology to streamline the plant specimens. The process of
identification process. Image- comparing physical
based identification, for example, characteristics and consulting
uses digital image processing reference materials can be labor-
techniques to analyze plant intensive and inefficient.
images and extract features for
Limited Accessibility:
classification. Machine learning
Traditional methods require
algorithms, such as convolutional
access to specialized knowledge,
neural networks (CNNs), are
reference materials, and
often employed to improve
sometimes, physical specimens.
accuracy and efficiency. DNA This limits their accessibility to
barcoding is another modern
non-experts and those without
method, which involves access to botanical resources.
sequencing a short DNA region
from a plant specimen to identify Dependency on Physical
its species. Characteristics: Traditional
methods rely on the observation
of morphological features, which IV PROPOSED METHOD :
can vary depending on factors
The proposed method for this
such as growth stage,
project involves using MATLAB
environmental conditions, and
for the automated identification of
genetic variation. This can lead to medicinal plants from images.
misidentification, especially in
The process begins with acquiring
cases where plants exhibit similar a dataset of images containing
physical characteristics.
various medicinal plant species.
Limited Scalability: Traditional These images are preprocessed to
methods are not easily scalable to enhance their quality and
large-scale plant identification normalize them for analysis.
projects, such as biodiversity Next, a convolutional neural
surveys or conservation efforts,
network (CNN) architecture is
due to their manual and labor- designed and implemented using
intensive nature.
MATLAB's Deep Learning
Lack of Standardization: Toolbox. The CNN is trained on
Traditional methods can lack the preprocessed images to learn
standardization, with different the features that distinguish
experts or resources providing different plant species. Transfer
varying identification results. learning techniques may also be
This can lead to inconsistencies in employed to leverage pre-trained
plant identification across models and improve
different studies or regions. classification performance.
Cost: DNA barcoding, while Once the CNN is trained, it can be
highly accurate, can be expensive used to classify new images of
and require specialized medicinal plants into their
equipment and expertise, making respective species. The
it less accessible for routine plant classification results can be
identification projects. visualized and analyzed to
evaluate the performance of the
model. The proposed method
aims to provide a fast, accurate, field of botanical sciences and
and automated approach to healthcare. Through this project,
identifying medicinal plants, with we have demonstrated the
potential applications in feasibility and effectiveness of
healthcare, biodiversity using machine learning
conservation, and traditional algorithms, particularly
knowledge preservation. convolutional neural networks
(CNNs), to automate the
IV BLOCK DIAGRAM :
identification of medicinal plants
from images.
By leveraging MATLAB's
powerful tools for image
processing and deep learning, we
have created a robust and accurate
system capable of classifying
medicinal plant species with high
accuracy. This system has the
potential to revolutionize the way
medicinal plants are identified,
providing a faster, more efficient,
and more accessible alternative to
traditional methods.
Overall, the development of a
medicinal plant identification
V RESULT :
system using MATLAB and deep
VI CONCLUSION: learning represents a significant
In conclusion, the development of step forward in our ability to
a medicinal plant identification harness technology for the study
system using MATLAB and deep and conservation of medicinal
learning techniques represents a plants, ultimately contributing to
significant advancement in the
improved healthcare outcomes Trends in Electronics and
and environmental sustainability. Informatics (ICOEI).
VII REFERENCES : [4] Automated Plant
Identification Through Deep
[1] MPInet: Medicinal Plants
Learning with Particular Focus
Identification using Deep
On Medicinal Plants.Sunil
Learning.
Bhutada; Ch. Sreeja Reddy; R.
Preethi Salian K; Shrisha H.S.; Sparsha Reddy; S. Vaishnavi;
Supriya Salian; Karthik K. 2023 Goutham Kumar 2023
7th International Conference on International Conference on
Electronics, Communication and Advanced Computing
Aerospace Technology (ICECA). Technologies and Applications
[2] FloraMediVision: A (ICACTA).
Medicinal Plant Leaf [5] AI Based Indigenous
Identification System using Medicinal Plant Identification.
Computer Vision. AnuPaulson; S
PradnyaPatil;Kaustubh Ravishankar.2020AdvancedCom
Kulkarni;Minal Sonkar; Priyanka putingandCommunicationTechno
Deshmukh 2023 6th International logiesforHighPerformanceApplic
Conference on Advances in ations (ACCTHPA).
Science and Technology
[6] Ayurvedic Plants
(ICAST).
Identification based on Machine
[3] Identification of Medicinal Learning and Deep Learning
Plant's and Their Usage by Using Technologies . Rajani S; Veena
Deep LearningC. M.N . 2022 Fourth International
Amuthalingeswaran;M. Conference on Emerging
Sivakumar;P. Renuga;S. Research in Electronics,
Alexpandi;J. Elamathi;S. Computer Science and
Santhana Hari2019. 2022 3rd Technology (ICERECT
International Conference on