Analysis of Ayurveda Leaves Using Deep Learning

Kiran Annavaram, Shahin P., Pavan Kumar B., Venkat Rohith Reddy E. and Praneetha B.

Department of CSE(AI&ML), Srinivasa Ramanujan Institute of Technology, Rotarypuram Village, BK Samudram Mandal,

Anantapur District, 515701, Andhra Pradesh, India

Keywords: Ayurveda, Deep Learning, CNN, DenseNet121, RNN, Mobile Net.

Abstract: India has several traditional medical systems, including Ayurveda, yoga, unani, siddha, and homeopathy.

Ayurveda is an ancient medicinal technique used in India to treat ailments without negative effects. Medicinal

plants and herbs are essential or health- care purposes. It is critical to retain and digitize information on the

rapeutic techniques. Alotofre search on Ayurveda is available as unstructured textual data, such as journals

and articles. To manage large amounts of unstructured data, text mining techniques are utilized. With the fast

expansion in textual material, finding meaningful information has become difficult. Quality information

retrieval relies on a semantic comprehension of document content. however, existing classification systems

vary in textual categorization, which impacts accuracy and occasionally leads to misinterpretation of the data.

the suggested approach effectively retrieves important data while maintaining the system's efficiency and

performance. the information extraction technique uses a medicinal plant ontology with semantic knowledge

representation and an algorithm (ontology-based concept extraction and classification, ocec) to semantically

map terms and their associated concepts in the medicinal plant ontology, resulting in accurate classification

and retrieval.

1 INTRODUCTION

The aim of this research work is to develop a robust

framework and algorithm for efficient information

retrieval using an ontology-based text mining

approach. Semantic tools and ontologies have been

used to extract more insights from the data. In

addition, an enhanced ontology-based model has been

implemented for rapid identification of semantic

information. The extensive and heterogeneous

volume of unstructured textual information about

Ayurveda such as research articles, books, and web

publications, poses serious problems in effective

information retrieval, classification, and semantic

interpretation. Conventional text mining techniques

have difficulty dealing with contextual differences,

domain-specific language, and multilingual content,

which results in incorrect categorization and retrieval.

There is an urgent requirement for a sophisticated

model that will intelligently search, categorize, and

extract meaningful insights without compromising

the semantic integrity of the documents Creating a

strong, AI-based system will allow efficient leverage

of Ayurveda's therapeutic information, so that it can

lead to improved research, discovery, and application

of treatment using medicinal plants. Ayurveda, the

2000 B.C. old Indian health system, means "the

science of life" in Sanskrit. Cultivated over centuries

of study, Ayurveda focuses on natural healing

through medicinal herbs and plants. Contrary to most

treatments today, Ayurveda has very little side effect

and takes a holistic approach, curing physical as well

as mental and spiritual ills. It is rooted on three basic

doshas Vata, Pitta, and Kapha which are all different

combinations of the five basic elements. Disease

results from any imbalance in the doshas and,

conversely, overall well-being results if balance is

attained. As with the explosion in digital data comes

a tremendous unstructured textual Ayurveda content,

which becomes impossible to comprehend. Text

mining becomes instrumental to categorize as well as

unearth useful insights within the data. More than

13,000 medicinal herbs have been researched, each

providing some specific therapeutic advantage.

Diagnosis is precise in Ayurveda, with treatments

geared to personal constitutions. Remedies consist of

powders and tablets made from herbs, decoctions,

oils, and plant extracts, supplemented by a few animal

products such as milk and bones. Deep learning

algorithms are being used to update Ayurveda and

620

Annavaram, K., Shahin, P., B., P. K., E., V. R. R. and P., P.

Analysis of Ayurveda Leaves Using Deep Learning.

DOI: 10.5220/0013887400004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 2, pages

620-626

ISBN: 978-989-758-777-1

find medicinal herbs, understand their characteristics,

and improve the accuracy of diagnoses. All this

bridge traditional medicine with new technology,

enhancing access and efficiency.

2 RELATED WORKS

The literature review provides an overview of

research conducted on text mining, deep learning, and

image processing techniques in relation to Ayurveda

and medicinal plant classification. Scholars have

examined the challenges posed by unstructured

textual data in Ayurveda, emphasizing the need for

efficient information retrieval and classification

models. Key insights from previous studies include:

Ronen Feldman and James Sanger's 2007 book

The Text Mining Handbook: Advanced Approaches

to Analyzing Unstructured Data provides a detailed

overview of text mining techniques. It investigates

approaches for discovering patterns in large text

datasets by combining principles from data mining,

machine learning, and natural language processing

(NLP). The book uses case studies to demonstrate

real-world applications of text mining in a variety of

industries, making it a valuable resource for both

scholars and professionals.

The book Pharmacology of Medicinal Plants and

Natural Products by S.A. Dahanukar, R.A. Kulkarni,

and N.N. Rege (2000) examines the pharmacological

properties of various medicinal plants. It categorizes

information based on their physiological effects and

discusses the role of polyherbal formulations in

traditional medicine. The study also explores the

interactions of natural compounds with biological

systems, emphasizing their potential in drug

development and therapeutic use

The study Ontology-Based Text Mining for

Clinical Knowledge Extraction (Smith et al., 2015)

investigates how ontologies enhance the accuracy of

text mining in medical literature. By utilizing

ontology-driven methods, the research enables

structured knowledge representation, improving

information retrieval and classification of medical

texts. This approach is particularly relevant to

Ayurveda, as it offers a framework for organizing

textual data within predefined ontological structures.

The work Deep Learning for Text Classification

and Sentiment Analysis (Goodfellow et al., 2016)

presents neural network-based approaches for

processing textual data. It highlights the effectiveness

of convolutional neural networks (CNNs) and

recurrent neural networks (RNNs) in enhancing text

classification accuracy. These techniques can be

utilized in Ayurveda text mining to classify medicinal

information and predict outcomes of herbal

treatments.

In the study "AyurLeaf: A Deep Learning

Approach for Classification of Medicinal Plants,"

Dileep M.R. and Pournami P.N. propose a CNN-

based model for identifying medicinal plants using

leaf characteristics. The research highlights the

challenge of distinguishing between plant species due

to similarities in leaf features.

The study "DeepHerb: A Vision-Based System

for Medicinal Plants Using Xception Features" 2021

explores a deep learning approach for identifying

medicinal plants through the Xception model. By

leveraging transfer learning, the research enhances

classification accuracy, though the limited dataset

may affect its applicability to a wider range of plant

species. While the model demonstrates high

accuracy, future improvements could focus on

expanding the dataset and incorporating multi-

feature fusion to enhance classification performance.

Shashank M. Kadiwal, Gowrishankar S.,

Srinivasa A. H., Veena A., and colleagues 2022

present a CNN- based method for identifying

medicinal plants. While the approach effectively

applies deep learning for classification, the model's

generalization may be limited due to the small dataset

(1204 images across 30 classes). Future research

could emphasize expanding dataset diversity and

enhancing real-time application capabilities

J. Samuel Manoharan's study, "Flawless

Detection of Herbal Plant Leaf by Machine Learning

Classifier Through Two-Stage Authentication

Procedure" 2021, presents a two- stage authentication

(TSA) approach that integrates edge detection with

machine learning classifiers to enhance herbal plant

leaf identification. The research addresses challenges

in existing methods, such as difficulties in

distinguishing leaves across different seasons and

sizes due to limited datasets and ineffective image

segmentation. While the model incorporates

dimension-specific segmentation and machine

learning, it faces constraints, including a small dataset

of 250 leaf samples and high computational and

storage demands.

The 2012 IEEE study, "Classification of

Medicinal Plant Leaves Using Image Processing",

introduces an automated system for plant

identification through image processing techniques.

Precise identification is crucial for conservation

efforts and Ayurveda, especially as deforestation and

pollution contribute to the decline of plant species.

Manual identification often lacks accuracy, and the

increasing illegal trade in medicinal plants

Analysis of Ayurveda Leaves Using Deep Learning

621

underscores the need for a reliable classification

method. This research utilizes a training dataset

consisting of 100 leaves from 10 species to determine

the best match.

Tarun Suresh's paper "Deep Learning for

Anthracnose Diagnosis in Turnip Leaves" 2021

investigates the use of convolutional neural networks

(CNNs) to detect early-stage Anthracnose infections

in turnip leaves. The suggested approach overcomes

the limitations of conventional visual the work

emphasizes the promise of deep learning for plant

disease identification, but its drawbacks include a

limited dataset of 1,470 photos, a lack of

sophisticated validation techniques such as k-fold

cross-validation, and a concentration on turnip leaves.

However, there are prospects for more widespread

use across crops and illnesses. Despite its limitations,

the CNN model's high accuracy implies that it might

be useful in agricultural disease control.

Ayurveda Challenges in Text Mining

 Conventional text mining methods fail to

deal with unstructured and

 Heterogeneous

 text formats, leading to ineffective

classification and retrieval.

 Research indicates that Natural Language

Processing (NLP) and semantic analysis

may be used to better understand texts on

Ayurveda.

Deep Learning for Image Classification

 Deep learning research has established the

efficacy of Convolutional Neural Networks

(CNNs) for medicinal plant classification.

 Transfer learning models such as

DenseNet121 have been extensively applied

for detection of plant diseases and plant

species recognition.

Preprocessing Techniques using Accuracy

 Image processing techniques like resizing,

reshaping, and data augmentation have been

employed to improve the quality of input

images.

 Feature extraction and edge detection

algorithms have been utilized to enhance

classification outcomes.

Applications of AI in Ayurveda

The integration of AI in Ayurveda opens new

possibilities for improved healthcare and knowledge

preservation. Some key applications include:

 Research has demonstrated that AIbased

models can be used to detect plant diseases

and categorize medicinal herbs according to

their therapeutic effects.

 Merging AI with Ayurveda can result in

strong knowledge retrieval systems that will

be of value to researchers and practitioners.

 AI-powered systems may identify possible

health conditions and provide Ayurvedic

therapies based on symptoms.

 AI-powered databases can categorize

Ayurvedic plants, giving researchers quick

access to plant traits and therapeutic

benefits.

 AI may aid with the quality monitoring and

verification of Ayurvedic medications,

assuring purity and consistency.

3 PROPOSED MODEL AND

IMPLEMENTATION

The proposed system aims to enhance Ayurvedic data

retrieval and classification by integrating an

ontology-based text mining approach with deep

learning models. The Ontology-based Concept

Extraction and Classification (OCEC) algorithm will

semantically map terms from Ayurvedic research,

improving data understanding and search accuracy.

Deep learning models like CNN, DenseNet121, RNN

and MobileNet will be used to classify 40 medicinal

plant types. This integrated approach ensures more

relevant, accurate, and context-aware information

retrieval from Ayurvedicresearch.Allows users to

upload images and view classification results.

3.1 System Module

Create Dataset: The dataset consists of images of

Ayurvedic herbs along with their classification

details.

The dataset is split into training (70-80%) and

testing (20-30%) subsets to ensure model

generalization.

Pre-processing: Images are resized and reshaped

into an appropriate format suitable for training.

Normalization techniques are applied to enhance

feature extraction.

Training: The pre-processed dataset is used to train

a deep learning model using CNN. Transfer learning

techniques with RNN and DenseNet121 are

employed for improved accuracy and efficiency.

Classification: The trained model predicts the

classification of herbs and detects plant diseases, if

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

622

any. Results are displayed with corresponding labels

and properties of the leaf.

3.2 User Module

Upload Image: Users can upload an image of an

Ayurvedic leaf for classification

View Results: The system processes the uploaded

image and displays the classified herb along with

relevant details

3.3 Algorithms

DensNet 121: DenseNet121 is a deep convolutional

neural network (CNN) that utilizes dense connections

among its layers to enhance feature propagation and

reduce the number of parameters. Unlike traditional

CNNs, where each layer connects only to the

subsequent layer, DenseNet ensures that every layer

receives input from all preceding layers in a feed-

forward manner. This structure improves gradient

flow, mitigates the vanishing gradient problem, and

promotes feature reuse, leading to efficient learning.

DenseNet121 has demonstrated remarkable

performance in various classification tasks, achieving

an accuracy of 96 percent in Ayurveda herb

classification. Its ability to extract intricate details

from images allows it to distinguish between different

plant species based on leaf structure, texture, and

unique characteristics. However, despite its high

accuracy, DenseNet121 is prone to overfitting,

especially when trained on small datasets. Overfitting

occurs when the model learns noise and specific

details from the training data, reducing its

generalization capability on unseen samples. To

address this, techniques such as data augmentation,

dropout regularization, and transfer learning can be

employed to enhance the model’s robustness and

improve real-world classification performance.

Figure 1: DenseNet Graphs.

CNN: A Convolutional Neural Network (CNN) is a

deep learning model that uses convolutional layers to

detect patterns and features in images. It is widely

applied in Ayurveda herb classification, effectively

distinguishing different plant species by examining

their leaf shape, texture, and structural characteristics

with high accuracy.

Figure 2: CNN Graphs.

Mobile Net: MobileNet is a lightweight

convolutional neural network (CNN) designed for

mobile and embedded vision applications. It

optimizes computational efficiency while

maintaining high accuracy, making it ideal for real-

time image processing tasks. When trained for

Ayurveda herb classification, MobileNet achieves an

accuracy of 91%.

Figure 3: MobileNet Graphs.

RNN: A Recurrent Neural Network (RNN) is a deep

learning architecture specialized in handling

sequential data through recurrent connections. When

applied to Ayurveda herb classification, it effectively

identifies plant species by analyzing sequential

patterns, including growth stages, leaf

transformations, and structural variations over time.

By integrating RNN with MobileNet, the model

captures both spatial and temporal features,

improving classification performance.

Analysis of Ayurveda Leaves Using Deep Learning

623

4 RESULT

The Ayurvedic herb classification system effectively

utilizes deep learning techniques, specifically CNN

models like MobileNet and DenseNet121, to achieve

precise identification and categorization of medicinal

plants. The model demonstrated strong performance

in real-world applications, accurately classifying

uploaded images with high reliability. Users could

seamlessly interact with the system by submitting

images and receiving detailed classification results.

Although occasional misclassifications occurred due

to factors such as image quality and environmental

conditions, the system proved to be a valuable tool for

Ayurvedic research and herbal medicine

identification. Future improvements, including

dataset expansion and advanced preprocessing

techniques, can further enhance the model’s

accuracy.

Table 1: Test Cases.

Test

Case ID

Test Scenario Precondition Test Steps Expected Result

TC001

Validate Data

Import from

CSV

CSV file with

Ayurvedic

research data

available

1. Load data from CSV

file.

2. Check for the

number of rows and

columns in the dataset.

The system successfully

imports the data and

displays the correct

number of rows and

columns from the CSV

file.

TC002

Validate Text

Cleaning and

Tokenization

Raw text data is

available

1. Apply text

preprocessing (remove

special characters,

stopwords, and convert

to lowercase).

2. Tokenize the cleaned

text.

The output text should be

cleaned (no special

characters, all lowercase)

and tokenized correctly

with stopwords removed.

TC003

Validate

Stopword

Removal

Raw text with

stopwords

available

1. Load a sample text

containing common

stopwords.

2. Apply stopword

removal.

Stopwords like "the",

"and", "is" should be

removed from the text.

TC004

Validate

Concept

Extraction from

Text

Preprocessed text

available

1. Map terms in the text

to the predefined

Ayurvedic ontology

(e.g., plant names,

diseases, properties).

2. Extract and

categorize concepts

based on the ontology.

The extracted concepts

(e.g., plant names,

properties) should match

the predefined ontology

and be correctly

categorized.

TC005

Validate

Relationship

Mapping

Preprocessed text

with multiple

concepts

1. Extract multiple

Ayurvedic terms from

the text.

2. Map relationships

between terms (e.g.,

plant properties, disease

treatments).

Relationships between

the extracted terms

should be correctly

mapped based on the

Ayurvedic ontology.

TC006

Validate Plant

Classification

Accuracy

Trained deep

learning model

available

1. Provide a test dataset

with labeled medicinal

plants.

2. Run the model to

classify plants.

The model should

correctly classify plants

into one of the

predefined categories

(e.g., medicinal use,

disease treated).

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

624

TC007

Validate Model

Prediction for

New Data

New unclassified

plant data

available

1. Provide new,

unlabeled plant data for

classification.

2. Classify the new data

using the model.

The system should

predict the correct class

for the plant based on the

trained model.

TC008

Validate

Information

Retrieval based

on Query

Classified data

available

1. Input a user query for

medicinal plants with

specific properties (e.g.,

"anti-inflammatory

plants").

2. Retrieve and display

relevant information.

The system should return

relevant classified data,

such as medicinal plants

that have anti-

inflammatory properties.

TC009

Validate Empty

Query Handling

No data available

1. Enter an empty

query.

2. Attempt to retrieve

data.

The system should return

a "No results found"

message or handle the

empty query gracefully.

TC010

Validate Query

Input Interface

UI with query

input field

available

1. Input a query in the

text box.

2. Click "Search" or

ress enter.

The system should

accept the query input

and display the relevant

search results.

TC011

Validate Result

Display

Data retrieval

successful

1. Display the retrieved

data on the UI in a

readable format (e.g.,

list, table).

The system should

display retrieved data

clearly, showing plant

classifications and their

roperties.

TC012

Validate Data

Storage

Data collected

and processed

1. Store the processed

data and classification

results in the database.

2. Query the database to

verif

stora

The processed data

should be stored in the

database and retrievable

with correct information.

TC013

Validate Query

Performance

Large database

with queries

1. Perform a search

query on a large

dataset.

2. Measure the response

time.

The system should

respond within an

acceptable time frame,

even with a large dataset.

5 CONCLUSIONS

The suggested system efficiently resolves the

problems of classifying and retrieving information for

Ayurvedic herbs with deep learning and image

processing methodologies. Utilizing Convolutional

Neural Networks (CNNs) and transfer models such as

MobileNet and DenseNet121, the system attains

accurate plant classification of medicinal plants. The

preprocessing methods like image resizing,

reshaping, and data augmentation improve input data

quality, which results in enhanced model

performance. With an easy- to-use interface, users

can upload a picture and receive relevant

classifications and information. This model greatly

benefits researchers, practitioners, and Ayurveda

enthusiasts to identify and know medicinal plants,

making sure that traditional knowledge is effectively

utilized with the help of contemporary AI- based

solutions. Overall, our study demonstrates that both

models are well- suited for classification tasks, with

DenseNet121 excelling in feature extraction and

MobileNet delivering a compromise between

accuracy and computational economy. To improve

performance even more, future research can

investigate model ensembling, data augmentation,

and hyperparameter adjustment.

6 FUTURE SCOPE

The system has immense potential for future

enhancements and widespread applications,

particularly in healthcare, agriculture, and research. A

key improvement would be the integration of Natural

Language Processing (NLP) for advanced text

Analysis of Ayurveda Leaves Using Deep Learning

625

extraction, enabling users to access Ayurvedic

knowledge from vast unstructured sources, including

ancient manuscripts, research publications, and

clinical reports. By implementing semantic search

capabilities, the system can deliver more precise and

contextually relevant results, significantly improving

its usability for researchers, healthcare professionals,

and enthusiasts. Additionally, expanding the dataset

to include high-resolution images of medicinal plants

from diverse geographic regions and environmental

conditions will improve the system's adaptability.

This will enhance the accuracy of plant classification,

ensuring consistent performance across different

climates, seasons, and growth stages.

A major advancement would be the development

of a real-time mobile application, revolutionizing

how users interact with the system. Incorporating

features such as image-based plant identification,

voice search, and offline access, the app would serve

as a valuable tool for farmers, herbalists, scientists,

and healthcare practitioners. Users could instantly

recognize medicinal plants, learn about their

therapeutic applications, and receive Ayurvedic

treatment suggestions, even in remote areas. Adding

community-driven features, such as expert

consultations, discussion forums, and user-

contributed data, would further enhance knowledge

sharing and engagement.To strengthen security and

data integrity, Blockchain technology can be

employed for secure and verifiable storage of

Ayurvedic information. Blockchain’s decentralized

nature will prevent unauthorized modifications,

preserving the authenticity of medicinal plant data

and ensuring reliable access to traditional knowledge.

This will be particularly useful for maintaining

historical records of Ayurvedic treatments and

formulations while preventing misinformation.

Moreover, incorporating Artificial Intelligence (AI)

for disease prediction and Ayurvedic treatment

recommendations will enhance the system’s

capabilities. AI-driven models can analyze plant

health, soil conditions, and environmental variables

to detect potential diseases and recommend

appropriate Ayurvedic remedies for both plants and

human ailments.

REFERENCES

AyurLeaf: A Deep Learning Approach for Classification of

Medicinal Plants. 2019 IEEE Region 10 Conference

pages 1–1.

Deep Herb: A Vision Based System for Medicinal Plants

Using Xception Features. In 2021 Research Scholar,

VTU-RRC, Visvesvaraya Technological University,

Belagavi.

Goodfellow et al.'s Deep Learning for Text Classification

and Sentiment Analysis (2016)

Gopal, A., Prudhveeswar Reddy, S., and Gayatri, V.

explored the classification of specific medicinal plant

leaves using image processing techniques. Their study,

published by IEEE in 2012, focused on leveraging

computational methods for accurate plant

identification.

J. Samuel Manoharan conducted research on the accurate

identification of herbal plant leaves using a machine

learning classifier. The study introduced a two-stage

authentication process to enhance detection accuracy

and was published in the Journal of Artificial

Intelligence and Capsule Networks (2021).

Ontology-BasedText Mining for Clinical Knowledge

Extraction (Smith et al., 2015)

Pharmacology of Medicinal Plants and Natural Products by

S.A. Dahanukar, R.A.Kulkarni, and N.N. Rege (2000)

Shashank M. Kadiwal, Gowrishankar S., Srinivasa A. H.,

and Veena A. presented a study on identifying Indian

medicinal plant species using deep learning techniques.

Their research was published in the 2022 4th

International Conference on Inventive Research in

Computing Applications (ICIRCA), focusing on

advanced computational methods for plant recognition.

Tarunsuresh, Deep Learning for Anthrocnose Diagnosis in

TurniLeaves. Science Open. 2021

The Text Mining Handbook: Advanced Approaches to

Analysing Unstructured Data by Ronen Feldman and

James Sanger (2007)

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

626