Disease Identification & Classification in Millet Crops Using ML

Techniques

Meena Kumari

, Devansh Kulshreshtha

, Anshul Agrawal

and Aniket Sharma

Department of Computer Science & Engineering (IoT), ABES Institute of Technology, Ghaziabad, Uttar Pradesh, India

Department of Computer Science & Engineering (AI), ABES Institute of Technology, Ghaziabad, Uttar Pradesh, India

Keywords: Machine Learning, Deep Learning, CNN, Transfer Learning, Image Classification, Millet Disease Detection,

Precision Agriculture, Smart Farming, Computer Vision, Crop Health Monitoring, Weed Detection.

Abstract: Millets, including pearl, finger, and sorghum varieties, are essential crops known for their adaptability to harsh

conditions and significant contribution to food security. However, their cultivation is frequently disrupted by

plant diseases and weed infestations, which affect yield and quality. Traditional methods of addressing these

issues often require manual effort and are limited in scalability, making them inefficient for large-scale

farming. This project aims to provide a comprehensive solution through Machine Learning (ML) and

Computer Vision technologies. Using the pre-trained VGG16 model, the system identifies and classifies

diseases in millet crops, determining whether the plants are healthy or affected by specific conditions such as

rust, mildew, or blast. Additionally, a weed detection feature is incorporated to facilitate effective

management of weeds. The solution is deployed as a user-friendly application designed to deliver real-time

insights, improving agricultural practices and promoting sustainable millet farming.

1 INTRODUCTION

1.1 Significance of Millets

Millets are resilient and nutritious crops, widely

cultivated in regions where other staples struggle to

grow. Pearl, finger, and sorghum millets, in

particular, provide essential nutrients and are critical

to addressing food security challenges. They are also

environmentally sustainable, requiring minimal water

and chemical inputs.

1.2 Challenges in Millet Cultivation

Despite their resilience, millets are not immune to

agricultural challenges. Two primary issues

impacting millet productivity are:

• Diseases: Common diseases such as rust,

smut, and blast can severely affect millet

growth, resulting in reduced yields.

• Weed Infestation: Weeds compete with

crops for essential resources like sunlight,

water, and nutrients, further reducing

productivity.

1.3 Drawbacks of Current Practices

Farmers traditionally rely on manual inspection for

disease and weed management. This technique has

drawbacks such as: -

• Being time-intensive and unsuitable for

large-scale farming.

• Prone to errors and inconsistencies.

• Inability to provide timely interventions for

early-stage disease or weed management.

1.4 Role of Technology

The integration of ML and Computer Vision in

agriculture presents an opportunity to address these

challenges efficiently. Pre-trained models such as

VGG16 enable quick and precise identification of

diseases and weeds, allowing for automated solutions

tailored to millet cultivation. Table shows the

proposed Diseases.

328

Kumari, M., Kulshreshtha, D., Agrawal, A. and Sharma, A.

Disease Identiﬁcation & Classiﬁcation in Millet Crops Using ML Techniques.

DOI: 10.5220/0013897500004919

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 3, pages

328-334

ISBN: 978-989-758-777-1

Table 1: Proposed Diseases.

S No. Title Year Author Diseases

Proposed

Diseases

Deep-millet: a

deep learning

model for pearl

millet disease

identification to

envisage precision

agriculture

2024

Jhonson et

Rust

Disease

Blast

Disease

Powdery

Mildew

 Downy

Mildew

 Blast

Disease

 Rust

Disease

A Smart and

sustainable

framework for

millet crop

monitoring

equipped with

disease detection

using enhanced

predictive

intelligence

2023

Mishra et

Blast Disease

Rust Disease

IoT and

Interpretable

Machine Learning

Based Framework

for Disease

Prediction in Pearl

Millet

2021

Nidhi

Kundu et al

Blast Disease

Rust Disease

2 LITERATURE SURVEY

In (I Johnson, 2024), The paper presents the Deep

Millet model, a CNN-based solution for detecting

pearl millet diseases with 98.86% accuracy, enabling

real-time disease identification through a mobile app

to support precision agriculture. (Johnson et al,

2022). In (

Mohamed Salama, 2024), Agriculture faces

threats from pests and diseases. This paper reviews

AI's role in automating detection, noting its efficiency

and accuracy. It discusses implementation challenges

and solutions, and calls for ongoing research to

enhance AI in pest and disease management. (Salama

et al, 2024). In (

Nivargi Anil Basavant, 2024), The paper

presents a machine learning-based agricultural

system for disease classification, crop prediction, and

fertilizer recommendation. It helps optimize farming

practices, improve productivity, and promote

sustainability. Future enhancements aim to refine

algorithms and integrate real-time data for

broader adoption. Basavant et al, 2024).

In (

Sanika kadam, 2024), The research uses machine

learning and optimized CNN models, such as AlexNet

and GoogleNet, to identify soybean diseases with high

accuracy. It underscores the potential of AI to enhance

agricultural productivity and sustainability. Kadam et

al, (2024). In (

Rushikesh Pawar,2024), The research

focuses on using machine learning, particularly CNN

models such as LeNet and VGG16, for identifying

diseases in soybean leaves. It introduces the SoyNet

dataset to enhance accuracy, supporting better crop

health and productivity. Pawar et al, (2024). In (

Payam

Delfani,2024)

, This paper explores how smart

technologies like IoT, machine learning, and AI are

transforming modern farming. It focuses on how these

tools help farmers predict plant diseases early, make

better decisions about resource use, and boost crop

yields—especially as climate change adds new

challenges to agriculture. Delfani et al, (2024).

In (

K. Sai Susheel et al,2023), This review explores

intelligent techniques like machine learning, image

processing, & IoT in identifying, monitoring, &

managing crop pests & diseases, aiming to enhance

agricultural productivity and sustainability. This

review includes different crops & diseases. Susheel et

al, (2023). In (

Sushruta Mishra et al,2023), The paper

proposes a smart millet crop monitoring system using

IoT and a Customized CNN, achieving 98.8%

Disease Identiﬁcation & Classiﬁcation in Millet Crops Using ML Techniques

329

accuracy in disease detection to support farmers

and enhance yield. Mishra et al, (2023). In (

Riya Walia

et al., 2023)

, The paper addresses sugarcane’s

susceptibility to Top Borer disease and introduces an

AI system using CNNs and high-resolution images

for precise detection. It advocates sustainable farming

and highlights the need for ongoing updates to tackle

emerging challenges. Walia et al, (2023). In (

Md.

Mehedi Hasan et al., 2023)

, The document discusses

machine learning and image processing techniques

for detecting rice diseases, focusing on their role in

improving accuracy and boosting

agricultural productivity. Hasan et al, (2023).

In (

Wanjie Feng et al., 2024)

, This paper shows how

AI models like SoyDNGP are changing crop breeding

by helping predict plant traits, choose the best parent

plants, and blend genetic and environmental data to

grow better crops faster. Feng et al, (2024). In (

Bita

Parga Zen et al., 2022)

, This study will discuss the

implementation of Artificial Intelligence-based plant

disease detection software. At this stage, deep

learning models are created using cameras matched.

Zen et al, (2022). In (

Md. Ashraful Haque, et al,2022)

The paper presents a deep learning-based approach

for identifying three major maize diseases using in-

field images. It applies an Inception-v3 CNN model,

achieving a classification accuracy of 95.99%. The

approach demonstrates improved disease detection

performance, even with varied backgrounds and

enhanced brightness conditions. Haque et al, (2022).

In (

Tiago Domingues, et al., 2022)

, The paper

reviews machine learning techniques for detecting,

classifying, and predicting crop diseases and pests. It

emphasizes the potential of ML in sustainable

farming, using weather, image, and spectral data to

improve crop yield and reduce pesticide use.

(Dominques et al, 2022). In (

Sana Akbar, et al., 2022)

The study integrates IoT and machine learning for

wheat disease detection, using MobileNet and

EfficientNet-B3. Techniques include image

preprocessing, augmentation, and CNN-based

classification, emphasizing IoT's role in monitoring

and improving crop yield management. Akbar et al,

(2022).

In (

Nidhi Kundu, et al., 2021)

, The shift to high-yield

grains has worsened malnutrition, prompting India to

promote millets as "Nutri Cereals" for food security.

This paper proposes using machine learning and IoT

for automated disease detection in pearl millet.

(Kundu et al,2021). In (

Muhammad Hammad Saleem, et

al., 2019)

, The paper reviews the application of deep

learning (DL) techniques for detecting and

classifying plant development of a fuzzy expert

system for integrated disease management in finger

millet crops, leveraging fuzzy logic to diagnose

diseases and recommend control measures for

effective crop management. Roseline et al, (2012).

3 METHODOLOGY

The project follows a structured approach divided

into three key modules:

• Disease Detection: Identifying whether

millet crops (Pearl, Finger, Sorghum) are

healthy or diseased.

• Disease Classification: Categorizing

diseases like Rust, Smut, or Blast.

• Weed Detection: Differentiating crops from

weeds for better yield management.

Figure 1: Disease Detection.

As shown if figure 1 The Disease Detection

Module follows a systematic approach, starting with

data collection from public datasets, field surveys,

and research institutions. Images are labeled as

healthy or diseased (Rust, Smut, Blast). In data

preparation, preprocessing (resizing, normalization,

noise reduction) and augmentation (rotation, flipping,

brightness adjustment) enhance model performance.

The disease detection phase employs VGG16 fine-

tuning and custom CNN models for better accuracy.

Model training uses Categorical Cross-Entropy loss,

the Adam optimizer, and dynamic learning rate

scheduling, with data split into training, validation,

and test sets. Evaluation relies on accuracy, precision,

recall, F1-score, and k-fold cross-validation. The

workflow includes image collection, preprocessing,

disease detection, prediction, and result output with

diagnosis and treatment recommendations.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

330

As shown if figure 2 below the Disease

Classification Module categorizes millet diseases

(Leaf Spot, Rust, Downy Mildew, Blight) using crop

images. It involves dataset collection, pre-processing

(resizing, normalization, augmentation), and model

selection (pre-trained CNNs like ResNet, VGG16, or

custom models). Feature extraction captures color,

texture, and edge patterns. Model training optimizes

performance using categorical cross-entropy and the

Adam optimizer, while evaluation relies on accuracy,

precision, recall, and F1-score. Post-processing

assigns labels based on probability thresholds and

visualizes results. The workflow follows dataset

collection, preprocessing, model selection, training,

evaluation, and result visualization for accurate

disease identification.

Figure 2: Disease Classification.

As shown if figure 3 The Weed Detection Module

automates weed identification in Pearl millet, Finger

millet, and Sorghum millet fields, improving crop

management. It collects images from drones,

smartphones, and cameras, supplemented with open-

source datasets. Preprocessing includes resizing

(224x224 pixels), noise reduction, brightness

adjustments, and data augmentation. The system uses

U-Net/Mask R-CNN for segmentation and CNN

models like ResNet, MobileNet for classification,

with transfer learning for better accuracy. The dataset

is split (70% training, 20% validation, 10% testing),

optimized using Adam, and evaluated with IoU, Dice

Coefficient, Precision, Recall, and F1-Score.

Figure 3: Weed Detection.

3.1 Components

3.1.1 Image Acquisition & Preprocessing

Components Used:

• Digital Camera / Smartphone / Web

Scraping (for collecting images).

• Image Dataset (124 images of diseased &

healthy millet).

• OpenCV / PIL (for image processing).

Functions:

• Capture and store images of healthy and

diseased millet crops.

• Resize images to 150×150 px for input into

the CNN.

• Convert images to RGB format if required.

3.1.2 Data Augmentation

Components Used:

• Keras Image Data Generator

• NumPy, OpenCV

Techniques Applied:

• Rotation (Randomly rotating images).

Disease Identiﬁcation & Classiﬁcation in Millet Crops Using ML Techniques

331

• Flipping (Horizontal & Vertical).

• Zooming (Rescaling images).

• Brightness Variation (Simulating real-world

lighting).

Purpose:

• Increase dataset size (from 124 to 711

images).

• Reduce overfitting.

3.1.3 Transfer Learning with VGG16

Components Used:

• VGG16 Pre-Trained Model (Keras /

TensorFlow)

• Feature Extraction Layers (Convolutional &

Pooling layers).

Function:

• Freeze the initial layers of VGG16 to retain

learned features.

• Extract features from input images to detect

disease patterns.

3.1.4 Fully Connected Layers (Custom

Classification Head)

Components Used:

• Dense Layers (Fully Connected Layers)

• Dropout Layer (To prevent overfitting)

• Activation Functions: ReLU & Sigmoid

Structure:

• Two fully connected layers added on top of

VGG16.

• Final Layer: Sigmoid Activation → Binary

Classification (Healthy / Diseased).

3.1.5 Model Training & Optimization

Components Used:

• Optimizer: Stochastic Gradient Descent

(SGD) with Momentum (0.9)

• Loss Function: Binary Crossentropy

• Early Stopping (To prevent overfitting).

• 80-20 Split (Training & Validation).

Training Configuration:

• Learning Rate: 1e-4

• Epochs: 30 (Early stopping applied).

3.1.6 Disease Classification & Prediction

Components Used:

• Trained CNN Model

• Softmax / Sigmoid Activation for

Classification

• Performance Metrics: Accuracy, Precision,

Recall, F1-Score

Function:

• Input: New millet leaf image.

• Processing: Feature extraction &

classification.

• Output: Healthy or Mildew-Affected (with

95% accuracy).

4 FUTURE SCOPE

• The future of this model holds great

potential for smart agriculture. It can be

expanded to detect multiple diseases in

millet crops like finger millet, sorghum,

and foxtail millet, as well as wheat, rice,

and maize.

• Integrating it into mobile apps and IoT

devices will allow farmers to snap pictures

for instant disease detection, making it

accessible and affordable.

• Using advanced deep learning models like

Efficient Net and Vision Transformers will

improve accuracy, while segmentation

techniques (U-Net, Mask R-CNN) can

help classify disease severity.

• Drones and remote sensing can be used for

large-scale farm monitoring, optimizing

pesticide use and resource management.

• Adding multi-language support will make

it easier for farmers worldwide, and fine-

tuning for regional diseases will improve

effectiveness.

• This AI-driven approach will help farmers

detect diseases early, boost crop yields,

and ensure sustainability, ultimately

contributing to global food security.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

332

5 CONCLUSIONS

This project offers an integrated solution for millet

farming by combining disease detection and weed

identification into a single platform. Leveraging

VGG16 and state-of-the-art ML techniques ensures

high accuracy and efficiency, addressing the

limitations of traditional methods. The deployable

system promises to enhance millet productivity,

reduce losses, and promote sustainable farming

practices. Future work could include extending the

model to additional crops and incorporating IoT-

based monitoring for continuous field data collection.

ACKNOWLEDGEMENTS

We would like to profoundly express sincere

gratitude to the ABES Institute of Technology,

Ghaziabad for their significant support, which

essentially made this really important research

possible. We also extend our deepest appreciation to

Dr. Upasana Pandey and Ms. Meena Kumari for their

extremely valuable guidance and insights throughout

this awesome project Finally, but certainly, not least,

we give thanks to all the wonderful participants who

kindly volunteered their precious time and expertise

for data collection and testing, without whom this

groundbreaking research would maybe not have been

even a little bit feasible.

REFERENCES

Bita Parga Zen, Iqsyahiro Kresna A, Diandra Chika

Fransisca, Applications for Detecting Plant Diseases

Based on Artificial Intelligence, Volume 6, Number 4,

October 2022.

https://doi.org/10.33395/sinkron.v7i4.11833

Habib Khan, Ijaz Ul Haq, Muhammad Munsif, Mustaqeem,

Shafi Ullah Khan and Mi Young Lee, Automated

Wheat Diseases Classification Framework Using

Advanced Machine Learning Technique, Agriculture

2022, 12(8).

https://doi.org/10.3390/agriculture12081226

I Johnson, X Anitha Mary, A Peniel Winifred Raj, J

Chalmers, M Karthikeyan and Andrew J, Deep-millet:

a deep learning model for pearl millet disease

identification to envisage precision agriculture– a

review, Environ. Res.Commun.6(2024)105031.

https://doi.org/10.1088/2515-7620/ad8415

K. Sai Susheel, R. Rajkumar, A Comprehensive Review on

Intelligent Techniques in Crop Pests and Diseases,

IJRITCC | September 2023.

https://doi.org/10.17762/ijritcc.v11i9.8328

Md. Ashraful Haque, Sudeep Marwaha, Chandan Kumar

Deb, Sapna Nigam, Alka Arora, Karambir Singh

Hooda, P. Lakshmi Soujanya, Sumit Kumar Aggarwal,

Brejesh Lall, Mukesh Kumar, Shahnawazul Islam,

Mohit Panwar, Prabhat Kumar & R. C.Agrawal, Deep

learning‑based approach for identifcation of diseases of

maize crop, (2022) 12:6334.

https://doi.org/10.1038/s41598-022-10140-z

Md. Mehedi Hasan, A F M Shahab Uddin, Mostafijur

Rahman Akhond, Md. Jashim Uddin, Md. Alamgir

Hossain 2 and Md. Alam Hossain 1, Machine Learning

and Image Processing Techniques for Rice Disease

Detection: A Critical Analysis, 2023, Int. J. Plant Biol.

2023, 14, 1190– 1207.

https://doi.org/10.3390/ijpb14040087

Mohamed Salama, Ali Retha Hasoon, Mohamed K Zidan,

Sarah M. Ayyad, Role of AI for Plant Disease Detection

and Pest Detection, 22- 25 July 2024.

https://doi.org/10.1109/ITC-

Egypt61547.2024.10620496

Mr. Rushikesh Pawar, Mr. Rushikesh Awatade , Mr. Omkar

Shedge , Mr. Saif Shaikh, Crop Leaf Disease Prediction

Using Machine Learning, 2024, IJSREM.

https://doi.org/10.55041/IJSREM30693

Muhammad Hammad Saleem, Johan Potgieter and Khalid

Mahmood Arif, Plant Disease Detection and

Classification by Deep Learning, Plants 2019, 8, 468.

http://dx.doi.org/10.3390/plants8110468

Nidhi Kundu, Geeta Rani 1, Vijaypal Singh Dhaka, Kalpit

Gupta, Siddaiah Chandra Nayak, Sahil Verma,

Muhammad Fazal Ijaz and Marcin Wozniak, IoT and

Interpretable Machine Learning Based Framework for

Disease Prediction in Pearl Millet, Sensors 2021, 21,

5386. https://doi.org/10.3390/s21165386

Nivargi Anil Basavant, Gururaj V, Kavya N L, Crop

Prediction Analysis and Plant Disease Detection Using

Machine Learning, Volume 6, Issue 3, May-June 2024.

https://doi.org/10.36948/ijfmr.2024.v06i03.20123

Payam Delfani, Vishnukiran Thuraga, Bikram Banerjee,

Aakash Chawade, Integrative approaches in modern

agriculture: IoT, ML and AI for disease forecasting

amidst climate change, The Author(s) 2024.

https://doi.org/10.1007/s11119-024-10164-7

Philomine Roseline, Clarence J M Tauro, N Ganesan, PhD,

Design and Development of Fuzzy Expert System for

Integrated Disease Management in Finger Millets,

Volume 56– No.1, October 2012.

https://doi.org/10.5120/8857-2815

Riya Walia, Dr. Satender kumar, Advancing Sugarcane

Disease Detection through CNN-Based Deep Learning,

International Journal of Membrane Science and

Technology, 2023, Vol.10, No. 1, pp1851-1861.

https://doi.org/10.15379/ijmst.v10i1.3501

Sana Akbar, Dr. Khawaja Tehseen Ahmad, Muhammad

Kamran Abid, Dr. Naeem Aslam, Wheat Disease

Detection for Yield Management Using IoT and Deep

Learning Techniques, Volume 10, Number 3, July-

September 2022.

https://doi.org/10.21015/vtse.v10i3.1108

Disease Identiﬁcation & Classiﬁcation in Millet Crops Using ML Techniques

333

Sanika kadam, Priya Kadam, Rutuja Karvekar, Rutuja

Mohite, AI Based Soybean Crop Disease Detection,

2024, IJSREM.

https://doi.org/10.55041/IJSREM34409

Solemane Coulibaly, Bernard Kamsu-Foguem, Dantouma

Kamissoko, Daouda Traore, Deep neural networks with

transfer learning in millet crop images, S.Coulibaly et

al./ Computers in Industry 108 (2019) 115–120.

https://doi.org/10.1016/j.compind.2019.02.003

Sushruta Mishra, Dayal Rohan Volety, Navdeep Bohra,

Sultan Alfarhood, Mejdl Safran, A smart and

sustainable framework or millet crop monitoring

equipped with disease detection using enhanced

predictive intelligence– a review, Alexandria Enginee-

ring Journal 83 (2023) 298– 306.

https://doi.org/10.1016/j.aej.2023.10.041

Tiago Domingues, Tomás Brandão and João C. Ferreir,

Machine Learning for Detection and Prediction of Crop

Diseases and Pests: A Comprehensive Survey, Agricu-

lture 2022, 12, 1350. https://doi.org/10.3390/agricultu-

re12091350

Wanjie Feng, Pengfei Gao, Xutong Wang, AI breeder:

Genomic predictions for crop breeding, New Crops 1

(2024) 100010.

https://doi.org/10.1016/j.ncrops.2023.12.005

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

334