Final Project: AI-Powered
Plant Disease Prediction
Developed for IBase Technologies, Nagpur
Submitted To: Mr. Nakul Deshmukh
Submitted By: Group codeIQ - Ankit Sarode, Shivam Shende, Pranay
Meshram
Common Okra Plant Diseases
Our project focuses on identifying and predicting solutions for prevalent okra plant diseases and pest infestations that significantly
impact crop yields.
Verticillium Wilt Bacterial Leaf Spot
Damping Off Shoot and Fruit Borer
Enation Leaf Curl Virus Leaf Roller
Rhizoctonia Root Rot Blister Beetle Damage
Deep Learning Models for Accuracy
We evaluated several cutting-edge deep learning architectures to achieve high accuracy in disease detection, specifically for Okra,
Cotton, and Hibiscus plants.
VGG1C RESNET50 CNN
A deep convolutional neural network Utilizes residual connections to Convolutional Neural Network, the
known for its simplicity and enable training of much deeper foundational architecture for image
effectiveness in image recognition networks, enhancing feature analysis and classification.
tasks. extraction.
CLD+CNN EHD+CNN
Combined approach leveraging Color Local Descriptors Utilizes Edge Histogram Descriptors combined with CNN
with CNN for enhanced disease feature identification. for robust detection of disease patterns.
Descriptive Paper
Our comprehensive descriptive paper outlines the theoretical framework,
methodology, and detailed findings of our plant disease prediction system.
Introduction: Revolutionizing Crop Health
The agriculture sector faces significant challenges from plant diseases, leading to substantial crop losses. Early and accurate
disease detection is crucial for sustainable farming and food security.
This project introduces an innovative Machine Learning and Deep Learning solution to predict and identify solutions for various
plant diseases, focusing on vital crops like Okra, Cotton, and Hibiscus.
Literature Review: Building on Foundational
Research
Our research involved a thorough review of existing studies in plant disease detection using computer vision and AI.
Image Processing Techniques: Analysis of traditional methods for feature extraction and disease symptom identification.
Machine Learning Applications: Examination of SVM, Random Forest, and other ML algorithms in agricultural diagnostics.
Deep Learning Advancements: Review of CNN, RNN, and transfer learning applications for high-accuracy plant disease
classification.
Dataset Challenges: Understanding the complexities and limitations of publicly available plant disease datasets.
Problem Statement: The Need for Timely
Intervention
Farmers often struggle with the timely and accurate identification of plant diseases due to lack of expertise and resources. This
leads to:
Significant Crop Losses Inefficient Resource Use Lack of Accessible
Diseases spread rapidly, decimating Broad-spectrum pesticides and Expertise
yields before effective measures treatments are often applied, Agricultural experts are not always
can be taken. increasing costs and environmental available to provide immediate
impact. diagnoses.
Proposed Work: An AI-Driven Solution
We propose an intelligent system that leverages Deep Learning to accurately identify plant diseases from images and suggest
appropriate solutions.
Deep Learning
Solution
Recommendation
Image Input
Disease Prediction
This system will empower farmers with rapid, reliable diagnoses, enabling targeted interventions and fostering healthier crops.
Methodology: Our Approach to AI Training
Our methodology involves a systematic process from data collection to model deployment.
Model Selection & Training
Data Acquisition & Preprocessing Implementing VGG16, ResNet50, and custom CNN
Collecting diverse images of healthy and diseased plant architectures, fine-tuning for optimal performance.
leaves, followed by augmentation and normalization.
Solution Recommendation Engine
Performance Evaluation Developing a rule-based or knowledge-based system to
Rigorous testing using accuracy, precision, recall, and F1- provide practical remedies for identified diseases.
score to validate model effectiveness.
Result Analysis: Impact and Future Scope
120
80
40
0
VGG16 ResNet50 C N N (Custom) CLD+CNN EHD+CNN
Our models demonstrate high accuracy, with ResNet50 achieving the best performance. Future work includes expanding the
dataset, real-time detection, and integration with drone technology.
Thank you!
