Real-Time Plant Disease Detection System
Computer vision image classification pipeline utilizing fine-tuned CNN architectures for automated agricultural pathology
AI/ML Engineer & GenAI Systems Architect • Savitribai Phule Pune University (SPPU)
The Problem Statement
Solution Architecture & Implementation
Built a deep Convolutional Neural Network utilizing TensorFlow and Keras trained on thousands of plant leaf photographs from the PlantVillage corpus. Implemented aggressive image augmentation (random rotations, horizontal flips, Gaussian blurs, and brightness variations) to immunize the model against erratic real-world field lighting conditions. Trained with Adam optimizer and categorical cross-entropy, reaching high precision across 38 categories including tomato blight, apple scab, and corn rust.
Key Engineering Innovations
Overfitting Immunity: Employed aggressive Dropout (0.5) and Batch Normalization layers to prevent memorization of laboratory background artifacts.
Color Space Preservation: Retained essential leaf pigmentation cues by feeding full 3-channel RGB imagery with normalized floating-point tensors.
Actionable Treatment Prescriptions: Mapped each disease class to an automated agronomic treatment guide, empowering farmers with immediate remedies.
Performance Benchmarks & Empirical Impact
| Metric | Before / Baseline | Engineered | Net Impact |
|---|---|---|---|
| Validation Accuracy | 76.4% (Shallow CNN) | 94.8% (Deep CNN) | +18.4% Classification Accuracy |
| False Negative Rate on Blight | 18.2% | 3.1% | High Agricultural Safety |
| Model Size (Quantized) | 148 MB | 18.4 MB | 8x Reduction for Mobile Deployment |
Technology Stack Architecture
Deep Learning Framework
Computer Vision
Data & Evaluation
Looking for an AI/ML Engineer or GenAI Consultant?
Yash Jondhale delivers production-ready RAG pipelines, FastAPI microservices, and computer vision systems. Available for machine learning roles and enterprise AI advisory.