Plant Disease Detection via Computer Vision

An efficient CNN architecture for detecting diseases in tomato leaves, designed for deployment on edge devices.

Problem Formulation

Early detection of plant diseases is crucial for food security. Farmers in remote areas often lack access to agricultural experts.

Data

Used the PlantVillage dataset containing 16,000 images of healthy and diseased tomato leaves across 10 classes.

Methodology

We utilized Transfer Learning with MobileNetV2 to ensure the model is lightweight enough for mobile deployment.

Results

Achieved 98.5% validation accuracy with an inference time of 45ms on a Raspberry Pi 4.

Limitations & Future Work

Performance drops significantly under poor lighting conditions.