Description of Invention
Plant diseases continue to threaten global food security, contributing to 20–40% losses in agricultural production annually. Traditional disease detection methods are often slow, labor-intensive, and prone to error, delaying timely action. This project presents a deep learning-based crop health monitoring system utilizing the ResNet101V2 architecture with transfer learning to detect diseases in both tomato leaves and fruits. The model was trained and evaluated on three datasets: approximately 16,000 tomato leaf images from PlantVillage, 7,226 tomato fruit images classified into four quality stages (Unripe, Ripe, Old, Damaged), and 2,400 fruit images graded into Reject, Ripe, and Unripe based on OECD and USDA standards. Preprocessing techniques, including contrast enhancement, edge sharpening, and normalization, were selectively applied to improve feature extraction, alongside data augmentation strategies such as rotation, zoom, and brightness adjustment. Model performance was assessed using Accuracy, Precision, Recall, and F1-Score across raw, preprocessed, and augmented datasets, demonstrating strong classification capability across diverse conditions.