Tomato Leaf Pathology Classification Model

Milestone Retrospective

Accurate detection of plant diseases is critical for sustainable agricultural production. In this project, I developed an end-to-end deep learning classification system capable of identifying 10 distinct tomato leaf diseases alongside healthy controls.

Using transfer learning with a VGG16 backbone pretrained on ImageNet, I designed custom dense classification heads with dropout regularization. To validate model robustness, I benchmarked performance against DenseNet121, InceptionV3, Xception, and MobileNetV2 using precision, recall, and F1-score metrics. The finalized pipeline achieved a 96.6% validation accuracy score.

Key Deliverables & Engineering Outcomes

• Preprocessed and augmented a 29,000+ image dataset to simulate variable field lighting and leaf orientations.
• Benchmarked 5 convolutional backbones, identifying VGG16 as the optimal precision-recall balance for plant spot diagnosis.
• Implemented automated confusion matrix generation and per-class diagnostic evaluation scripts.

Technologies & Methodologies