A Robust Intelligent System for Detecting Tomato Crop Diseases Using Deep Learning

The tomato crop is a strategic crop in the Egyptian market with high commercial value and large production. However, tomato diseases can cause huge losses and reduce yields. This work aims to use deep learning to construct a robust intelligent system for detecting tomato crop diseases to help farmers and agricultural workers by comparing the performance of four different recent state-of-the-art deep learning models to recognize 9 different diseases of tomatoes. In order to maximize the system's generalization ability, data augmentation, fine-tuning, label smoothing, and dataset enrichment techniques were investigated. The best-performing model achieved an average accuracy of 99.12% with a hold-out test set from the original dataset and an accuracy of 71.43% with new images downloaded from the Internet that had never been seen before. Training and testing were performed on a computer, and the final model was deployed on a smartphone for real-time on-site disease classification.