Plant Diseases Detection Optimization with Deep Learning and Data Augmentation

Plant disease detection is crucial for food security and sustainable agriculture. This study focuses on Tomato plants with five classes: Bacterial Spot, Yellow Leaf Curl Virus, Septoria Leaf Spot, Tomato Healthy and Late Blight. Utilizing Convolutional Neural Network (CNN) architectures including simple CNN with dropouts and regularization, CNN with Data Augmentation (DA), and CNN with Autoencoder or Convolutional Autoencoder (CAE), trade-offs between Accuracy, F1 score, Precision and Recall are explored. While the CNN achieves 98.12% accuracy, DA boosts it to 98.75%, albeit with increased training time. The hybrid CNN & CAE model strikes a balance at 97.08% accuracy. Extending the evaluation to Potato plants, the CNN & DA model achieves 90% accuracy. These findings demonstrate model adaptability across different crops and disease classes.