RAI-Net: Tomato Plant Disease Classification Using Residual-Attention-Inception Network

Pests and pathogens cause significant losses to tomato crop yield, resulting in billion-dollar economic impact worldwide. Early and accurate detection of tomato plant disease is important for an effective intervention and improved crop yield. In this work, a deep learning-based RAI (Residual-Attention-Inception)-Net model has been proposed for an effective feature extraction by deploying channel attention on the output obtained from fine-tuned ResNet18 model for improved feature extraction. Inception module has been integrated with the ResNet18 model augmented with channel attention module enhances the multi-scale feature analysis capability of the proposed model for tomato plant leaf disease detection task. The interpretability of the proposed network has been improved with the Gradient-weighted Class Activation Mapping (Grad-CAM) method which highlights the regions focused by the proposed RAI-Net model in making its predictions. RAI-Net achieves an accuracy of 97.88% on a test set comprising 4595 images across ten different classes of tomato disease, demonstrating its effectiveness in automated detection of tomato leaf diseases.