AI Based Tomato Leaf Disease Detection Using a Hybrid Modified AlexNet for Food Security Application

Abstract

Tomato leaf diseases are of great importance in reducing the yield and quality of tomato crops affecting the productivity and food security. Hence, early and accurate diagnosis of diseases is a crucial component of sustainable management of agriculture. In this study, a Hybrid Modified AlexNet model is proposed for automatic classification of tomato leaf diseases using real-field images. The proposed architecture employs a hybrid feature fusion mechanism for combining the flattened spatial features and Global Average Pooling (GAP) features to efficiently extract both the local disease characteristics and global contextual information. The dataset has sixteen healthy and diseased tomato leaf classes collected in real agricultural conditions. To improve model generalisation all images were resized to 100 × 100 pixels and augmented by rotation, flipping, zooming and shearing operations. The network consists of three convolutional blocks, followed by hybrid feature fusion and fully connected classification layers. Grad-CAM visualisation was added to enhance interpretability by highlighting the disease-affected regions responsible for the model predictions. Experimental results confirm that the proposed Hybrid Modified AlexNet obtained a classification accuracy of 99.70% with high values of precision, recall, F1-score, and ROC-AUC. The model has around 19.5 million trainable parameters and needs lower inference time and energy consumption compared to several existing deep learning architectures. The lightweight and explainable design makes the proposed model fit to be deployed on mobile and edge devices for real-time disease diagnosis.