A Lightweight Convolutional Neural Network for Real-time Tomato Leaf Disease Detection

Tomato leaf diseases significantly impact crop yield and threaten global food security, particularly in resource-constrained regions with limited access to diagnostic tools. This research introduces an optimised and efficient disease identification framework comparing a streamlined custom convolutional neural network (CNN) with the larger VGG16 model. Both models were trained and validated on 6,124 high-resolution images of tomato leaves representing five categories (bacterial spot, early blight, yellow leaf curl virus, mosaic virus, and healthy specimens). Both models underwent extensive preprocessing and data augmentation techniques to enhance the model's robustness. Experimental results demonstrated that the custom CNN achieved a classification accuracy of 96.9% and a macro F1-score of 0.97, significantly outperforming the performance of 92.8% accuracy and a 0.93 F1-score. The custom CNN model demonstrated a remarkable strength in identifying early blight and yellow leaf curl virus. With only 136,000 trainable parameters, fewer than VGG16's 14.9 million, the model enables real-time deployment on mobile and Internet of Things (IoT) devices with limited computational resources. This lightweight system has a wide range of applications, including instant disease diagnosis for farmers and precision agriculture through drone- or sensor-based monitoring embedded in smart greenhouses for real-time monitoring. These findings demonstrate that a proposed lightweight architecture offers superior performance and scalability for agricultural artificial intelligence applications, providing an accessible, low-cost, edge-compatible solution for tomato disease identification and monitoring for smallholder farmers.