ACHC: a novel deep learning model for tomato leaf disease

The expansion of commercial tomato cultivation creates a pressing need for accurate, automated disease diagnosis to support modern agriculture. To address this, we developed an enhanced AlexNet model integrated with Coordinate and Hybrid Channel-spatial attention (ACHA) for tomato leaf disease recognition. The classical AlexNet was first streamlined to reduce computational overhead, tailored to the characteristics of tomato disease imagery. A coordinate attention module was then inserted after Conv1 and Conv3 to improve spatial localization of lesion areas. Additionally, a hybrid channel-spatial attention module was incorporated after Conv2 and Conv4 to strengthen discriminative feature learning across channels. Experimental results demonstrate that our approach achieves an average test accuracy of 99.62%, outperforming several state-of-the-art deep learning networks such as MobileNetV3 and ResNet50. This work provides an effective and lightweight solution for automated disease monitoring, contributing to more precise and sustainable tomato production systems.