EfficientNetB4-SECA: A Deep Learning Model for Detecting Leaf Diseases in Tomato and Chili Plants

Abstract

Tomatoes and chilies are essential crops vulnerable to foliar diseases that significantly reduce yield and quality. Accurately identifying these diseases is essential for effective precision agriculture. Despite advances in deep learning, few studies have systematically evaluated state-of-the-art models on crop-specific datasets. To address this gap, we assessed the performance of leading architectures, including EfficientNet-B4, Vision Transformer, InceptionV3, and EfficientNet-B4 variants with attention, on Tomato Leaf Disease and COLD Chili datasets. Motivated by the evaluation results, we propose EfficientNet-B4-SECA, a novel deep learning 
model tailored for tomato and chili leaf disease classifications. It builds upon the EfficientNet-B4 backbone and incorporates two attention mechanisms: Squeeze-and-Excitation, which enhances key feature channels, and Coordinate Attention, which captures spatial and channel dependencies. Experimental results indicate that EfficientNet-B4-SECA achieves the highest classification accuracy on both datasets, with 84.97% for tomato and 87.83% for chili leaf diseases, surpassing that of all compared models. These findings demonstrate the effectiveness of the model and its potential for real-world deployment in agricultural disease monitoring.