Intelligent Leaf Disease Diagnosis Using EfficientNetB0 Combined with CBAM in a Multi- Task Framework

Abstract

Automatic plant leaf disease detection plays a crucial role in smart agriculture by reducing crop losses and enabling timely treatment. In this study, a multi-task deep learning model based on EfficientNetB0 is proposed to simultaneously predict three outputs: leaf health status, plant species, and disease type. To enhance the feature representation of discriminative regions, the Convolutional Block Attention Module (CBAM) is integrated into the network. Furthermore, a specialized Gated Disease Head is designed to exploit the semantic dependency between plant species and disease type, while a Consistency Supervision mechanism is introduced to prevent contradictory predictions among the outputs. The model is trained on a dataset containing approximately 4,000 leaf images from barberry, jujube, and pomegranate plants, using a two-stage optimization strategy including warm-up and fine-tuning phases. Experimental results demonstrate that the proposed approach achieves 97% accuracy in health classification, 93% in species identification, and 93% in disease recognition, while maintaining a significantly lower parameter size compared to state-of-the-art methods. Overall, the proposed framework provides an efficient, lightweight, and reliable solution for real- world intelligent plant disease monitoring systems.