MultiCotNet: a novel multispatial attention-based deep learning architecture for cotton leaf disease classification

Abstract

Cotton is one of the most economically significant cash crops worldwide, and its productivity is severely affected by various leaf diseases that reduce crop quality and yield. Early and accurate disease identification is therefore essential for effective crop management and sustainable agricultural practices. Conventional disease diagnosis methods largely depend on manual inspection by experts, making them time-consuming, subjective, and less suitable for large-scale real-time monitoring under field conditions. To address these limitations, this study proposes MultiCotNet, a novel multispatial attention-based deep learning framework for robust classification of cotton leaf disease. The proposed architecture integrates multilevel attention mechanisms to enhance feature representation by capturing both local and global contextual information, thereby improving classification performance in challenging real-world scenarios involving illumination variations, occlusions, and complex backgrounds. Experimental evaluation conducted on a comprehensive real-world cotton leaf dataset demonstrates the effectiveness of the proposed model, achieving an accuracy of 96.6%, precision of 96.84%, recall of 97.21%, and F1-score of 96.96%. Comparative analysis further shows that MultiCotNet outperforms several existing baseline models in terms of classification accuracy and robustness. The proposed framework provides a scalable and reliable solution for early cotton disease detection and can support intelligent agricultural monitoring systems for timely disease management and improved crop productivity.