USST: Utilizing SimAM and SGA Techniques to Cassava Leaf Diseases Classification in Real Cultivation Environments

Cassava leaf diseases devastate crop yield and quality, posing a significant threat to food security in Africa. Limited by accuracy and processing speed, traditional detection methods hinder effective disease management. We introduce USST, a novel cassava leaf disease detection approach to address this challenge. USST employs a synergistic combination of two attention mechanisms for enhanced feature learning: optimized self-guided attention (SGA) captures global disease patterns. In contrast, a simple and efficient attention mechanism (simAM) precisely identifies disease-related regions in cassava leaf images. Our extensive experiments demonstrate that USST's comprehensive integration of global and local perspectives is highly effective for cassava leaf disease identification, especially in real plant environments. This approach surpasses previous methods in both accuracy and efficiency. On the Cassava Leaf Diseases Dataset, USST achieved an industry-leading accuracy of 88.48%, significantly outperforming established methodologies. This automated disease identification method presents an effective solution for farmers, with the potential to revolutionize agricultural practices.