Cassava Leaf Disease Detection using Seresnext-50 with Attention

Cassava, a crucial staple crop faces persistent threats from various diseases, significantly undermining food security and economic stability in the region. Traditional methods of disease detection rely heavily on manual inspection by agricultural experts, presenting challenges of scalability and accessibility for smallholder farmers. In response, this research endeavors to introduce a data-driven solution by utilizing deep learning techniques for automated classification of cassava leaves. In this research a novel methodology is proposed for automatic detection of the cassava leaves into five classes (four diseased and one healthy) using deep learning techniques. The proposed methodology incorporates transfer learning strategy to train SEResNext50 32x4d model with self-attention mechanism. The experiments are performed over cassava leaf disease dataset. The results confirm the efficiency of the approach introduced, showcasing significant accuracy of 0.9647 for “Cassava Mosaic Disease” and an overall accuracy of 0.8988 in image classification leading to efficiently utilize the method in detecting diseased leaves through computer-aided-diagnostic systems.