Abstract
Precise and prompt detection of potato leaf diseases is crucial to minimize agricultural losses and maximize crop productivity. This research presents a novel perception-based framework to address this challenge efficiently. The proposed model leverages the strengths of MDSCIRNet and SEResNet101V2, incorporating depthwise separable convolutions, multi-head attention, and squeeze-and-excitation blocks to enhance feature extraction and classification. Advanced data augmentation techniques improved the training dataset and improved model generalization. Performance evaluations against state-of-the-art architectures demonstrated the superior accuracy and efficiency of the proposed model, achieving a training accuracy of 99.89% and a test accuracy of 99.67%. The results, validated through confusion matrices, classification reports, and performance graphs, confirm the effectiveness of the model in accurately detecting various potato leaf diseases. This novel approach provides a promising solution for early disease diagnosis, enabling timely interventions and reducing economic losses for farmers.
