Sugarcane Leaf Defect Identification by CNN with Attenuation Modeling

Abstract

Sugarcane leaf defect identification is critical for early disease management and ensuring high crop yield. In this work, we propose a novel convolutional neural network (CNN) integrated with attenuation modeling to accurately classify nine different classes of sugarcane leaf defects. Attenuation modeling is incorporated to enhance feature extraction by emphasizing subtle lesion patterns and reducing background noise interference, which often hinders defect recognition. The proposed hybrid approach leverages CNN’s ability to capture spatial features alongside attenuation coefficients that modulate the input representation, improving model sensitivity to variations in leaf texture and color caused by different defects. Experimental evaluation on a comprehensive sugarcane leaf image dataset demonstrates a classification accuracy of 94%, significantly outperforming traditional CNN methods that lack attenuation enhancement. The model also benefits from optimized training procedures, requiring only 20 minutes of training time on standard GPU hardware, making it practical for real-time agricultural monitoring applications. This approach offers robustness against varying illumination and complex backgrounds commonly found in field conditions. Our results indicate that combining CNN with attenuation modeling provides a powerful and efficient solution for sugarcane leaf disease classification, aiding farmers and agronomists in timely diagnosis and effective disease management strategies. This study contributes to precision agriculture by advancing automated visual inspection systems tailored to crop-specific challenges.