Deep Learning-Based Classification of Sugarcane Leaf Disease

Plants are seriously threatened by crop diseases. Prompt preventive action can reduce the likelihood of crop loss and contamination. This study explores using deep learning to classify sugarcane leaf diseases, crucial for effective agricultural disease management. Fungal infections like red rot, rust, yellow, and mosaic viruses can harm sugarcane health. Deep learning models help extract complex information from images due to their multi-level structures. CNNs are also able to dramatically reduce computation time by taking advantage of GPU for computation. In this study, we utilize advanced deep learning models, including ResNet-50, VGG-16, DenseNet-201, VGG-19, and Inception V3, for the classification of sugarcane leaf diseases. We carefully curate a diverse dataset from mendeley of 2511 images comprising 5 classes: Healthy, Mosaic, Red Rot, Rust and Yellow to train and evaluate these models. After fine-tuning the parameters and implementing preprocessing techniques such as gamma correction and contrast stretching, ResNet-50 emerges as the top performer with 95.69% accuracy, followed by VGG-16 at 93.26% and DenseNet-201 at 89.62%. VGG-19 and Inception V3 achieve accuracies of 79.62% and 74.88%, respectively. These findings provide valuable guidance for selecting suitable models to improve disease management in agriculture, leading to enhanced productivity and sustainable sugarcane farming practices.