Enhanced Sugarcane Leaf Disease Prediction Using Deep Learning Models

The developed system marks a notable improvement in the prediction of sugarcane leaf diseases prediction through the integration of an expanded dataset and cutting-edge deep learning architectures. Traditional approaches, often restricted to five disease categories and smaller datasets, achieved a maximum accuracy of 86.53% with models like ResNet50 and XceptionNet but faced challenges in scalability and generalization. Addressing these limitations, the updated dataset introduces a new disease class, Sugarcane BacterialBlight, and incorporates a total of 17,982 images. This expansion is achieved through systematic data augmentation techniques, including image cropping, scaling, and rotation, ensuring a balanced and comprehensive dataset. Advanced deep learning models, including ResNet50, EfficientNet_B0, and MobileNet_V2, were meticulously trained and estimated on the augmented dataset using key metrics, including classification accuracy, precision, recall, and the F1score. EfficientNet_B0 model achieved a classification accuracy of 95.11%, with precision and recall both at 95.10% and 95.11%, respectively, and an F1-score of 95.09%. ResNet50 and MobileNet_V2 also demonstrated marked improvements compared to baseline performance. This research underscores the value of combining enriched datasets with efficient model architectures to deliver highly accurate disease prediction systems. These advancements hold substantial promise for precision agriculture, offering scalable, real-time solutions for sustainable farming practices.