EfficientNet-Based Approach for Enhanced Sugarcane Disease Classification: Implementation and Evaluation

Classifying plant diseases is essential for maintaining crop health and agricultural productivity. Technological advancements have shifted plant disease classification from traditional manual inspections to automated, precise techniques. This evolution is crucial for sugarcane, a vital crop impacting numerous economies. Deep learning techniques have yet to be applied to classify sugarcane plant diseases. Early methods relied on handcrafted features and basic algorithms, but sophisticated machine learning and deep learning techniques have revolutionized this field. Deep learning models such as Inception, ResNet, DenseNet, and MobileNet, trained on large datasets, have been instrumental in plant disease classification. With its compound scaling method, EfficientNet achieves superior accuracy while maintaining computational efficiency, outperforming other models in generalization and classification accuracy. Standard datasets like Plant Village and Leaf Snap have become benchmarks, but no comprehensive dataset has existed for sugarcane diseases until now. This research introduces the sugarcane disease dataset, implementing the EfficientNet model for classification that has reached 92.50%, 96,81%, and 91,26% in test and validation accuracy also mobileNetV3 Large reach 98,86 for validation accuracy on augmented sugarcane dataset. We compare these models with other established methods, demonstrating EfficientNet's superior performance through detailed visualizations of confusion matrix and accuracy graphs, which significantly contribute to sugarcane disease management.