Efficient Classification of Sugarcane Leaf Disease Using Fine-Tuned MobileNetV2

In this work, we propose a deep learning approach to classify the health of sugarcane leaves using the transfer learning on MobileNetV2 architecture. This research aims to improve crop disease detection through the exploitation of a pretrained model on ImageNet tailored to the specific task of classification into diseased or healthy sugarcane leaves. The dataset was split into three sets namely training, validation, and test with data augmentation and caching strategies to improve the model efficiency and performance. We froze some of the layers of the model MobileNetV2. This will allow selective extraction of features while other layers are open and adaptable to the dataset of the sugarcane leaf. After 50 epochs of training, the model performed very well. Loss and accuracy were very low, indicating minimal overfitting on the test set. A final confusion matrix was then derived in order to measure the reliability of the predictions across classes. The approach will thus validate the application efficacy of transfer learning for crop disease detection and its promotion of accurate, scalable classification in crop disease management.