Enhanced Classification of Potato Leaf Disease Using Xception and ReduceLROnPlateau Callbacks
Andi Sunyoto, Yoga Pristyanto, Anggit Ferdita Nugraha
Abstract
The potato requires advancements in early crop blight classification and detection. Artificial intelligence technologies offer promising solutions for enhancing plant protection. This research paper introduces an optimal deep-learning model for classifying potato leaf diseases. The training dataset consists of three categories: healthy leaves, early blight, and late blight, with 1000, 1000, and 152 samples, respectively, totaling 2152 images. The selected architecture is Xception, utilizing stochastic gradient descent as the optimizer and employing ReduceLROnPlateau to adjust the learning rate during training dynamically. Standard data augmentation techniques are applied before classification. Notably, this approach significantly enhances the overall testing accuracy, achieving an impressive 97.22%. The study employs accuracy, precision, and AUC to ensure the validity of the results. In comparative testing against the existing methods, the proposed architecture outperforms existing models in terms of accuracy.