Orange Disease Detection Using Pretrained ResNet50 and LIME for Explainable AI

This paper presents a method for detecting diseases in oranges using a pretrained ResNet50 model. The diseases targeted are citrus canker, black spot, and citrus greening, in addition to fresh oranges. The goal is to achieve high classification accuracy while making the model more explainable to humans by using LIME (Local Interpretable Model-Agnostic Explanations) for interpretability. The dataset consists of 991 training images and 99 test images, divided across the four categories. The model achieved a test accuracy of 96.97%, a test loss of 3.6373, and a ROC AUC score of 0.978..