Explainability in Orange Disease Detection through a Res-Inception Framework Integrating Deep Learning Techniques

This paper presents a deep learning architecture Res-Inception model, that combines InceptionV3 and ResNet50V2 to classify different orange diseases with high accuracy. The images selected for the present study were 1614 in number and composed of four classes, namely, black spot disease, citrus canker, citrus greening, and healthy oranges, and chosen to enable precise disease identification for agriculture purposes for orange farming. To improve the detection accuracy, more sophisticated processing of input data was performed, such as resizing, flipping, rotation, shifting, zooming, and standardization by using the Z-score method combined with Median Absolute Deviation. The proposed Res-Inception model was found to have an accuracy of 98.93%, which is better than that of other CNNs, CNN-SVM, and YOLOv7. One of the significant differences in the proposal of this work is based on the explainability of the model. It has provided the integration of explainable AI such as Grad-CAM, Score-CAM, and LIME which enables The comprehension of how the model makes decisions in real-world applications in agriculture. This capability not only enhances transparency but also empowers users to validate predictions in real-time disease management scenarios. Thus, the Res-Inception approach is an innovative and effective method for early detection of diseases of orange plants in precision agriculture and further enhances the use of artificial intelligence to help farmers worldwide improve crop yields and increase food security.