Citrus diseases are the main issue for agricultural productivity, which cause a reduction in yield and quality, thus causing economic loss; also, we see that traditional methods of diagnosis, which are very subjective and prone to error, and thus we need to suggest accurate automated detection systems. In this study, we present a hybrid model that integrates conventional machine learning with deep learning models. We used the Gray Level Co-occurrence Matrix (GLCM) for feature extraction, which is classified via machine learning algorithms like k-Nearest Neighbors (KNN), Decision Tree (DT), Linear Discriminant Analysis (LDA), and Random Forest (RF). At the same time, deep convolutional neural networks (DCNNs), GoogLeNet, and ResNet18 were used for comprehensive classification. To improve robustness, we integrated multilayer features from GoogLeNet and ResNet18 with machine learning classifiers; in particular, we combined GoogLeNet with RF. The performance and quality of the proposed system were evaluated using three datasets, each made up of four classes: Blackspot, Canker, Fresh, and Greening. Dataset 1 performed the best results, whereas datasets 2 and 3 were used to test the models' generalizability. In dataset 1, RF and GoogLeNet achieve 99.44% accuracy, and ResNet18 achieves better accuracy at 99.72%. In the hybrid framework, which combined features of GoogLeNet and RF, achieved 100% accuracy, which is considered a very strong indication of its very high dependability and performance in the field of identifying citrus diseases.
