A Hybrid Computer Vision Framework Integrating Deep Learning with Optimized Machine Learning for Rice Leaf Disease Diagnosis

Rice leaf diseases cause high yield loss and are a major threat to food security especially in rice production area. The early and accurate identification of these diseases is therefore very important; however, the current methods of manual inspection are time-consuming and subjective and are often inaccessible for farmers living in remote areas. In order to solve this problem, we propose a hybrid framework of computer vision which combines feature extraction based on deep learning algorithm with optimized machine learning classifiers for automatic diagnosis of rice leaf diseases in this study. To begin with, images are enhanced by preprocessing techniques in order to augment visibility of the disease. Subsequently, deep features are extracted based on pre-trained convolutional neural network models, that is, ResNet101, ResNet152, EfficientNetB7, and DenseNet201. Thereafter, these features are classified using Support Vector Machine (SVM), Random Forest, and Gradient Boosting classifiers which are optimised using Hyperparameter Tuning. Experiments have been conducted on a publicly available rice leaf data containing eight diseased and healthy classes, the experiment result showed that DenseNet201 based SVM achieved the best performance with an accuracy of 95.95%, high accuracy of recall level, and an accuracy of the area under curve (AUC) approaching to 1. These results show that the hybrid models possess the capabilities of high model accuracy while maintaining computational efficiency, making the models applicable to real life scenarios in agricultural applications with marginal computational resources.