Efficient Detection and Classification of Orange Diseases using Hybrid CNN-SVM Model

Abstract

Orange is an important citrus fruit grown globally, and its consumption is encouraged by health-conscious individuals due to its nutritional value. Classifying oranges is important for quality control, sorting, and grading in the food industry. For the production of high-quality oranges, farm-based disease prediction is not utilizing technology to its full potential. A hybrid version is proposed in this research paper for the categorization of six common disorders of oranges, namely Penicillium, Scab, Anthracnose, Melanose, Phytophthora, and Citrus Canker, using a blend of the classifier - Support Vector Machine and ANN prototype - Convolutional Neural Network. With CNN being accustomed for feature derivation and SVM being utilized for classification, the suggested model leverages the best aspects of both algorithms. Using a dataset of 4,864 orange photos, the suggested hybrid model’s performance is assessed, and as a result, an accuracy of 88.13734% is achieved. Our sensitivity analysis indicates that the form, size, and texture of the lesions were the most crucial characteristics for categorizing orange-colored illnesses, followed by their texture and color. The effectiveness of utilizing a hybrid model for illness diagnosis in citrus fruits is shown by the postulated hybrid model’s superior performance over existing classification models like SVM, Random Forest, and K-Nearest Neighbor (KNN). The impeccable competence of the proposed hybrid model makes it suitable to be employed in automated disease detection systems to make prompt and well-informed decisions about disease management and prevention, thereby enhancing citrus crop productivity and quality.