Orange Disease Classification Using Hybrid Deep Feature Extraction and Machine Learning Classification

Abstract

Orange cultivation is a major crop worldwide. However, citrus plants are vulnerable to many diseases, including Greening, Black Spot and Canker, which lead to crop quality deterioration and yield loss. Traditionally, detection of disease relies on human scouting that is hardly repeatable, so there is a growing interest in automated techniques to address the problem. The present work suggests a new hybrid method for orange disease detection that integrates deep feature extraction with backbones EfficientNetB3, and classical machine learning classifiers - K-Nearest Neighbors (KNN), Support Vector Machine (SVM) and Random Forest (RF). This proposed experiment utilized deep features in conjunction with traditional classifiers to gain a tradeoff between good performance, low computing power demand, and model transparency thus making the solution feasible for developing countries or rural areas. Testing was done with a publicly available dataset on Kaggle which comprises four classes: Greening, Canker, Black Spot, and Healthy oranges. The experimental results revealed that SVM recorded more overall accurateness of 94.6% while the instances causing the misclassification were mostly between visually very similar Canker and Black Spot samples. Hence, the proposed method is a practical example of combining deep learning and classical machine learning approaches for an efficient, correct and deployable orange disease identification tool that can support precision agriculture.