A novel ensemble deep learning approach for detection and classification of onion diseases

Abstract

Abstract Ethiopia, highly dependent on agriculture, places significant economic value on onions, a pivotal vegetable crop. However, onion production in Ethiopia faces considerable challenges due to various diseases such as purple blotch, downy mildew, damping off, and the iris yellow spot virus. Current disease management relies on manual inspection methods, leading to variability in effectiveness. In this study, we introduce an ensemble model that extracts and integrates features from VGGNet and AlexNet to classify multiple onion diseases from agricultural onion images. The extracted deep features are then classified using Softmax, k-Nearest Neighbor, Random Forest, and Support Vector Machine classifiers with accuracy of 93%, 75.11%, 94.31%, and 96%, respectively. The model achieved a remarkable accuracy of 96% with the SVM classifier. As a future work, increasing the dataset size and the scope to include other types of onion diseases will improve the robustness and necessity of the model.

Keywords

• Onion diseases
• Ensemble model
• Support vector machine