Mango fruit disease detection and classification using MobileNetV3_large model

Mangoes are appreciated globally for their flavor, nutritional value, and economic significance, which are one of the most significant tropical fruit crops. Various fruit diseases also negatively impact mango production, decreasing yield and quality. This shows the critical necessity for precise and timely disease identification. This study targets the issue of a lack of effective automated diagnosis tools and a lack of research on diseases affecting the mango fruit rather than the leaves. To effectively identify and categorize the main diseases of mango fruit, such as Alternaria, Anthracnose, Black Mold Rot, and Stem-End Rot, a deep learning model had to be developed. A publicly available Kaggle dataset was used and augmented to 8,297 samples to enhance generalization. The preprocessing phase involved background removal, image resizing, noise reduction, and z-score normalization to improve image quality and maintain consistency. Features were obtained using a customized MobileNetV3-Large architecture that significantly reduced the number of trainable parameters from 5,507,432 to 566,325 and the model size from 21.8 MB to 6.29 MB, making it highly suitable for deployment on mobile and resource-constrained devices. Featured extracts were classified using the XGBoost algorithm, developing a hybrid MobileNetV3-Large–XGBoost model. Experiment results showed 98.95% test accuracy, reflecting excellent reliability and precision in disease detection. Findings show that the proposed model not only enhances automatic fruit disease detection but also provides a viable solution for real-time disease monitoring and control. The deployment's potential helps farmers and agricultural experts with the capability to detect early infection, reduce crop loss, improve disease management approaches, and enhance sustainable mango growth.