MobileNetV3 for Mango Leaf Disease Detection:An efficient Deep Learning Approach for Precision Agriculture
Sukruth S Puranik, Siddharth R Hanamakkanavar, Anupama P Bidargaddi, Vighnesh V Ballur, Pratham T Joshi, S. M. Meena, Uday Kulkarni
Abstract
India, renowned for its vast agricultural landscape, especially in mango cultivation, faces substantial challenges with leaf diseases that significantly affect mango yield and quality. These issues pose economic challenges for the agricultural sector. The task of accurately diagnosing these diseases is complex and time-intensive. Addressing this, our study leverages advanced machine learning techniques and high-processing computational resources. We utilize the publicly available MangoLeafBD(MBD) dataset, which contains 4000 high-resolution images of mango leaves, sourced from diverse orchards in Bangladesh using mobile phone cameras. This dataset, with its meticulously edited images, is crucial for effectively training machine learning models. Using the MobileNetV3(MV3) architecture, known for its enhanced accuracy and efficiency, we have developed a mobile application for real-time mango leaf disease diagnosis. This application, integrating Android's Camerax API, facilitates immediate, on-site disease detection. The retraining of MV3 on the MBD dataset has achieved an impressive accuracy of 98%, demonstrating its potential in agricultural technology. This advancement not only contributes significantly to plant pathology but also offers a practical solution for farmers, enhancing disease management and promoting sustainable agricultural practices.