Automated Banana Leaf Disease Classification and Treatment System Using DenseNet-201 Architecture and Edge-Based IoT Platform

Banana growing is complicated by diseases like Black Sigatoka, Fusarium Wilt, and various viral infections such as BBMV and BBTV. All these together accounts for a worldwide economic loss of more than $10 billion yearly, with the percentage of damage of different areas ranging from 20% to 80%. Manual visuals inspection which is the prevalent method for detecting these diseases is slow, very much dependent on the inspector's moods as well as the place he/she is located, and is also quite likely to mix-up or reject the correct early-stage infections. Only about 65-75% of the inspections are accurate. Automating the use of the IoT-enabled Raspberry Pi for real-time disease classification and treatment delivery is the main goal of this research. Urban contributions include: (1) DenseNet-201 that has a classification performance of over 98.51% accuracy, 98.62% precision, 99.58% specificity, and 98.19% F1-score which makes it the best choice; (2) Raspberry Pi-based real-time monitoring platform with automated diagnosis (<2 seconds inference time) and precise treatment dispensing, leading to pesticide saving of 68.3%; (3) comprehensive evaluation on a curated dataset of 500 banana leaf images across five classes with rigorous cross-validation. Statistical significance testing (McNemar's test: χ<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>=18.47, p<0.001) confirms the superiority of our approach. The system fills a major need in the field of precision agriculture by being an affordable solution that smallholder farmers can easily access, thus facilitating early disease intervention, minimizing crop loss, and supporting the world's green agricultural practices