Banana leaf disease detection (BLDD) is an automated approach used to identify diseases in banana leaves for improving crop health and agricultural productivity. Accurate and real-time diagnosis of banana leaf diseases remains difficult due to environmental variability, visual similarity of symptoms, and the need for computationally efficient models. To address these issues, a YOLOv9 model is proposed, using advanced deep learning architectures for accurate banana leaf disease classification. Adaptive trilateral filtering is used to improve banana leaf image quality by lowering noise while preserving sharp edge information and disease lesion patterns. YOLOv9-based detection is used to correctly locate diseased patches in banana leaves, allowing for exact identification of infection sites. Furthermore, a Deep Belief Network (DBN) is developed to improve multi-level disease classification by learning discriminative representations from identified regions. The proposed framework classifies banana leaf conditions into healthy, cordana, pestalotiopsis, and sigatoka categories based on the extracted features, ensuring accurate and reliable disease severity assessment for precision agriculture applications. The YOLOv9 model achieves an accuracy of 91.9%.