The inability to accurately and efficiently identify tobacco diseases can significantly impact both the yield and quality of tobacco crops. Based on Deep learning, this study focuses on enhancing the precision, efficiency, and accessibility of tobacco disease identification while minimizing associated costs. The research explores the application of deep learning techniques for this purpose. Initially, samples of 19 prevalent tobacco diseases were gathered from tobacco cultivation regions in Henan Province. These samples were categorized based on expert diagnoses. Following data augmentation, a comprehensive dataset was compiled. Next, the YOLOv5 network model was examined. To facilitate real-time, user-friendly identification suitable for mobile deployment, the model underwent pruning and optimization to reduce its complexity without compromising accuracy. The model was then trained using the prepared dataset. After training, the model was adapted for the Android platform, and a dedicated application was developed. This application not only identifies diseases but also offers insights into their causes and prevention strategies. The final phase involved experimental validation, which demonstrated that the optimized model operates effectively on Android devices, achieving a recall rate exceeding 90% for the majority of the diseases studied. This advancement represents a significant step forward in the practical application of AI for agricultural disease management.