earticle

영어

This study aims to develop a lightweight object detection model that ensures both real-time responsiveness and high accuracy for rice disease diagnosis. The model targets major rice leaf diseases such as rice blast (Riceblast), bacterial blight , and brown spot . A total of 3,234 high-resolution images (2400× 1080) were collected and used as training data. Based on the YOLOv5s architecture, the proposed YOLOv5-V2 model removes the Focus layer and integrates the ShuffleNet V2 backbone to reduce computational complexity and improve inference speed. The model achieved a size of 6.45MB, with a performance of mAP-0.5 = 89.6%, mAP-0.5:0.95 = 66.7%, precision = 91.3%, and recall = 0.85. These results demonstrate the model's potential for real-time disease detection and automated diagnostic systems in agricultural settings.