To address the complexities,multi-scale nature,and efficiency issues related to rice disease detection,we developed the DEFFN-YOLOv5 rice leaf disease algorithm using deep learning.We focused on studying rice diseases,including bacterial leaf blight,rice blast,and brown spot.To enhance disease detection accuracy,we improved the original YOLOv5 algorithm by introducing the PixelShuffle up sampling module to restore image details.Additionally,we enhanced feature extraction capabilities by incorporating deformable convolutions and a light weight ECA channel attention module.We further improved context information capture across different scale sand channels using BiFPN to enhance information interaction within the model,thereby improving object understanding and localization.Experimental results demonstrate that the enhanced DEFFN-YOLOv5 algorithm achieved an average precision(mAP)of 86%in target detection,a 3%improvement over the original YOLOv5 algorithm.Meanwhile,computational requirements were reduced by 4.6 GFLOPs,a 27.85%reduction compared to the original YOLOv5 algorithm.These enhancements make DEFFN-YOLOv5 a superior performer in rice disease detection.
维普
万方数据