基于改进FasterNet-YOLOv8的焊缝表面缺陷检测算法OA
Weld Surface Defect Detection Algorithm Based on Improved FasterNet-YOLOv8
针对焊缝缺陷复杂背景干扰性强,检测精度和效率较低的问题,提出了一种改进的FasterNet-YOLOv8缺陷检测算法.在Backbone端更换FasterNet轻量级模型主干,捕获重要特征信息.将FasterNet-Block和卷积注意力融合模块(Convolution and Atten-tion Fusion Module,CAFM)开发到网络的特征提取模块中,设计了一种新颖的C2f-Faster-CAFM轻量级架构,减少网络的冗余通道的同时自适应捕捉全局关键信息.设计采用特征聚焦扩散金字塔网络(Feature Focused Diffusion Pyramid Network,FDPN)来增强多尺度信息融合提取能力,从而提高网络在多尺度场景中的鲁棒性和检测精度.实验结果表明,与原YOLOv8算法相比,Faster-Net-YOLOv8的精确率达到94.9%,召回率达到93.5%,平均检测精度提升至97.4%,提高了3.1%.
Aiming at the problem that weld defects have strong interference from complex backgrounds,and their detection accuracy and efficiency are low,an improved FasterNet-YOLOv8 defect detection algorithm is proposed.The FasterNet lightweight model backbone is replaced on the Backbone side to capture important feature information.FasterNet-Block and convolution and attention fusion module(CAFM)are developed into the feature extraction module of the network,and a novel C2f-Faster-CAFM lightweight architecture is designed to reduce the redundant channels of the network while adaptively capturing global key information.The feature focused diffusion pyramid network(FDPN)is designed to enhance the multi-scale information fusion extraction capability,thereby improving the robustness and detection accuracy of the network in multi-scale scenes.Experimental results show that compared with the original YOLOv8 algorithm,the precision of FasterNet-YOLOv8 reaches 94.9%,the recall reaches 93.5%,and the average detection accuracy is increased to 97.4%,with an increase of 3.1%.
李冠胜;阮景奎;王宸;闫伟伟
湖北汽车工业学院机械学院,湖北 十堰 442002湖北汽车工业学院机械学院,湖北 十堰 442002湖北汽车工业学院机械学院,湖北 十堰 442002驰田汽车股份有限公司,湖北 十堰 442000
矿业与冶金
缺陷检测YOLOv8FasterNet注意力机制特征聚焦扩散金字塔网络
defect detectionYOLOv8FasterNetattention mechanismfeature focused diffusion pyramid network
《机电工程技术》 2026 (2)
78-83,6
国家自然科学基金青年项目(51907055)湖北省教育厅项目(B2022275)中建科技研发课题(CSCEC-2022-Q-52)湖北省科技厅区域科技创新计划(2023EHA018)
评论