基于可变形卷积和注意力机制的生丝疵点检测算法OA

In order to solve the problem of false detection and missed detection caused by small raw silk defects and variable morphology,a raw silk defect detection algorithm based on deformable convolution and attention mechanism was proposed.Taking YOLOv8n as the benchmark model,firstly,the deformable convolutional DCNv2 was integrated into C2f to form a new C2f-DCN module in the backbone network,and the arbitrary sampling shape characteristics of the deformable convolution was used to adaptively fit the geometry of the defects,so as to improve the feature extraction ability of the model for irregular defects.Secondly,the ECA attention mechanism was added at the end of the backbone network to suppress useless information such as background noise through cross-channel interaction,so as to improve the attention of the model to the defect feature information.Finally,a P2 detection head was added to the neck to capture shallow semantic information,and a four-branch detection layer structure was constructed to enhance the response ability to small targets.Experimental results showed that the mAP@0.5 and mAP@0.5∶0.95 of the proposed algorithm were reached 95.4%and 75.9%respectively,which were increased by 3.3 percentage points and 9.0 percentage points higher than that of the original algorithm,and the inference speed was 65.2 frames/s.The algorithm not only effectively realizes defect detection and reduces the phenomenon of false detection and missed detection of defects,but also has a better detection speed,which can meet the requirements of real-time detection.