A Fabric Appearance Defect Detection Method Based on PCS-YOLOv5 Lightweight Model
In response to the problems of large number of parameters,high computational complexity,and slow detection speed when deployed on ordinary industrial computers in existing fabric appearance defect detection models,this paper proposes a lightweight model PCS-YOLOv5.Firstly,PP—LCNet is used to replace the YOLOv5 backbone network to achieve model lightweight and accelerate inference speed.It introduces the CBAM attention module into the Neck network to suppress interference and focus on important features,thereby improving the accuracy of object detection.It modifies the bounding box regression loss function to SIoU to enhance the accuracy of defect localization.The experimental test results show that compared to the YOLOv5 original model,PCS-YOLOv5 performs better in mAP@0.5 under the condition of basic consistency,the detection speed is increased by 10.2%,the number of parameters is reduced by 56.8%,the computational complexity is reduced by 63%,and the model weight is reduced by 56%,which can meet the requirements of online detection of fabric appearance defects on site.
国家科技期刊平台
NSTL
万方数据
