YOLO-VDCW:A new lightweight surface defect detection algorithm for strip steel
The defect detection technology of strip surface is very important to ensure the product quality up to the standard of strip steel.Aiming at the problems of complex structure,insufficient utilization of computing resources and low detection accuracy in current strip surface defect detection models,a new lightweight strip surface defect de-tection algorithm named YOLO-VDCW was proposed.Firstly,the VanillaNet module was introduced into YOLOv8 to improve the efficiency of computing resources utilization and realize the lightweight of the model in the backbone network.Secondly,the C2f module was replaced by C2fDSConv to accurately transmit gradient information and fur-ther improve computing resource utilization and performance.In addition,the coordinate attention module was em-bedded after C2fDSConv,and the coordinate information was introduced to enhance the target positioning accuracy and perception ability.Finally,the CIoU loss function was replaced by Wise-IoU loss function to accurately measure the similarity between target frames and improve the defect detection performance of the model.On the NEU-DET dataset,the average detection accuracy(mAP)of YOLO-VDCW reaches 79.8％,compared with YOLOv8n,the average detection accuracy is increased by 3.8 percent point,the calculation amount and parameter number are re-duced by 34.1％and 36.8％respectively,the detection speed is increased by 37.9％,and the model volume is only 4.9 MB.The experimental results show that compared with other algorithms,YOLO-VDCW can effectively im-prove the accuracy and speed of strip surface defect detection while ensuring lightweight.
NETL
NSTL
万方数据
