Aiming at the problems of large number of parameters,poor detection effect of small objects,and instability of the quality of artificial annotation datasets in the detection of surface defects of continuous casting billet,this paper proposes a surface defect detection model of continuous casting billet based on the improved YOLOv7 algorithm.Firstly,the MobileNetv3 block module is introduced to reduce the amount of calculation,parameters and complexity.Integrate the two-layer routing attention mechanism to achieve more flexible computing distribution and content awareness.Finally,the WIoU loss function is used to replace the original loss function to make the model obtain better generalization ability.The detection accuracy of the improved model reached 89.3%,and the detection accuracy of corner transverse cracking was increased by 6.6%.The weight file size has been reduced to 56.5%,and the detection speed has reached 98 fp/s.The results show that the proposed algorithm has small parameters,fast detection speed and high precision,which can meet the requirements of online detection of surface defects of continuous casting billet.
surface defect detectionYOLOv7continuous casting billetattention mechanismloss function
NETL
NSTL
万方数据
