An Algorithm of Detecting GSS-YOLO Object Geared to Surface Defects of Strip Materials

XIAO Yilei; WANG Cheng; QU Yi; KONG Yakang; CHEN Xiancong; WANG Xiaoxu

doi:10.3969/j.issn.2097-1915.2025.05.004

Abstract：In view of the problems that intelligent detection technology in the process of strip surface defect detection is low in accuracy,missing in detection and false in detection,a GSS-YOLO object detection op-timization algorithm is proposed based on YOLOv8n.The model is to organically integrate the Neck net-work of the GOLD-YOLO algorithm with the Ultralytics to improve the detection accuracy of the model for defects of different sizes and shapes.In order to balance the gap in identifying the effect of different de-fect types and reducing the complexity of the network structure,two lightweight modules in the Slim-Neck structure are introduced,i.e.lightweight convolution VoVGSCSP and efficient channel attention mechanism SimAM,enable to improve the detection accuracy and generalization ability of the model,and to simultaneously limit the expansion of the computational and weight volume of the model.It appears from relying on the classical strip surface defect dataset NEU-DET,thaw experiment,lateral comparison experiment that the average accuracy of the model is 3.7%higher than that of the benchmark model,and the gap between the detection accuracy of various defects is reduced,thus the detection accuracy meeting not only the requirements,but also guaranteeing at running speed.In comparison with the current main-stream models,this model has a certain advantage in detection accuracy,and is in reference value to the application of defect detection in actual industrial production.