YOLOv8s-GSW: a real-time detection model for hexagonal barbed wire breakpoints

The detection of defects in Hexagonal barbed wire is a crucial aspect within the realm of building structure safety and monitoring. Traditional manual detection methods for Hexagonal barbed wire defects suffer from long detection times and low efficiency. Existing deep learning models such as YOLOv5 and YOLOv8 fail to strike a balance between detection accuracy and speed in the context of Hexagonal barbed wire detection. Therefore, this paper proposes a Hexagonal barbed wire defect detection model based on YOLOv8s named YOLOv8S-GSW, which includes replacing part of the convolution of the backbone network with GSconv to improve the detection speed of the model, and replacing C2f Block(C2f) in the feature extraction network with VOV-GSCSP structure. It further improves the detection speed and balances the detection accuracy. Additionally, the WIoUv3 loss function is introduced as a replacement for the original CIoU loss function, resulting in accelerated network convergence along with enhanced detection accuracy. Experimental results demonstrate that YOLOv8S-GSW achieves mAP@0.5 and mAP@0.5 0.95 scores of 94.32% and 45.70%, respectively-exceeding those achieved by YOLOv8s by 3.26% and 2.75%. Furthermore, this improved model reduces its volume by 10.97%. Consequently, the enhanced YOLOv8S-GSW exhibits superior capability for detecting Hexagonal barbed wire defects, while meeting industrial requirements.