Abstract
Aiming at the problem of complex background interference and insufficient weld detection accuracy in steel structure weld recognition task, this paper proposes a target detection algorithm based on improved YOLOv5. By introducing the Coordinate Attention (CA) mechanism in the backbone network and integrating the CA with the C3 structure, the C3CA module is designed to effectively enhance the model's perception of the spatial position of the weld and the detection accuracy of the weld. Experiments show that the improved YOLOv5s-C3CA model achieves 93.79% mAP@0.5 on the self-made weld data set, which is 2.48% higher than the original model. At the same time, the amount of model parameters is reduced by 8.53%, and the floating point operation is reduced by 10.6%, which achieves a balance between detection accuracy and calculation efficiency. This study verifies the effectiveness of the coordinate attention mechanism in improving the feature expression ability of the model, and provides a solution for the automatic detection of welds in industrial scenarios.