Optimization of YOLOv8 Algorithm for Welding Image Defect Detection Based on Spatial Attention Mechanism

This paper discusses the research of welding image defect detection using YOLOv8 algorithm combined with spatial attention mechanism. YOLOv8 is a leading technology in the field of object detection, which we have improved by introducing a spatial attention mechanism to improve the detection of defects in welding images. During the experiment, the loss value of the model gradually decreases from the initial 0.12, which shows the learning ability and optimization effect of the model in the training process. In particular, when the training iteration is about 100 times, the loss value drops to 0.02, indicating that the model has achieved good convergence. Although accuracy has decreased as recall rates have increased, this reflects an increase in the number of false positives as coverage of defect detection has increased. However, the results of the test set showed that both accuracy and recall improved with training, and the accuracy improved at a faster rate. Finally, the improved YOLOv8 algorithm based on the spatial attention mechanism can effectively detect and locate welding defects, showing excellent prediction performance. In conclusion, this study successfully combined the advanced YOLOv8 algorithm with the spatial attention mechanism to develop an efficient welding image defect detection method. This method not only improves the accuracy of defect detection, but also improves the convergence speed and