SF-YOLOv8: A New Metal Weld Defect Object Detection Method Using Natural Images

Current object detection algorithms often face a trade-off between accuracy and speed. While some algorithms achieve sufficient accuracy, they lack precision, and others are fast but fail to meet the accuracy requirements in real-time defect detection of aluminum alloy weld seams in natural images. To address this issue, this paper proposes an improved object detection algorithm based on YOLOv8, named SF-YOLOv8. First, the initial convolutional layer is replaced with a Focus module to enhance the model's ability to detect small defects. Second, the SimAM attention mechanism is integrated into the last layer of the CNN to improve the model's overall performance. Finally, the loss function is optimized and restructured. Experimental results show that the SF - YOLOv8 algorithm improves Precision, Recall, F1-score, and other evaluation metrics for detecting defective welds, cracks, excessive reinforcement, good welds, porosity, and spatter, compared to other detection methods. However, further research is needed to enhance detection speed and accuracy through algorithmic improvements or the application of new algorithms.