Enhancing Welding Defect Detection: An Improved YOLOv8 Model with Feature Augmentation and Scale Fusion

In real-world scenarios, the detection of welding defects encounters challenges posed by complex background interference and multi-scale target variations. To address these issues, this paper proposes an enhanced YOLOv8 model. Firstly, the Convolutional Block Attention Module is introduced into the backbone network to bolster the model's feature extraction capabilities. Secondly, the adoption of the Bidirectional Feature Pyramid Network (BiFPN) enhances the PAN structure, thereby improving the model's ability to fuse targets of different scales. Finally, three Efficient Channel Attention (ECA) mechanisms are incorporated into the model's detection layers to enhance its ability to extract meaningful information. Experimental results demonstrate that our proposed model achieves precision and mean average accuracy rates of 89.7% and 90.1%, respectively, representing a 3.1% and 2.6% improvement over the original model. This indicates that the improved model is more effective in detecting welding defects.