Research on automobile welding quality inspection and defect identification algorithm based on deep learning

Welding quality is an important factor affecting safety and reliability in automobile manufacturing. Traditional detection methods have some limitations, such as low efficiency and easy interference. In this paper, an algorithm for automobile welding quality inspection and defect identification based on deep learning is proposed. The improved ResNet-34 network architecture is adopted, and data preprocessing methods such as rotation enhancement and Gaussian noise are combined. The multi-task loss function is used to comprehensively classify, regress and segment tasks, and the model performance is optimized. Experiments show that the algorithm is superior to the comparison models such as Faster R-CNN, U-Net and YOLOv8n in classification accuracy, [email protected] detection and IoU segmentation, especially in the precise location of tiny defects. The model further improves the feature extraction ability and generalization performance by improving strategies such as ASPP module and CutMix enhancement.