An Improved YOLOv7 Algorithm for Solder Defect Detection of Lithium Battery

The structural integrity of welded poles in power lithium batteries is closely related to the driving safety of the electric vehicles. Aiming at the problem of missed detection resulted from minor defects such as small coverage area and low pixel during the process of laser welding of traditional-pole for lithium battery, an improved YOLOv7 welding defect detection algorithm is proposed in this paper. First, an efficient channel attention mechanism C3SE module is introduced to improve the model's ability of extracting deep important features. Then, the improved BiFPN structure replaces PANet to improve the model's efficiency of feature utilization and the ability to express multi-scale targets. Finally, the MP structure uses the separate-merge operation and cascades the SE module in the subsequent convolutional layers to avoid losing the fine granularity of features. The experimental results show that the mAP of the improved algorithm for detecting defects reaches 96.4%, which is 1.2% higher than the original algorithm. Our work can provide important reference for similar tasks of welding defects detection using target detection scheme.