Abstract: To meet the demand for automated detection of welding quality in lithium battery tabs in production enterprises, a computer vision-based lithium battery tab welding quality detection system is designed. Considering the optical characteristics and precision requirements of the positive and negative electrode materials, two optical systems are designed to image the positive and negative electrodes of the lithium battery, respectively. A multi-adaptive template matching algorithm is proposed to extract the welding seam areas of the positive and negative electrodes. An adaptive threshold segmentation method is used to gray the welding seam areas. Gray morphological operations are employed to suppress noise in the grayscale image and highlight the weld pad feature points.The contour extraction algorithm is used to extract the weld pad feature points, and the number of feature points is used to detect lithium battery pole ear cutting defects. The U²-Net semantic segmentation model is introduced into the welding quality detection task. The dataset for the positive and negative electrode welding seams, as well as the burnthrough, faulty welding, and welding oxidation defects, is augmented using data enhancement techniques to address the small sample size and small target problem in welding defect detection. The U²-Net semantic segmentation model and contour extraction algorithm are used to measure welding seam dimensions and detect defects such as faulty welding defects. The system trial operation results show a defect detection rate of 99.98% and a false detection rate of only 0.06%, meeting the actual needs of enterprises and achieving rapid, non-contact, and accurate automated detection.