Multi-task Deep Learning Based Defect Detection For Lithium Battery Tabs

After the welding process of Lithium battery tabs, it is necessary to detect the surface defects of the welded products. The Gap is one of the common defects, and the defect forms are changeable, which brings a great challenge to the detection. This paper proposes a lithium battery tab gap defect technology based on multi-task deep learning model. The model takes U-Net as the architecture, ResNet as the encoder backbone network, and the decoding end connects up to three task-related networks, including defect area detection task network, defect contour detection task network, and defect binary classification network. The method is verified on the polar ear data set collected from the production line. The results of defect detection and product judgment show that the multi-task method has achieved good results, and the accuracy of product judgment can reach 98.8%. Compared with the single task method, it only needs 50% samples to obtain the same performance, which can effectively alleviate the problem of the insufficient number of on-site samples.