In this study, an automatic defect detection method is proposed for screen printing in battery manufacturing. It is based on stationary velocity field (SVF) neural network template matching and the Lucas-Kanade (L-K) optical flow algorithm. The new method can recognize and classify different defects, such as lacking, skew, and blur, under the condition of irregular shape distortion. Three critical processing stages are performed during detection: (1) Image preprocessing was performed to acquire the printed region of interest and then image blocking was carried out for template creation. (2) The SVF network for image registration was constructed and the corresponding dataset was built based on oriented fast and rotated brief feature matching. (3) Irregular print distortion was rectified and defects were extracted using L-K optical flow and image subtraction. Software and hardware systems have been developed to support this method in industrial applications. To improve environment adaptation, we proposed a dynamic template updating mechanism to optimize the detection template. From the experiments, it can be concluded that the method has desirable performance in terms of accuracy (97%), time efficiency (485 ms), and resolution (0.039 mm). The proposed method possesses the advantages of image registration, defect extraction, and industrial efficiency compared to conventional methods. Although they suffer from irregular print distortions in batteries, the proposed method still ensures a higher detection accuracy.