To address the issue of traditional algorithms being limited in type and having substantial debugging difficulties in metal surface defect detection, a distributed aperture detection imaging system was designed to achieve super-resolution reconstruction. On this basis, an improved template-difference detection algorithm was proposed. Firstly, adaptive non-local mean filtering and large-scale median filtering were applied for preprocessing, followed by difference operation. Then, adaptive binarization was realized based on the structure similarity index measure. Finally, the scratches, void defects, and rivets were classified by utilizing the prior feature information of defect area, shape, and color. Experimental results indicate that the improved template-difference method achieves the best performance in both recall and precision. The average recall of scratch defects reaches 98% and average precision is 62.57%, significantly superior to traditional algorithms such as Sobel, Prewitt, and Laplacian (with records of 53.74%, 47.78%, and 25.72%, respectively). This system scheme enhances the efficiency and accuracy of metal plate defect detection, possessing significant practical application values.