A Feature Difference Network for Defect Detection on the Surface with Complex Pattern of Ceramic Tile

The surface defects of ceramic tile greatly affect the service life of ceramic tile. At present, many detection methods of ceramic tile surface defects are mostly used for ceramic tiles with monochrome background or simple texture. However, many tiles with complex and irregular surface patterns are used in practical applications, but many methods cannot effectively detect surface defects in such tiles. This paper presents a double input feature difference network structure to overcome the limitation. First, a double input channel is constructed to extract features from the template image and the defect image respectively. Next, a method of feature difference is performed at different depths to suppress the background interference and prevent misclassification between different defect categories. Then a parameter-free attention module is embedded in the backbone to improve the ability of feature extraction. Experimental results show that this model effectively improves the mean average accuracy of 8.3% and the recall rate of 11.7%.