Automatic Anomaly Mark Detection on Fabric Production Video by Artificial Intelligence Techniques

In the previous era, humans played important roles in all aspects of industrial work. However, they indisputably made many errors that can be mitigated by automated manufacturing, thus revealing the importance of the latter. In this paper, an autoencoder-based fabric-defect detection method via video is presented. The fabric-production video is segmented using frames to produce images, and then a VGG16-based autoencoder is applied to reconstruct the original image. In the proposed scheme, each fabric-production image is normalized to 256 x 256 pixels, which provided excellent results compared with using various margin sizes in our experiments. We used the structural similarity index (SSIM), which measures similarity when checking whether image regions are normal or defective. Moreover, a masking algorithm is utilized to improve detection accuracy. Based on our experiments, we found that 0.5 is an appropriate value for setting the SSIM threshold as it produced the best detection performance with a defect detection accuracy of ~99%.