PROJECT TITLE :

Multistage GAN for Fabric Defect Detection

ABSTRACT:

Fabric flaw identification is a fascinating yet difficult topic. The complexity of both fabric textures and defects has resulted in a plethora of ways for detecting fabric defects, yet these methods are still far from ideal. A GAN-based framework for fabric defect identification is presented in this paper. The suggested fabric defect detection system is capable of learning existing fabric defect samples and automatically adapting to varied fabric textures during different application periods, taking into account the current challenges in real-world applications. For fabric defect detection, we use a deep semantic segmentation network that can identify a wide range of defect kinds. A multistage GAN was also trained to create realistic flaws in previously defect-free samples. For starters, a texture-conditioned GAN is trained to look at the conditional distribution of defects on a variety of textures. We want to be able to make reasonable-looking defects in new fabrics. Once the faults have been formed, a GAN-based fusion network fuses them to specified regions. Last but not least, the well-trained multistage GAN contributes to the fine-tuning of the semantic segmentation network in order to better detect faults under various scenarios. A wide range of typical fabric samples are used in our extensive trials to test the detection efficiency of our new approach.