Development Of Defects Detection Algorithms For Digital Textile Printing Based On Embedded GPU

Textile printing defects detection is one of the most critical aspects of quality control.At present,the traditional manual inspection is mainly adopted,but there are some shortcomings such as high cost,low accuracy,and high labor intensity.Therefore,it is of great significance to develop a robust defects detection system to replace the traditional manual inspection,which may greatly accelerate the development of the automated printing defects detection.This thesis develops a textile printing defects detection system based on the embedded GPU and the digital printer platform.Firstly,the printed fabric images are preprocessed,which can effectively enhance the contrast of the images of defect part and the normal part while reducing the image noise.Following that,for different types of defects,two detection algorithms are developed.One is developed based on statistical projection and multi-scale LBP,the other is based on GMM.By using the image texture information and modeling the image,the two proposed algorithms can accurately locate the defects.In addition,in order to meet the real-time requirements in the industrial process,this paper designs optimization methods with parallel computation using CUDA,which greatly improves the performance of the detection algorithm.Experiments show that the printing defects detection system developed in this paper can well meet the actual production line requirements.The maximum time consumption can be controlled within 1s and the detection accuracy reach up to 95%or more,which can effectively replace the traditional manual inspection.