Design And Implementation Of Defect Detection Software For Warp Knitted Fabrics Based On Deep Learning

Warp knitted fabric is the most significant production facility in the textile industry,This method has the disadvantages of low detection efficiency and high work intensity.With the development of image processing technology in the production of warp-knit fabrics,In the textile sector,some companies have developed an error detection system based on image processing which improves to some extent the error detection of workers.However,most existing error detection systems use statistical methods,spectral methods,and other error detection algorithms,Based on expert experience and unable to meet the requirements of industrial production of warp-knit fabrics.Therefore,research and development of warp-knit fabric fault detection software based on a deep-line algorithm have an important industrial application value and importance.Firstly,the camera is selected according to the detection requirements of warp fabric,The warp defect detection system and image acquisition platform are built by using an image acquisition card and other hardware to improve the robustness of training to the deep learning model.A data enhancement method based on image geometry and image pixel transformation is designed.Through the comparative analysis of object detection and semantic segmentation algorithms based on deep learning,the number of warp defect data sets is increased.To reduce the cost of algorithm development and the difficulty of software development,the algorithm which accords with the characteristics of warpknitted fabric defect detection is selected.Then u-net is selected as the text and the basic network of the detection algorithm.The variant form related to the residual structure is studied.The cyclic residual structure is introduced to optimize the code and decoding process and optimize the network degradation gate caused by too deep encrypted volume network According to the characteristics of "small defect background" of the light braided prepuce,the weight loss function is introduced in the training process of a deep learning model to reduce the background tendency of inspection results and improve the speed of model training.We find that the improved algorithm has a better detection effect.Finally,C# and C++ programming languages with open source tools such as Libtorch,are used to develop the deep learning defect detection module in the defect detection software system.It also provided a comprehensive description of the detected warp knit fabric defects in the human-computer interface using detect albums,maps,and information tables.Based on the data transfer mode and data transfer protocol,design and implement the data transfer and control module in the defect detection software system to complete the accurate labeling function of the PLC marking machine.The results show that the defect detection software system designed in this paper can reach the detection speed of 0.5?0.7m/s and the labeling error within 5cm,which can meet the needs of enterprises.