Design And Implementation Of Fabric Intelligent Inspection Vision System For Flexible Garment Production Line

Garment industry as a major industrial chain is also an important pillar industry of the people’s livelihood.The traditional mode of production technology is backward,labor intensity is high,working environment is poor,and production efficiency is low,and it is facing fierce market competition pressure.Therefore,the introduction and integration of intelligent equipment and machine vision technology is an inevitable trend for the establishment of automated production lines,product upgrades,and the development of the clothing industry toward intelligent,green,and automated development.In the process of garment production,the fabric is prone to defects,folds,burrs,blurring and other phenomena due to its soft texture,different materials,and the influence of equipment and technology,which brings some difficulties to detection,identification and positioning technology.Based on the above background,combined with the needs of intelligent clothing production,this paper designs and implements a fabric intelligent inspection vision system based on deep learning.By carrying out related research work on fabric intelligent detection,to further improve the degree of intelligence of the clothing production line.The main research content of this paper includes the following aspects:(1)Research and compare several common camera calibration methods.Based on the analysis of traditional camera calibration method,active vision camera calibration method and camera self-calibration method,the camera pinhole model,distortion model and Zhang Zhengyou calibration principle are mainly introduced.Finally,Zhang Zhengyou calibration method is selected to establish the camera imaging geometric model and obtain the internal and external parameters of the camera in the system.(2)In view of the complex characteristics of fabric and the shortcomings of the existing defect detection technology,a fabric defect detection method based on deep learning is proposed.Based on Faster R-CNN model,this method uses deep residual network as feature extraction network,and introduces multi-scale feature detection to increase detailed shallow features.Then,the Soft Max classifier is regularized to reduce the intra-class spacing and increase the inter-class spacing.Finally,the Soft-NMS algorithm is used to replace the traditional non-maximum suppression algorithm to improve the detection accuracy of the model.The experimental results show that the optimized model has faster convergence speed and has better detection effect on small targets and dense defects.(3)In this paper,based on the analysis of the current situation of pocketing technology in garment production,combined with the existing fabric edge detection technology,a bidirectional cascade network structure of multi-scale edge detection is used for edge detection of pocketing workpiece.After obtaining the edge map of the workpiece,the template matching algorithm based on edge feature is used to extract the edge information of the workpiece.To verify the experimental results,the NMS algorithm is used to generate the final prediction results.The results show that the bidirectional cascade network can extract more clear and complete fabric edges,and the template matching method based on edge features can obtain accurate edge position information.Based on the above research results,the overall construction of fabric intelligent detection vision system is realized.The feasibility and application value of the vision system are proved by relevant experimental analysis and industrial application.