Research On Online Non-woven Fabric Defect Detection Technology Based On Machine Vision

At present most domestic enterprises still rely on human eye to complete the inspectionwork, however this not only slows inspecting speed, and creates a heavy intensity of labor,also the final results would be greatly affected by inspection personnel’s experience,proficiency etc, and this method lacks of reliability. In order to meet developmentrequirements of the current textile industry, it is urgent to develop a stable and fast accuratefabric defect on-line detection methods.This paper first introduces the trend of research and application of Non-woven fabricdefect detection; Describe overall design scheme of Non-woven on-line defect detectionsystem and related theoretical basis; Analysis real-time requirements of the practicalapplication of the algorithm; In order to utilize texture character of non-woven surface, comeup with a scheme based on Gabor filter detection methods combined with OTSU method forimage segmentation. When dealing with the key problems of Gabor filter parameters setting,conduct experiment comparison and analysis each variable to the final detect results, anddetermine reasonable filter parameters. Through experiment on different types of defectsverified the validity of the algorithm. Also makes a comparative study with typical defectdetection methods, and prove the superiority of defect detection method based on Gabor filter.On the other hand, defect classification is another key problem in the system design, we tookthe method based on BP neural network and conducted research for its composition and thenumber of different layer neurons, the final classification result on commonly appeareddefects verify the effectiveness of the proposed algorithm. In the end, made conclusion of themethod and pointed out existing problems and improvement ideas.This paper puts forward and realizes relevant solutions of the non-woven fabric defectdetection using the OpenCV library of image processing. The experimental results show thatthe system runs stably, and the calculated results are accurate enough to meet the real-timerequirements.