Research On Machine Vision Inspection Algorithm For Warp-knitted Fabric After Finalization

The setting link is the last processing process in the dyeing and finishing workshop.In this process,the hot air tenter is used to set the warp knitted fabric.In this process,there will be excessive pattern correction or failure to meet the correction requirements,and the pattern detection after setting needs to be completed manually.The efficiency of manual detection is low,the consciousness of subjective judgment is strong,and only part of the pattern sampling of embryo cloth can be completed.To realize the construction of digital chemical plant,this subject takes the finalized warp knitted fabric pattern as the research object,and uses machine vision algorithm to process the image,segment and extract the pattern and compare the template.First of all,the actual investigation of the factory workers to detect the pattern process of warp knitted embryonic cloth,and improve the low efficiency process in the detection process.According to the characteristics of warp knitted fabric as flexible fabric,the shortcomings of pattern extraction and recognition of warp knitted fabric at home and abroad are analyzed,and the establishment of pattern detection system for different kinds of warp knitted fabric is proposed.It includes hardware detection system and algorithm detection system.Hardware detection includes camera,camera lens,light source,image acquisition card,encoder,etc.pattern detection algorithm mainly includes pattern image preprocessing,pattern complete contour effective extraction and template matching detection algorithm.Then,the image collected by the hardware system is spliced into a complete pattern,and the pattern stitching algorithm based on the combination of template matching method and RANSAC algorithm is used.The stitched pattern is denoised by multi-dimensional filter combined with median filter and mean filter,so as to overcome the disadvantages of traditional filter denoising algorithm,such as poor recognition of embryonic cloth pattern noise.In order to enhance the bright and dark areas of the fabric,a low complexity image enhancement algorithm based on cuckoo algorithm for parameter optimization is adopted.This method has high contrast while maintaining the details of embryo cloth pattern image.Then,an improved three-dimensional maximum inter class variance(Otsu)method is proposed to effectively segment the warp knitted fabric pattern after pretreatment,so as to overcome the shortcomings of the traditional three-dimensional Otsu method,such as poor noise resistance,high computational complexity and long running time.Compared with the traditional three-dimensional Otsu method,the improved three-dimensional Otsu method has strong robustness,high accuracy and reduced calculation time by 60%.Canny edge detection algorithm is used to connect the segmented pattern edge pixels into effective edges,eliminate irrelevant false edges,and extract the complete and effective pattern contour of embryonic cloth.In order to ensure the accuracy of the subsequent pattern template matching algorithm,the pattern contour of warp knitted fabric needs to be corrected horizontally,and the pattern tilt correction is completed by Hough transform algorithm.Finally,PN net deep learning framework is used to complete the pattern feature learning training.Resnet-18 network based on inception mechanism,relu activation function are used to improve PN net network.Design pattern matching process and interface development.Compared with matchnet pattern matching network,improving PN net network will improve the accuracy of pattern recognition.The accuracy of network recognition is 98%,and the recognition time of single pattern is about 100 ms.