Research On The Method Of Fabric Color Difference Detection Based On Machine Vision

Fabric color difference is the primary reference index in fabric quality evaluation.The color difference problem will cause the fabrics to be returned to the factory for remanufacturing,and this process requires multiple stripping and re-dyeing procedures,which makes the high production cost and low resource utilization rate of the company.Therefore,it is of great significance to study the method of fabric color difference detection.At present,most domestic small and medium-sized enterprises still use manual detection to judge the color difference of fabrics,and a few large and medium-sized enterprises with economic capacity mainly use foreign detection instruments to detect the color difference.In view of the actual demand of enterprises for fabric color difference detection,this thesis adopts the method combining machine vision and image processing technology to realize the fabric color difference detection and the automatic classification of fabric color,which effectively improves the efficiency and accuracy of fabric color difference detection.This thesis mainly carries out the following research work:Firstly,in order to obtain high-quality fabric images,the hardware selection and design of the image acquisition system were carried out.The use of high-precision color linear array camera improves the image quality of the fabric from the source;at the same time,the conveyer belt motion device was designed and used to meet the acquisition speed required by the color linear array camera;the D65 standard light source was selected at a 45 degree irradiation angle to make the fabric under appropriate lighting conditions,so that the color of the collected fabric image is closer to the actual fabric color,which reduced the detection error of the system.Secondly,research on the detection algorithm of fabric color difference was carried out.Several color spaces and filtering algorithms were compared and analyzed,and the bilateral filtering algorithm was selected in RGB color space to preprocess the image,which can remove the influence of noise generated in the process of image acquisition.Then the RGB color features of the fabric image were extracted and converted to the CIELAB color space with better uniformity.The CMC(1:c)color difference formula with better visual consistency was used to calculate the color difference value,and the color difference level was classified according to the color difference evaluation criteria hierarchy,which can detect the color difference between different fabrics,and realize the dyeing inconsistency detection of fabric.In order to detect the color difference in the same fabric,the qualitative and quantitative algorithms were studied.The superpixel segmentation method,saliency algorithm and adaptive threshold method were studied,and the qualitative analysis of fabric dyeing unevenness was carried out.The experimental results showed that these three methods can visually and quickly distinguish whether there is a color difference on the same fabric.In order to quantitatively detect the dyeing unevenness of the fabric,the fabric was equally divided into blocks and the color difference of each region was calculated,and the main color difference in the same fabric was judged according to the size difference of the color difference value.At the same time,the visual interface of fabric color difference detection was designed,which can simply and intuitively display the results of fabric color difference detection.Finally,using Inception_v3 and ResNet50 neural network classification algorithm for fabric color classification,and the comparison experiment results showed that the classification accuracy of ResNet50 network was higher than Inception_v3 network,with an accuracy of 99%and good stability.Therefore,ResNet50 network was selected for fabric color classification.This thesis mainly studies the algorithm of color difference detection and color classification of pure cotton fabrics,which has a certain reference value for realizing the automatic control of fabric color quality.