Research On Grey Cloth Defect Detection Based On MB-LBP And Improved Faster R-CN

Fabric defects seriously affect product quality and corporate efficiency,and traditional manual inspection restricts the improvement of production efficiency.The rapid development of machine vision and deep learning has brought new development opportunities to the field of defect detection,greatly improving detection accuracy and efficiency.Given the various types of flaws,large-scale differences,and complex interference factors in the research on fabric flaw detection,this paper proposes a detection method combining shallow feature extraction and deep learning based on existing research.The main research contents are as follows:First of all,to solve the problem of poor image contrast and unobvious target features captured by industrial cameras,this paper proposes an algorithm that can not only improve image contrast,highlight defective targets,but also enhance target details.Use guided filtering as an algorithm framework to protect the details.The CLAHE contrast-improved image is used as the input image,the unsharp mask detail enhancement image is used as the guiding image,and guided filtering is performed to achieve image enhancement.This verifies the superiority of the proposed algorithm through experiments.Then,aiming at the problems of many types of defects,a large amount of feature extraction data,and poor generality of the algorithm,a defect location algorithm based on MB-LBP shallow feature extraction operator and SVDD single classifier is proposed.Start with a normal fabric sample,divide the normal sample into several non-overlapping sub-blocks,extract the MB-LBP features of the sub-blocks and train the corresponding single classifier,divide the detected samples into blocks and send them to the classifier for classification,identify and mark the defective sub-blocks Coordinates to realize defect location.This method effectively avoids the difficult problem of feature extraction from multi-defect samples and transforms the positioning problem into a classification problem.Experiments show that the flaw location accuracy of this method can reach more than 96%.Finally,aiming at the slow speed and low accuracy of traditional classification methods,a defect classification algorithm based on an improved deep learning model Faster R-CNN was proposed.The main improvements are: 1)Replace the feature extraction network with Res Net50,increase the network depth,improve the feature extraction ability,use the residual structure and bottleneck structure to optimize the network model,and reduce the amount of calculation;2)Adjust the size and proportion of the RPN network anchor box,Fit the defect distribution of the data set,and improve the algorithm’s ability to detect multi-scale targets;3)Use hole convolution to offset the problem of multiple down-sampling lost targets and improve detection accuracy.Experiments show that the m AP value of the improved algorithm on the test set is 0.94,which effectively improves the detection accuracy of the data set defects.