Research Of Raw Silk Defect Detection System Algorithm Based On Machine Study

The production and consumption of raw silk are mostly applied in China. At present, rawsilk inspection still employs the traditional blackboard inspection. The major drawbacks of thismethod are strong subjectivity, poor repeatability and labor costs since it mainly adoptsmacroscopic view or artificial visual inspection. With the rapid development of computertechnology, intelligent recognition technology and digital image processing technology, usingmachine vision technology to detect the quality of raw silk will become the trend in the future.This topic refers to raw silk blackboard test of national standard GB/T1797-2008raw silkand the results of silk electronic inspection. Firstly, raw silk defect classification standards will beidentified in this experiment. Secondly, several commonly used image-filtering methods, such asgaussian filtering, average filtering and median filtering are analyzed and compared. Theexperiments show that gaussian filter not only can eliminate the raw image noise but alsopreserve the raw silk size, shape and texture characteristics. Lastly, the three threshold algorithminvolving global threshold, adaptive threshold and manually threshold are analyzed. Comparedthe experimental results, manually threshold method is the optimal threshold algorithm.According to the experimental results, image segmentation is the best when the threshold takes188. In addition, the defect outline of raw silk is extracted on the basis of image segmentation.The four major characteristics of the raw silk defect value, which are area, duty cycle, length andwidth are extracted based on the defect outline.This paper based on OpenCV platform, three kinds of defect will be converted into threetwo-class classification problem. Experiments show that the SVM as a typical of two-classclassifier performs a good classification results. Expanding the SVM to more class classificationproblem that based on RBF kernel function of SVM, the default parameters of SVMclassification method and PSO algorithm are applied in optimizing parameters. Compared twokinds of classification, the effect of optimizing by PSO algorithm of penalty factor C and theparameter ? of RBF kernel function of the SVM classification is better than the defaultparameters of SVM classification method.