The Study, Based On Genetic Algorithms And Support Vector Machine Blended Yarn Cross-sectional Image Segmentation Method

Textile is one of the most important traditional industries in China, of which textile testing is the critical part. Most of the traditional textile testing relies on manual operation and subjective assessment. How to improve the objective assessment and the degree of automation in testing is the key point in achieving sustainable development in this field. The rapid development of the computer graphic processing and the machine learning technology is pouring in fresh vitality to many fields. How to combine the computer graphic processing technology with the machine learning theory to get a better testing result is the most important technical difficulty.This article mainly centers on the method to train the support vector machine (SVM) to separate yarns into two categories of wool and viscose by catching the cross section graphic of wool/viscose blended yarns. The main research is as follows:(1) A commonly used graphic preprocessing flow was proposed based on the segment objects stated in this article.(2) An extraction was put forward, judging by their characteristics of wool and viscose in the blended yarn. In order to obtain the sample collection used for the training of SVM, a comparison was set up between their characteristics and the following aspects such as regular circle, ellipse, rectangle and circumscribing polygons.(3) A basic framework was discussed about the genetic algorithm (GA) and the SVM. With the aid of the framework, the selection operator of GA was improved and the type of kernel function of the SVM was selected as well. This improvement and selection was proved reasonable by an applicable example.(4) A separation method was analyzed through demonstration on the basis of the image of GA and SVM. Grounded on the application of separation according to the cross section graphic of the blended yarn (wool and viscose), it proved that this separation method is feasible.(5) A comparative analysis of parameter optimization between a non-heuristic grid search and a heuristic genetic algorithm was dealt with. It mainly comparatively analyzed the speed of obtaining the categorizer model between them. The result is the combination of the genetic algorithm and the SVM method is appreciated. In the end, the structural parameters of the blended yarn (wool and viscose) were extracted and analyzed.