Research On Detection And Classification Algorithms Of Fabric Defects In Industrial Pipelining

Fabric defects detection is one of the most important parts of textile quality control and management, however, the study on automatic textile defects detection is only in its infancy and until recently, there's still no mature real-time industrial pipelining system of fabric defects detecting made in China. Therefore, according to the actual situation of the Chinese textile industry, it's of great significance to research novel technology and effective algorithm to develop automatic textile defects detecting system which is at low hardware cost.In this thesis, some fabric defects detecting and classifying algorithms are studied based on studies of the corresponding researches in existing literatures, and they are as follows: 1. How to develop the speed of the detecting system to meet the requirement of real-time detecting. 2. How to classify the defects with limited samples, nonlinearity and high-dimension.The main contributions of this thesis are listed as follows:Firstly, the hardware platform of the fabric defects detecting system based on DSP is built. The principle theory of the system is introduced, and the system functional modules are analyzed, besides the types of the functional defects detecting module based on DSP are selected.Secondly, the fabric defects detection algorithm which works on the DSP is designed. After pre-disposing, binary conversed by dynamic threshold and morphology eroded, whether the bitmap has defects or not is quickly judged by statistical strategy. An efficient algorithm to detect the Cloudiness is also well designed too.Thirdly, the algorithm used to extract the fabric defect characters is designed, and which is combined by the gray co-occurrence matrix and the Gabor Filter. Combining the Gabor Filter can capture much more useful information of the texture and using fabric defect characters of gray co-occurrence matrix can increase the veracity of the algorithm of the defects classifying.Fourthly, the algorithm of defects classifying based on SVM is designed and the software implementation on the PC is completed. The difficulties that classify the defects with limited samples, nonlinearity and high-dimension are resolved by the defects classifying algorithm based on SVM.Finally, the work carried out in this thesis is briefly summarized and some remarks on further improvement of the motion control system are also presented.