Fabric Defect Detection Base On Pulse Coupled Neural Network

Fabric defect detection technology is one of the most commonly used technology in the production of cloth. This technology can effectively improve the fabric quality and reduce the emergence of defects. Before the industrialization of defect detection, fabric defect detection is mainly depends on human eyes, which always be affected by some other factors, for example, lighting, environment, physical illness, psychological problems, etc. In the present stage of the fabric defect detection technology, the main research is on the plain weave fabric defect detection, and many research achievements have been applied in industrial production. But for the detection with complex patterns, there haven’t find a fast and efficient detection method yet. This paper will from the perspective of bionic vision mechanism, make use of the structure of pulse coupled neural network segmentation of complex patterns, extracting the high-level feature of human cognitive characteristics, and judge the flaw location pattern according to the correlation of image pattern.In this paper, in order to improve the defect detection technology, the pulse coupled neural network is applied to fabric defect detection. Many image processing technology are applied in the technology of fabric defect detection, such as image segmentation and image feature extraction. In image segmentation, we use the minimum cross entropy criterion to obtain the optimal iterative cross entropy, then put the optimal number of iterations into pulse coupled neural network for image segmentation, and get the two value image. The two value image sequence contains lots of feature information in the original image, we extract two kinds of characteristic sequences by mathematics changes, which are the ignition time sequence and the information entropy of time series respectively. In order to find whether the detection of fabric containing defects, we regard the two series as the feature vector of the image, and use the similarity measure to calculate Euclidean distance. In addition, this paper introduces a kind of pulse coupled neural networks model--intersecting cortical model, based on the output of the two value image analysis, we proposed the new detection method using the autocorrelation of image pattern.