Research On Fabric Defect Detection Based On Neural Network

China is the world’s largest expoter of textile production and trade.In the development of economy,the textile industry has always played an important role.However,recent reports indicate that China’s textile industry has less competitiveness due to the factors such as the cost of raw materials and rising labor costs.At the same time as the development of textile industry,the textile products quality problem get more attention from manufacturers and purchasers,one of the important factors that affect the quality defects is the defect of textiles.However,in China and some other developing countries,most of textile fabric defect detection process still rely on manual detection,it has many shortcomings,which mainly includes the following two points:(1)Low production efficiency and large labor cost.Under the traditional manual detection,it is difficult to meet the modern production requirements,and the increase of labor cost increases the cost of fabric production.(2)High rate of artificial missed detection.People will be tired after working for a period of time,especially in the visual sense,which causes many fabric defects to be omitted during the inspection,which can not guarantee the quality of the products.With the development of image processing technology,adding visual inspection system on the basis of traditional cloth inspection machine will be the trend of fabric defect detection system in the future.Although computer vision system can not replace human beings,it has advantages of continuous work and high production efficiency in the process of industrial production.At present,there are related fabric defect detection systems both at home and abroad,and there are many researches on defect detection algorithms,but there is no mature system that has been widely used.In this thesis,the existing defect detection system is studied,the current situation of defect detection system at home and abroad is studied,and the detection algorithm is researched and improved.The main research contents are as follows:(1)The existing defect detection system,defect segmentation and classification algorithm are introduced and analyzed.At the same time,the fabric defect detection system built in the laboratory is described in detail,including the components of the whole component,the selection of hardware models,the operation and use of the software part,etc.(2)The fabric image is pre processed to improve the quality of the fabric image.Such as removing image noise,improving contrast,etc.(3)Study and compare all kinds of defect segmentation algorithms.Firstly,the pulse coupled neural network is used to segment the pre processed image,and the particle swarm optimization algorithm is used to select the parameters.The experimental results show that the detection effect of this method is more obvious.(4)Feature selection and classification of defect images.In this thesis,the image feature extraction from multi angle,and based on the principal component analysis method,from the initial feature space to extract the most several features reflecting fabric defects.BP neural network and PNN neural network are used for classification and comparison.The innovations are as follows:(1)Particle swarm optimization is used to select the network parameters when using pulse-coupled neural network to segment the image.The experimental results show that using this network to perform defect image segmentation is better.(2)In the selection of defect features,the principal component analysis of the geometric features and texture features of defects has reduced the feature dimension,which not only retains the valid features but also reduces the complexity for the subsequent classification.(3)BP neural network and probabilistic neural network are respectively used for classification.There are many applications based on BP neural network for classification.In this thesis,an experimental comparison is made between the two networks.Experiments show that PNN is better than BP neural network.