Research On Surface Defect Detection On The Basis Of Deep Learning

Fabric defect detection has become an important part of fabric quality control,and defect detection algorithm plays an important role in improving product quality and production efficiency.The traditional machine vision detection method relies on feature selection and extraction,the quality of feature determines the performance of the whole detection method,and the shallow learning model cannot comprehensively extract the useful features for defect detection.Deep learning is originated from artificial neural network,which can extract the characteristics automatically and efficiently from layer by layer,it has a widely application prospect in the surface defect detection.As a result,an approach of surface defect detection based on deep learning is described in this paper.Firstly,the advantages and application scenarios of different structures are analyzed for the problem of how to select the suitable neural network model for surface defect detection.Meanwhile,fabric surface images are used as experimental data,and different network models are selected for experimental verification.After analyzing the experimental results,convolution neural network is taken as the basic model for further research.In order to determine the basic structure of the model,different parameters are selected for comparison testing to catch the optimal parameters of the model,which can minimize the recognition error rate due to unsuitable initial parameters.Secondly,a surface defect detection model based on convolution neural network is proposed to solve the problem of insufficient nonlinear feature extraction and detection accuracy.In this model,multi-layer perceptron is used to improve the traditional convolution layer,which can enhance the ability of network to extract nonlinear features.At the same time,the algorithm is optimized to improve the convergence speed and recognition accuracy rate.Experiments were carried out using the fabric defect data set,the results show that the proposed method has higher recognition accuracy than the existing detection algorithms.Then,the distribution balance of data sets is caused by some defect samples is difficult to obtain because of the production process and other problems,which will affect the final recognition rate of the model.Therefore,a data expansion model based on the Deep Convolutional Generative Adversarial Networks(DCGAN)is proposed to ensure the balance of data sets.The model includes two parts: the generative network and adversarial network,new samples that are consist with the size and characteristics of the original sample can be automatically generated without the loss of useful information.It has better adaption than other data enhancement methods.Next,in order to solve the problem of the loss of effective information in feature extraction process,a method of surface defect detection based on multi-feature fusion convolution neural network is proposed.The feature extracted from multiple stages are merged,and more feature information is retained and sent into classifier together.The experimental results show that the proposed model is superior to the original model and can achieve higher recognition accuracy rate.The reliability and superiority of the proposed model are proved by comparing with the traditional defect detection method.At last,what have been done in this research is summarized and some suggestions for the follow-up study are put forwarded.