Research On Recognition And Detection Of Fabric Defects Based On Cascade R-CNN

Textile industry has been playing an important role in China’s national economy.However,it is difficult to identify and detect the fabric defects in the textile industry.In recent years,the development of machine vision technology,especially the rapid development and successful application of deep learning in computer vision technology,has provided great research value for fabric defect detection in textile industry.Therefore,this paper will combine the advantages of deep learning in visual inspection technology to study the fabric surface defect detection method based on machine vision technology,and explore the application of feature pyramid network and deformable convolution in fabric surface defect detection.It is finally used for the location,recognition and classification of different fabric defect data.Experimental results show that the proposed neural network model with cascade structure achieves inspiring results in defect location and recognition.This paper research object detection and recognition method of machine vision system.In order to improve the recall rate of defect detection and reduce the missed detection in the detection process without reducing the accuracy,this paper designs the network model based on the two-stage cascade neural network model.First,in view of the difficulty of detecting small defect and how to extract the features of them in the deep convolutional layer to ensure that their semantic location information is rich,we study small defect detection method based on the feature pyramid network algorithm.Secondly,in view of the irregular shape,different shape and different position of defects,the paper proposes a method of feature extraction based on deformable convolution neural network for fabric defect detection.In order to concentrate feature extraction on the deformed object area and achieve the purpose of extracting the deformed defect feature area,the deformable convolution neural network is fused to learn the offset of feature points at different positions,and the position information of spatial sampling is further adjusted to make the convolution kernel offset at the sampling points of feature input.In order to solve the problem of large difference between similar defects and insignificant difference between different defects,improve the detection accuracy of network model and make it have applicability and robustness,this paper propose an algorithm which integrating deformable convolution and pyramid network in Cascade R-CNN(IDPNet)based on two-stage network.The IOU of the network model is gradually increased to obtain high-precision detection results.Finally,based on the research content of this paper,experiments are carried out on two different kinds of cloth data to verify the effectiveness of the network model.and IDPNet is compared with Cascade R-CNN and Faster R-CNN in the main indicators respectively.The experimental results show that the proposed method has better detection results.