Research And Software Development Of Fabric Defect Detection Method Based On Machine Vision

With the high-speed development of the current society and economy,it is an increasing demand for high quality fabric,but the fabric defect will greatly debase the quality of the products.In our country mainly adopts the traditional way of artificial detection,and there are some big shortcomings such as low detection accuracy,high abour costs and high labor intensity.In order to greatly reduce the cost of detection manpower and effectively improve the competitiveness of textile products,the dissertation studies the high-precision and high-speed fabric defect detection methods by using the machine vision technology and deep learning technology.Realizing the automatic detection of fabric defect has important theoretical significance and value of practical application.The main work and results are as follows:(1)Analyze the software function and performance requirements of fabric defect detection,design the hardware system scheme.The linear array light source with high concentrative property is selected and the same end irradiation is used.The light intensity is controlled by digital light source controller.According to the characteristics of fabric transmission and fabric width,linear array camera is selected as the image acquisition sensor.Choose a smooth surface,stable transmission and speed adjustable transmission device to ensure the reliability of the fabric in the process of transmission.(2)A new method based on frequency domain transform and image morphology is proposed to locate defects in fabric.Firstly,the original image is preprocessed.Texture features are extracted by texture filter,and defects are pre-identified by GMM classifier model.Then gaussian filter is used to construct a band-stop filter to perform frequency and filter on the spectrum after Fourier transform.Finally,on the filtered spatial domain image,the image morphology method is used to extract the defect location and extract the fabric defect.The image processing method combining spatial and frequency domain enhances the detection robustness,and the detection speed is faster,which can meet the real-time requirements.(3)Compared with the mainstream target detection neural network framework,Faster R-CNN with better performance is selected as the fabric defect classification network.The data sets are made by making public data on the Internet and actual fabric images,and the common fabric defects are divided into seven categories.The full connection layer in the network is modified,and the global average pooling layer is used instead to reduce the network parameters.The parameters of anchor frame are modified to improve the accuracy of network detection.(4)The deep convolution residual network is used to construct a deeper network structure to replace the VGG network in the original Faster R-CNN.The residual network can learn the shallow information and enhance the feature extraction of fabric defects.In order to make full use of the information of each layer of feature extraction network,this dissertation forecasts the results of pyramid feature fusion after sampling different feature layers and shallow layers,and makes suggestion boxes prediction for each scale.(5)Based on QT5 platform,the existing machine vision algorithm library and deep learning framework are used for software development.The software requirements are analyzed and the main function modules of the software are realized,including the functions of software login,collection and display,detection information statistics and human-computer interaction.