Zynq For Fabric Defect Detection System Design And Implementation

The fabric surface defects are main factors affecting fabric quality and price,defect detection is an important part in of textile quality control and quality inspection process.But the domestic market is not mature defect detection system.Therefore,according to the actual situation of China's textile industry,the use of efficient algorithms,lower hardware costs and research applicable to the actual production of the defect detection system is important.With the rapid development of electronic technology,embedded image processing system performance continues to improve,more feature-rich,collaborative design platform based Zynq hardware and software integrated ARM processor and FPGA programmable logic device,suitable for compute-intensive and feature-rich embedded type applications,has broad application prospects.Based on the embedded image processing applications,it is proposed based on ARM + FPGA defect classification system design and implementation.The main contents are as follows.(1)Design and Implementation classification of fabric defect image processing.In this paper,fabric defect as the main research object,the actual production of six common defect is classifited.First,median filtering and histogram equalization preprocessing,noise filtering is adopted in acquired image,to enhanced image defect information.Then,after the extraction of the image preprocessing feature values,using local binary pattern(LBP)and GLCM,and all of the local information of image texture features are described,and finally extracted feature value into a classifier to train and test.(2)The defect classification classifier selection.The paper for problem of small sample selection support vector machine(SVM)classifier,the classification of different kernel functions were selected for training and testing,and to strike the optimal kernel parameters through cross-validation.The final choice of the best classification is radial basis classification as a function of the kernel function.On the basis of improved SVM,using least squares support vector machines(LS-SVM),cross-validation through parameter optimization.Seek for cross-validation parameters in the process is very time-consuming,therefore bayesian framework is used to seek the best kernel parameters,and results were compared.(3)The system of Xilinx Zed Board as hardware systems,its core is a multi-core heterogeneous structure of ARM + FPGA.Zed Board running Linux on ARM systems,the use of USB camera image obtaining fabric by Qt and Open CV libraries tool defect classification algorithm,and algorithm implementation process via HDMI FPGA control and display.Defect classification algorithm by MATLAB first functional verification,then PC for programming and testing,and finally through the cross compiler,the migration execution algorithm development board,while the results of the test run,the system can achieve defect image acquisition and classification.The results show that the system real-time,high classification accuracy and low power consumption.