Research Of Real-time Visual Detection Technology For Fabric Surface Defect Based On SoC

In industrial product quality detection, machine vision detection technology gradually replace inefficient artificial detection method, traditional machine vision still exist low pertinence, low real-time and detection rate, the bloated system and other problems, in the increasing of market demand and the trend of enhanced competition. To make up for the shortcomings of traditional visual detection system, based on the advantages of embedded system like high-performance, low-power and tailoring design. This paper adopt nankin, pure color cloth as the main object of detection, analyzing in detail on the basis of visual detection equipment and technology domestic and overseas, design the detection system based on the embedded Zynq SoC and CMOS camera.The real-time operation system control camera image acquisition, image processing based on PL and PS, information transmission and data exchange between PS and PL, internal and external communication, and improve and optimize the design of the detection algorithm.Specific to the problems of image salient detection, like low accuracy, low resolution, low real-time performance, combined complementary of low-level image features, a salient region detection algorithm based on RGB and CIE LAB color features and edge feature is presented. Analysis the implementation mode of Algorithm, Design DoG filter, color space conversion, DCT/IDCT calculation module based on PL optimally via hardware-software co-design technology, and customize IP core, the rest of the algorithm and control processing can be implementated on ARM software. Experiment results show the algorithm suppress background better, compared with 5 kinds of popular algorithms, and has higher efficiency and rapidity in fabric defect detection.Specific to the problems of defect area fast positioning and texture extraction, a detection algorithm combined on salient analysis and optimized EGF is presented. Adopt differential revolution method by training free-defect images to gain the optimized EGF during the offline period, gain the optimized EGF; in the online period, realize median filter, color space conversion, three-channel EGF processing in the LAB space based on PL, and transfer image data to ARM via AXI VDMA bus, calculate feature vectors of the pre-filter and filtered images, extract defect salient region, determine the threshold and segment defect region. The algorithm suppress background effectively and highlight the defect region accurately, has high effectiveness as to block, line type fabric fabric detection, the average rate of four types fabric defect detection can reach 94.4%, improve the instantaneity.To verify overall performance of the embedded vision detection system based on SoC, build the IP integrated design of real-time image acquisition, processing and display, analysis the design and realization of detection algorithm based on PL and PS, construct real-time Linux system multi-task scheduling platform, and test the fabric defect detection system, experiment results demonstrate that it has good performance.