Research On The Machine Vision-based Platform For Real-time Fabric Defect Detection

Quality control plays an important role in enhancing the core competitiveness of enterprises. Recently, machine-vision-based inspection techniques have been widely used quality control applications due to its distinct advantages such as non-contact, anti-jamming, high-speed and high-precision. In textile industry, fabric defect detection is a major part of quality control for final products; at present, in china fabric defect detection is still performed by human inspectors in textile malls, a process suffering from low efficiency, high missing rate and high false detection. To overcome those drawbacks, an automated fabric inspection platform that can meet real-time and accurate requirements was designed and implemented.First of all, this thesis conducts an analysis about research background, summarizing some major challenges, i.e. high-speed transmission for large amounts of image data and real-time processing. Then the overall design scheme about the platform is presented, which consists of four parts:fabric drive system, illumination&imaging system, image acquisition&processing system and human-computer interaction system, and the structures and functions of these systems are detailed in turn.Secondly, starting with the hardware design, this thesis elaborates the image acquisition&processing system which is the core system of the platform. Aiming at the difficulty of the proposed platform, an integrated design scheme using Camera Link, FPGA and DSP is proposed:Camera Link high-speed transmission technique is used to process large volumes of data&high data transmission requirement; FPGA technique is used to reduce system development costs, shorten the development cycle, and enhance the flexibility of the system design; DSP technique is used to perform fabric defect detection algorithm in real-time. Around the integration of these three techniques, the system hardware design work focuses on the Camera Link module, FPGA module, and DSP module.When finishing the platform hardware design, this thesis details design and program the software part of the platform including the software functions and composition of the platform, software design and program of the image acquisition&processing system, human-computer interaction system and communication procedures between them.Finally, after debugging both platform’s hardware and software, the AR spectral analysis algorithm is selected to assess the performance of the proposed platform. The experimental results show that the platform can achieve91%of correct detection rate in the real-time requirement. At the same time, the experimental results verify that the hardware part of the platform is stable and reliable, and the software part of the design is competent for all expected functions, which means overall performance of the platform is able to meet the design expectations.The success of research and development of this real-time defect detection platform provides a good foundation for real-time verification of others defect detection algorithms, and also provides a platform of visual quality assurance for other kinds of textiles.