Design And Implementation Of Multi-computers Parallel Processing Fabric Defect Detection System

In the Process of production of textile industry, the control of fabric defect wasan important part which impacted on the quality assessment of fabric productsdirectly. Currently, the method of artificial defect detection were still used by majorityof domestic textile enterprises, with the expansion of production needs and theimprovement of the quality requirements, this method had not kept up with themodern requirements of the enterprises, thus an automatic defect detection systemwhich could rapidly detect defects was urgently needed.Defect detection system required large amount of image data (40MB/s or more)to process, high complexity image processing algorithm, high performancerequirements of processing system, therefore, the defect detection system that wasdesigned in this paper composed multiple computers to improve the processingperformance of the entire system by parallel processing of data which was based onmultiple computers. Pressing the server and client way among multiple computers,which was similar to the C/S mode. Computers connected via Ethernet, high-speedtransmission of data based on UDP protocol.As data reliability assurance did not provided by UDP protocol, this paper built asimple, practical and reliable transport protocol based on the existing protocols, andmade specific description on working mode and the composition of the structure ofthe protocol, etc.With reliable transmission, we could further build multi-computers platform. Bythe summary of image processing algorithms, this paper put forward a general modelof the image processing algorithms, and based on the model, decomposing thealgorithm process by the form of task, finally, the system was designed based onprimitive task. This paper introduced the definition of the task structure, the mutualconversion between task and data stream, assignment of tasks and track, and someother desired composition module of the system. About defect detection algorithm, this paper used multi-channels Gabor filters asthe basis of the detection algorithm to achieve the detection algorithm by combiningwith maximum entropy and morphological processing, it’s proved that the algorithmwas effective. This paper introduced in detail each step on defect detection algorithm,finally, making corresponding improvement to the detection algorithm to run onmulti-computers platforms. In the experiments, we got the required processing timeof different computer number, calculated speedup, made a significant accelerationeffect.On defect classification algorithm, this paper extracted three characteristics ofdefect area S, defect width W and defect height H, used the SOM unsupervisedlearning neural network to clustering, it’s proved that the algorithm was effective.