Research On Nonwoven Fabric Defect Detection System Using Gabor Filter

The defect detection is important to nonwoven producing. The traditional methodsdepend mainly on labor costs. With the development of digital image processing andcomputer science, real-time defect detection method based on machine vision is catchingmore and more attention. Visual detection system, which is able to adapt to a complexindustrial environment and has feature of high precision and real-time, has practicalsignificance to the problem of nonwoven fabric defect detection.Aiming at the different frequency characteristics of the various types of defects in thesurface of the nonwoven fabric, a Gabor wavelet filter method is used in nonwoven imagedefect segmentation. Gabor wavelet have the capacity for responding the change ofdefects gray value in local region from different scales and directions. In order to increasethe detection accuracy, a parameter selection scheme of optimal Gabor filters which havedifferent peak response to the defects of different frequency characteristics is proposed inthis dissertation. In this way, different kinds of defects exhibit different gray scale in resultimages. So defect area can be easily split up. Experimental results indicate that theproposed defect detection algorithm has good accurate performance, but whose real-timeis still insufficient.In order to improve the real-time of detection algorithm, a parallel computation basedon CUDA is proposed. The dissertation achieves a nonwoven fabric defect detectionparallel algorithm on the GPU platforms, which is based on the theory of Gabor waveletfilter. The method, that transmitting multiple threads in the GPU with one threadprocessing one image pixel, greatly improving the computational efficiency. Experimentalresults indicate that CUDA-based parallel algorithms greatly enhance the detection rate,while ensuring accuracy.The realization method of the system is presented from both hardware and softwareaspects. And the function and implementation methods of its components are described.specific solutions are given to many problems encountered in the implementation process.And the prospect to the possible industrial application of detection system is also present.