Research On Nonwoven Fabric Defect Online Detection System Using Machine Vision

Fabric defect is the primary factor affecting the quality of the fabrics and defect detection is very important in quality management of non-woven fabric. Traditional human inspection is devoid of consistency and reliability. With the development of modern industry and fabric quality and production demand is growing, defect detection based on machine vision technology has become the research hotspot and development trend.In this thesis, according to the background gray inconsistent problem of different images, image processing methods of gray-value normalization and homomorphic filtering are presented, and thus can basically eliminate the influence of illumination distribution which is inconsistent. This thesis which is based on the practical application, aiming at the common defects of non-woven fabric, uses the gray level co-occurrence matrix to describe the defect features. However this method can distinguish defect, but it can't provide more defect information. The thesis researches the two-dimensional Gabor filter detection algorithm with multi-resolution characteristic, and this can effectively separate defect after threshold segmentation. 2D Gabor filter algorithm is more universal, but due to its large amount of calculation the real-time performance is not insufficient, so it is an urgent need to solve issue in practical application.In order to improve the real-time performance of Gabor filter algorithm, the thesis has researched two aspects which are algorithm optimization and GPU acceleration realization:(1) real component of Gabor filter plays a more important role in the detection results, so we just apply the real component in convolution arithmetic when we realize the detection algorithm, thus can efficiently reduce half of the amount of calculation. The thesis proposes a fast implementation of Gabor filter, which is based on separable convolution method, and it can greatly improves the computational efficiency.(2) A fast realization of Gabor filter method based on GPU parallel computation has proposed, respectively using CUDA global memory, texture memory and shared memory to realize the Gabor filter in different ways. Comparative experiments show that using the shared memory optimization implementation of Gabor filter algorithm based on separation of convolution is the most efficient way, which can reach 8~20 times speedup.Finally, the thesis briefly introduces the structure and implementation of defect detection system from two aspects of hardware and software. Besides, all works of this thesis are summarized, and some efficient improvement solution has put forward at the end of the thesis.