Research On Image Recognition Algorithm Of Yarn-dyed Woven Fabric Density

Woven fabric density is a key indicator of woven fabric quality testing.Early detection of woven fabric density relied on a large number of manual use of some simple inspection tools,such as magnifying glasses,analytical needles and photo mirrors.Such work is boring and labor intensive,but if it is working for a long time,it is a great test for the visual and mental state of the staff.Under such circumstances,it is even more impossible to ensure the accuracy of product testing and effectiveness.Since the beginning of the 21 st century,the field of computational imagery has flourished,bringing convenience to the industry and bringing great improvements in accuracy,speed,workload,and labor intensity.Such a change not only reduces the burden on workers,but also improves the production efficiency of the entire assembly line.However,at present,the use of machine vision to detect the density of woven fabrics is more mature about the density detection of monochromatic woven fabrics,and the accuracy of multi-color woven fabric density detection still has a higher room for improvement.To this end,this paper has developed a study on the image recognition algorithm for color fabric density.In order to improve the accuracy of color fabric density detection,this paper firstly uses the multi-directional light source to illuminate the fabric when sampling,specifically providing the light source in four directions,and the rotation angle of the adjacent position is 90 degrees at each position.Sampling is performed separately during the lighting.For the four fabric images obtained by sampling,after selecting the appropriate wavelet base,the image is decomposed by wavelet transform,and then merged according to the optimal fusion rule.Finally,a new image is generated by inverse wavelet transform.The purpose of this is to bring together the characteristics of each picture,enhance the contrast of the yarn and the gap,and reduce the interference of color.The fused fabric image cannot be directly detected for density,but also requires pre-processing,including filtering and contrast enhancement.The filtering uses Butterworth low-pass filtering,while the contrast enhancement uses the Retinex method to eliminate noise in the image.Interference and enhanced contrast.After the filtering and contrast enhancement are completed,the tilt detection is also required,because the fabric is artificially placed during the detection process,and it is difficult to ensure the presence or absence of tilting.For the method of tilt detection,this paper proposes a method combining Gaussian pyramid and Hough transform.Compared with the original Hough transform,the accuracy of tilting is greatly improved.Finally,it is necessary to use the gray projection method to detect the density of the woven fabric.Compared with the traditional solid color fabric density detection,the only difference between the two is that for the solid color fabric,the Fourier transform can be fast and accurate.It is calculated that the Fourier transform cannot calculate the density of the colored yarn fabric because of the color interference.After obtaining the projection curve by using the gray projection method,it is found that there are many local peak points on the projection curve,which has great interference to the accuracy of density detection,so it needs to be eliminated.The method adopted in this paper is to use local weighted regression.The algorithm performs smoothing,but this algorithm requires a width value to be manually set.Therefore,this paper proposes an algorithm to effectively solve the problem,and finally accurately detects the density of the yarn-dyed woven fabric.