Research On The Test Method Of Fabric Wrinkle Resistance Based On CNN

The fabric wrinkle resistance affects the service life and appearance performance of the garment,effective and objective testing and evaluation methods are of great significance to guide garment design and production.However,the fabric wrinkle resistance test methods have shortcomings:1)The wrinkle recovery angle method has many manual measurement steps,which requires high requirements for the operator;2)Automatic measurement using laser power,poor measurement stability;3)It is difficult to characterize the wrinkle resistance of the whole fabric only by measuring the recovery angle of the warp and weft;4)The standard of appearance method is still subjective evaluation method,with low efficiency and low objectivity,and there is a big difference between the reference model and the wrinkles produced in the actual wearing process of the clothing.In view of the shortcomings of the wrinkle recovery angle method and the appearance method in measuring the fabric wrinkle resistance,this paper carries out related research from these two perspectives respectively,the specific research contents and results are as follows:(1)Aiming at the deficiencies in the manual measurement of the wrinkle recovery angle method,such as the manual measurement needs to transfer the sample,the reading is affected by subjective factors,and the automatic measurement reading is not accurate,a multi-directional dynamic wrinkle recovery test device for garment fabrics is designed and manufactured in this paper.This device can simultaneously measure the wrinkle recovery angle of the fabric in four directions.According to the characteristics of the device,a new angle index?_iis established.Through the analysis of the collected wrinkle recovery angle images,it is found that the shape of wrinkle recovery angle can be divided into four categories.Trigonometric function method,slope method,image registration method based on joint entropy and image registration method based on feature points are used to calculate the four types of wrinkle recovery angle.The results show that the slope method is suitable for all types of wrinkles.Taking 11 kinds of woven fabrics as the experimental objects,comparative experiments were carried out by using self-made device and laboratory instruments.The measurement results are highly correlated,which proves the reliability and rationality of the device.(2)In view of the shortcomings of appearance method,such as relying on subjective evaluation,difference with actual wrinkle,this papper made 35 fit pants,and selected a woman with middle body shape to wear them and carried out the wrinkle experiment,built a shooting device to collect the image of the wrinkle part of the pants changing with the time series,set up the dataset.The expert evaluation method is used to grade the wrinkled image,and the K-means clustering algorithm is used to correct the rating results.This paper constructs a multi-scale convolution neural network model,which adopts parallel structure and uses convolution kernels of different sizes to extract multi-dimensional features of images and fuse them,which can improve the accuracy of the model and reduce the computational complexity.The objective evaluation accuracy of the model is 92.69%,which is better than BP and CNN neural network.(3)The dynamic change process of wrinkle recovery angle,wrinkle image texture and their correlation are analyzed.The results are as follows:(a)The wrinkle recovery angle?₁has an important effect on the change of the recovery Angle in unit time after1s.The better wrinkle resistance of the fabric,the smaller the?₁,and the smaller the change of the recovery angle after 1s.(b)When testing fabric wrinkle recovery angle,if the difference of?₁between the two fabrics is greater than 26°,the strength of the wrinkle resistance of the two can be directly judged by the size of?₁;if the difference of?₁is less than 26°,it needs to be judged according to?₃₀₀.(c)Both the GLCM and the Tamura texture feature parameters can better extract the wrinkle texture features of the image.(d)The texture feature of the wrinkle image is highly correlated with the wrinkle recovery angle,and has a higher correlation with?₁.The larger the?_i(the worse the wrinkle resistance of the fabric),the more uneven,messy,deep grooves,low texture similarity,uneven grayscale distribution of the image,and obvious linear structure.(e)The fabric with better wrinkle resistance has few wrinkles and have a faster recovery speed.Based on computer vision technology,this paper explores wrinkle recovery angle measurement method and objective evaluation method of clothing smoothness.The proposed device and angle measurement algorithm can be used to test and calculate fabric multi-directional recovery angle,and the proposed multi-scale convolution network model can be used for objective evaluation of clothing smoothness,providing reference for the selection of clothing fabrics.