An unsupervised learning-based fabric online automatic matching method

Abstract

The application discloses an online automatic flower arrangement method of fabric based on unsupervised learning, which comprises the following steps: S1, using a flower-shaped electronic manuscript image of a flower-shaped non-deformation fabric as a template image, or collecting a full-width image of a static or running flower-shaped non-deformation fabric to perform image processing and obtaining a flower-shaped region image as the template image; S2, collecting a full-width image of a running fabric to be detected to perform image processing and obtaining a flower-shaped region image as a detected image; S3, a central processing unit extracts image feature points in S1 and image feature points in S2, performs matching, obtains matching data, and the image feature points are composed of two parts of key points and descriptors; S4, the central processing unit calculates a flower deformation amount according to the matching data; and S5, correcting according to the flower deformation amount. The application can automatically and accurately calculate flower inclination and flower bending deviation for the flower inclination or bending of the fabric with no repeated patterns, and automatically and accurately correct the fabric flower inclination and bending through machine vision.

Description

TECHNICAL FIELD

[0001] The present application relates to a fabric pattern straightening method, in particular to an online automatic fabric pattern straightening method based on unsupervised learning, and belongs to the technical field of textile printing and dyeing process. BACKGROUND

[0002] The fabric with a pattern of checks, stripes and prints, etc. in the post-finishing process such as washing, drying, tentering and preshrinking, etc. is in a continuous traction state and is affected by various mechanical movements and production operations, and the tension of each guide roller is uneven, etc. The fabric shows distortion of the pattern of checks, stripes and prints, etc. such as pattern tilting, bending and S-bending, etc. The distortion of the pattern of checks, stripes and prints, etc. will affect the processing quality of the subsequent process, and needs to be straightened.

[0003] The existing fabric pattern straightening method based on machine vision usually adopts an industrial camera to collect the motion image of the fabric, adopts a digital image feature extraction technology to extract the feature information of the fabric image, and relies on the fitting of several sampling points to obtain the deformation amount of the pattern of the fabric to be detected, and then the correction device of the weft straightener is used to correct the deformation of the pattern of the fabric, which has good adaptability and detection accuracy. However, for the fabric without repeated pattern of the pattern, that is, the number of pattern repetition in the fabric width direction is 1, or the fabric has only one pattern in the width direction, the existing pattern straightening method only relies on the fitting of several sampling points, and the straightening result often has a certain deviation, which cannot be solved by the existing pattern straightening method. Moreover, the existing pattern straightening method has a large labor intensity and a large operation difficulty. SUMMARY

[0004] The technical problem to be solved by the present application is to provide an online automatic fabric pattern straightening method based on unsupervised learning, which can accurately calculate the deformation amount of the pattern of the fabric with a pattern of checks, stripes and prints, etc., especially the pattern tilting or bending of the fabric without repeated pattern of the pattern, so as to accurately correct the tilting and bending of the pattern of the fabric.

[0005] To solve the above technical problem, the present application adopts an online automatic fabric pattern straightening method based on unsupervised learning, comprising the following steps:

[0006] S1: The central processing unit adopts the electronic image of the pattern of the fabric without deformation as a template image, or the central processing unit collects the full-width image of the static or running fabric without deformation, processes the full-width image of the fabric without deformation to obtain the pattern area image of the fabric without deformation, and takes the pattern area image of the fabric without deformation as a template image;

[0007] S2: The central processing unit collects a full-width image of the running fabric to be detected, and performs image processing on the full-width image of the fabric to be detected to obtain a pattern region image of the fabric to be detected, wherein the width of the pattern region image of the fabric to be detected is C W , and the height is C H , and the pattern region image of the fabric to be detected is taken as a detection image;

[0008] S3: The central processing unit extracts image feature points of the template image in step S1 and image feature points of the detection image in step S2, and matches the image feature points of the two to obtain matching data, wherein the image feature points are composed of key points and descriptors.

[0009] S4: The central processing unit calculates the pattern deformation of the fabric to be detected according to the matching data.

[0010] S5: According to the pattern deformation of the fabric to be detected, the pattern deformation of the fabric to be detected is corrected through the correcting device of the weft straightener.

[0011] As a preferred embodiment of the present application, in step S1, the central processing unit obtains the pattern electronic manuscript image through local copying or network transmission; the central processing unit collects a full-width image of the running pattern non-deformed fabric through an industrial camera, and performs image processing on the full-width image of the pattern non-deformed fabric, including performing filter denoising processing on the full-width image of the pattern non-deformed fabric, and performing edge detection or brightness threshold segmentation processing on the full-width image after filter denoising processing to obtain a pattern region image of the pattern non-deformed fabric.

[0012] As a preferred embodiment of the present application, the central processing unit performs filter denoising processing on the full-width image of the pattern non-deformed fabric through a filter, wherein the filter includes a mean filter, a median filter, a low-pass filter, a Gaussian filter in a spatial filter, or a wavelet transform filter, a Fourier transform filter, a cosine transform filter in a frequency filter, or a morphological filter performing denoising through morphological operations such as expansion, corrosion or opening-closing operation; the central processing unit performs edge detection on the full-width image after filter denoising processing through a sobel algorithm or a Roberts algorithm or a Prewitt algorithm or a Laplacian algorithm or a Canny algorithm to obtain a pattern region image of the pattern non-deformed fabric; or the central processing unit performs brightness threshold segmentation processing on the full-width image after filter denoising processing through a fixed threshold segmentation method or a threshold segmentation method based on a gray histogram or an adaptive threshold segmentation method or a maximum entropy threshold segmentation method or a maximum inter-class variance threshold segmentation method to obtain a pattern region image of the pattern non-deformed fabric.

[0013] As a preferred embodiment of the present application, in step S2, the central processing unit collects a full-width image of the running fabric to be detected by the industrial camera, and image processing of the full-width image of the fabric to be detected includes filtering and denoising processing of the full-width image of the fabric to be detected, edge detection or brightness threshold segmentation processing of the full-width image after filtering and denoising processing, and obtaining a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H .

[0014] As a preferred embodiment of the present application, the central processing unit performs filtering and denoising processing on the full-width image of the fabric to be detected by a filter, wherein the filter includes a mean filter, a median filter, a low-pass filter, and a Gaussian filter in a spatial filter, or the filter includes a wavelet transform filter, a Fourier transform filter, and a cosine transform filter in a frequency filter, or the filter includes a morphological filter for denoising by morphological operations such as dilation, erosion, or opening and closing operations; the central processing unit performs edge detection on the full-width image after filtering and denoising processing by a sobel algorithm or a Roberts algorithm or a Prewitt algorithm or a Laplacian algorithm or a Canny algorithm, and obtains a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H ; or the central processing unit performs brightness threshold segmentation processing on the full-width image after filtering and denoising processing by a fixed threshold segmentation method or a gray histogram-based threshold segmentation method or an adaptive threshold segmentation method or a maximum entropy threshold segmentation method or a maximum inter-class variance threshold segmentation method, and obtains a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H .

[0015] As a preferred embodiment of the present application, in step S3, the image feature points include corners, inflection points, and edges of the image; the central processing unit extracts image feature points of the template image in step S1 and image feature points of the image to be detected in step S2 by a SuperPoint or D2-Net or Aslfeat or R2D2 deep learning network model; and the central processing unit matches the image feature points of the template image with the image feature points of the image to be detected by a Neural-Guided Ransac or S2dnet or SuperGlue or Acne deep learning network model, obtains matching data, and excludes false matches.

[0016] As a preferred embodiment of the present application, the matching data is a left deviation amount ΔL, a middle side deviation amount ΔM and a right deviation amount ΔR of the pattern region image of the fabric to be detected, wherein the left side refers to a region of one side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, the right side refers to a region of the other side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, and the middle side refers to a region of the center of the pattern region image of the fabric to be detected occupying 1 / 8 width of the fabric width on the left and right sides; in step S4, the pattern deformation amount includes a pattern skew deviation amount and a pattern bow deviation amount, the pattern skew deviation amount Skew = (ΔR-ΔL) / C W , and the pattern bow deviation amount Bow = (ΔM-((ΔR-ΔL) / 2)) / (C W / 2).

[0017] As a preferred embodiment of the present application, in step S3, the image feature points include corner points, inflection points and edges of the image; the central processing unit extracts the image feature points of the template image in step S1 and the image feature points of the image to be detected in step S2 through the Patch2Pix deep learning network model, matches the image feature points of the two images, obtains the matching data, and excludes false matches.

[0018] As a preferred embodiment of the present application, the matching data is a left deviation amount ΔL, a middle side deviation amount ΔM and a right deviation amount ΔR of the pattern region image of the fabric to be detected, wherein the left side refers to a region of one side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, the right side refers to a region of the other side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, and the middle side refers to a region of the center of the pattern region image of the fabric to be detected occupying 1 / 8 width of the fabric width on the left and right sides; in step S4, the pattern deformation amount includes a pattern skew deviation amount and a pattern bow deviation amount, the pattern skew deviation amount Skew = (ΔR-ΔL) / C W , and the pattern bow deviation amount Bow = (ΔM-((ΔR-ΔL) / 2)) / (C W / 2).

[0019] As a preferred embodiment of the present application, the central processing unit includes a terminal device or a server, the terminal device includes a digital controller or an embedded control system or an industrial computer with a human-machine interface, and the server includes a cloud server.

[0020] After the above pattern matching method is used, the present application has the following beneficial effects:

[0021] The present application can automatically and accurately calculate the flower skew deviation and flower bending deviation for the fabric with flower pattern, especially for the flower skew or bending of the fabric with flower pattern without repetition, and automatically and accurately correct the flower skew and bending of the fabric through machine vision, has strong operability, can greatly improve the qualified rate of the fabric, solves the problem of flower distortion of the existing fabric in the finishing process such as washing, drying, tentering and preshrinking, and greatly reduces the operation difficulty.

[0022] The present application has good whole flower effect and high quality, can accurately correct the distortion such as flower skew and bending, and better meets the whole flower requirement of the fabric.

[0023] The present application greatly reduces the skill requirement and labor intensity of the operator, and improves the production efficiency.

[0024] The present application guarantees the genuine product rate of the printed product, and brings greater economic benefits to the enterprise. Embodiment

[0025] The present application will be further described below in combination with examples.

[0026] A fabric online automatic whole flower method based on unsupervised learning, preferably using an existing weft straightener as an automatic whole flower device, such as the weft straighteners disclosed in Chinese utility model patents with patent numbers 202121938486.5, 202121174391.0 and 202121175396.5, comprising the following steps:

[0027] S1: The central processing unit uses the flower electronic draft image of the flower pattern non-deformation fabric as a template image, or the central processing unit collects the full-width image of the static or running flower pattern non-deformation fabric, processes the full-width image of the flower pattern non-deformation fabric to obtain the flower region image of the flower pattern non-deformation fabric, and uses the flower region image of the flower pattern non-deformation fabric as a template image; in this step, the flower pattern non-deformation fabric can be the flower pattern of the fabric manually adjusted to the non-deformation state by the operator, can be the flower pattern slightly deformed fabric meeting the production requirement, or can be the fabric without deformation of the flower pattern itself;

[0028] S2: The central processing unit collects the full-width image of the running fabric to be detected, processes the full-width image of the fabric to be detected to obtain the flower region image of the fabric to be detected, the width of the flower region image of the fabric to be detected is C W , and the height is C H , and uses the flower region image of the fabric to be detected as a detection image;

[0029] S3: The central processing unit extracts the image feature points of the template image in step S1 and the image feature points of the image to be detected in step S2, and matches the image feature points of the two, to obtain matching data, wherein the image feature points are composed of two parts of key points and descriptors;

[0030] S4: The central processing unit calculates the pattern deformation of the fabric to be detected according to the matching data;

[0031] S5: According to the pattern deformation of the fabric to be detected, the pattern deformation of the fabric to be detected is corrected by the correcting device of the weft straightener.

[0032] As a preferred embodiment of the present application, in step S1, the central processing unit obtains the pattern electronic manuscript image by local copying or network transmission; the central processing unit collects the full-width image of the pattern non-deformation fabric in static or running state by an industrial camera, and performs image processing on the full-width image of the pattern non-deformation fabric, including filtering and denoising processing, edge detection or brightness threshold segmentation processing, to obtain the pattern region image of the pattern non-deformation fabric. The industrial camera includes a linear array camera or a plane array camera, and the industrial camera transmits the collected full-width image to the central processing unit for image processing.

[0033] As a preferred embodiment of the present application, the central processing unit performs filtering and denoising processing on the full-width image of the pattern non-deformation fabric by a filter, which includes a mean filter, a median filter, a low-pass filter, a Gaussian filter in a spatial filter, or a wavelet transform filter, a Fourier transform filter, a cosine transform filter in a frequency filter, or a morphological filter for denoising by morphological operations such as expansion, corrosion or opening-closing operation; the central processing unit performs edge detection on the full-width image after filtering and denoising processing by a sobel algorithm or a Roberts algorithm or a Prewitt algorithm or a Laplacian algorithm or a Canny algorithm, to obtain the pattern region image of the pattern non-deformation fabric; or the central processing unit performs brightness threshold segmentation processing on the full-width image after filtering and denoising processing by a fixed threshold segmentation method or a threshold segmentation method based on a gray histogram or an adaptive threshold segmentation method or a maximum entropy threshold segmentation method or a maximum inter-class variance threshold segmentation method, to obtain the pattern region image of the pattern non-deformation fabric.

[0034] As a preferred embodiment of the present application, in step S2, the central processing unit collects a full-width image of the running fabric to be detected through the industrial camera, and image processing of the full-width image of the fabric to be detected includes filtering and denoising processing of the full-width image of the fabric to be detected, edge detection or brightness threshold segmentation processing of the full-width image after filtering and denoising processing, and obtaining a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H .

[0035] As a preferred embodiment of the present application, the central processing unit performs filtering and denoising processing on the full-width image of the fabric to be detected through a filter, wherein the filter includes a mean filter, a median filter, a low-pass filter, and a Gaussian filter in a spatial filter, or the filter includes a wavelet transform filter, a Fourier transform filter, and a cosine transform filter in a frequency filter, or the filter includes a morphological filter for denoising through morphological operations such as dilation, erosion, or opening and closing operations; the central processing unit performs edge detection on the full-width image after filtering and denoising processing through a sobel algorithm or a Roberts algorithm or a Prewitt algorithm or a Laplacian algorithm or a Canny algorithm, and obtains a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H ; or the central processing unit performs brightness threshold segmentation processing on the full-width image after filtering and denoising processing through a fixed threshold segmentation method or a gray histogram-based threshold segmentation method or an adaptive threshold segmentation method or a maximum entropy threshold segmentation method or a maximum inter-class variance threshold segmentation method, and obtains a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H .

[0036] As a preferred embodiment of the present application, in step S3, the image feature points include corners, inflection points, and edges of the image; the central processing unit extracts image feature points of the template image in step S1 and image feature points of the image to be detected in step S2 through a known deep learning network model such as SuperPoint or D2-Net or Aslfeat or R2D2; and the central processing unit matches the image feature points of the template image with the image feature points of the image to be detected through a known deep learning network model such as Neural-Guided Ransac or S2dnet or SuperGlue or Acne, obtains matching data, and excludes false matches.

[0037] As a preferred embodiment of the present application, the matching data is a left deviation amount ΔL, a middle side deviation amount ΔM, and a right deviation amount ΔR of the pattern region image of the fabric to be detected, wherein the left side refers to a region of one side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, the right side refers to a region of the other side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, and the middle side refers to a region of the center of the pattern region image of the fabric to be detected occupying 1 / 8 width of the fabric width on both sides, and in step S4, the pattern deformation amount includes a pattern skew deviation amount and a pattern bow deviation amount, the pattern skew deviation amount Skew = (ΔR-ΔL) / C W , and the pattern bow deviation amount Bow = (ΔM-((ΔR-ΔL) / 2)) / (C W / 2).

[0038] As another preferred embodiment of the present application, in step S3, the image feature points include corner points, inflection points, and edges of the image; the central processing unit extracts the image feature points of the template image in step S1 and the image feature points of the image to be detected in step S2 through a known Patch2Pix deep learning network model, and directly matches the image feature points of the two at the same time to obtain the matching data and exclude false matches.

[0039] As a preferred embodiment of the present application, the matching data is a left deviation amount ΔL, a middle side deviation amount ΔM, and a right deviation amount ΔR of the pattern region image of the fabric to be detected, wherein the left side refers to a region of one side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, the right side refers to a region of the other side of the pattern region image of the fabric to be detected occupying 1 / 4 width of the fabric width, and the middle side refers to a region of the center of the pattern region image of the fabric to be detected occupying 1 / 8 width of the fabric width on both sides, and in step S4, the pattern deformation amount includes a pattern skew deviation amount and a pattern bow deviation amount, the pattern skew deviation amount Skew = (ΔR-ΔL) / C W , and the pattern bow deviation amount Bow = (ΔM-((ΔR-ΔL) / 2)) / (C W / 2).

[0040] As a preferred embodiment of the present application, the central processing unit includes a terminal device or a server, the terminal device includes a digital controller such as a DDC digital controller or an embedded control system or an industrial computer with a human-machine interface, and the server includes a cloud server. The industrial camera is connected to the central processing unit, which can be a central processing unit of a let-off machine or an independently arranged central processing unit.

[0041] During the rectification, the central processing unit calculates the pattern deformation of the fabric to be detected according to the matching data, and controls the action of the rectification device of the whole-weft machine to complete the automatic pattern rectification of the fabric.

[0042] Through tests, the present application can automatically and accurately calculate the pattern bending deviation and pattern oblique deviation for the pattern tilting or bending of the fabric without repeated patterns, and has good pattern rectification effect and high quality, solves the pattern deformation problem of the existing fabric during the finishing treatment processes such as washing, drying, tentering and preshrinking, and guarantees the product quality of the printed products, and achieves good results.

Claims

1. An unsupervised learning based online automatic pattern matching method for fabric, characterized in that, The method comprises the following steps: S1: The central processing unit takes the pattern electronic manuscript image of the pattern non-deformation fabric as a template image, or the central processing unit collects the full-width image of the static or running pattern non-deformation fabric, and carries out image processing on the full-width image of the pattern non-deformation fabric to obtain the pattern region image of the pattern non-deformation fabric, and takes the pattern region image of the pattern non-deformation fabric as a template image; S2: The central processing unit collects the full-width image of the running fabric to be detected, performs image processing on the full-width image of the fabric to be detected, and obtains a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H , and the pattern area image of the fabric to be detected is taken as a detection image; S3: The central processing unit extracts the image feature points of the template image in step S1 and the image feature points of the image to be detected in step S2, and matches the image feature points of the two to obtain matching data, wherein the image feature points are composed of key points and descriptors; S4: The central processing unit calculates the pattern deformation amount of the fabric to be detected according to the matching data; S5: According to the pattern deformation amount of the fabric to be detected, the pattern deformation of the fabric to be detected is corrected through the correcting device of the weft straightener. In step S3, the image feature points include corner points, inflection points, and edges of the image; the central processing unit extracts image feature points of the template image in step S1 and image feature points of the image to be detected in step S2 using a SuperPoint, D2-Net, Aslfeat, or R2D2 deep learning network model; the central processing unit uses Neural-Guided... Ransac, S2dnet, SuperGlue, or Acne deep learning network models match the image feature points of the template image with the image feature points of the image to be detected to obtain matching data and eliminate incorrect matches. The matching data consists of the left-side deviation ΔL, the middle-side deviation ΔM, and the right-side deviation ΔR of the pattern area image of the fabric to be detected. The left side refers to the area occupying 1 / 4 of the width of the pattern area image on one side; the right side refers to the area occupying 1 / 4 of the width of the pattern area image on the other side; and the middle side refers to the area occupying 1 / 8 of the width of the pattern area image on both sides of the center. In step S4, the pattern deformation includes the pattern slant deviation and the pattern bend deviation. The pattern slant deviation Skew = (ΔR - ΔL) / C W The deviation of the flower bend, Bow = (ΔM - ((ΔR - ΔL) / 2)) / (C) W / 2); Alternatively, in step S3, the image feature points include the corner points, inflection points, and edges of the image; the central processing unit extracts the image feature points of the template image in step S1 and the image feature points of the image to be detected in step S2 using the Patch2Pix deep learning network model, and matches the image feature points of the two to obtain matching data and eliminate incorrect matches; the matching data is the left deviation of the pattern area image of the fabric to be detected as ΔL, the middle deviation as ΔM, and the right deviation as ΔR, where the left side refers to the area on one side of the pattern area image of the fabric to be detected, which occupies 1 / 4 of the width of the fabric width; the right side refers to the area on the other side of the pattern area image of the fabric to be detected, which occupies 1 / 4 of the width of the fabric width; and the middle side refers to the area on both sides of the center of the pattern area image of the fabric to be detected, which occupies 1 / 8 of the width of the fabric width. In step S4, the pattern deformation includes the pattern slant deviation and the pattern bend deviation, where the pattern slant deviation Skew = (ΔR - ΔL) / C W The deviation of the flower bend, Bow = (ΔM - ((ΔR - ΔL) / 2)) / (C) W / 2).

2. The unsupervised learning based fabric online automatic matching method according to claim 1, characterized in that: In step S1, the central processing unit obtains the pattern electronic manuscript image through local copying or network transmission; the central processing unit collects the full-width image of the static or running pattern non-deformation fabric, and carries out image processing on the full-width image of the pattern non-deformation fabric, including filtering and denoising processing on the full-width image of the pattern non-deformation fabric, and carrying out edge detection or brightness threshold segmentation processing on the full-width image after filtering and denoising processing to obtain the pattern region image of the pattern non-deformation fabric.

3. The unsupervised learning based fabric online automatic filling method according to claim 2, characterized in that: The central processing unit carries out filtering and denoising processing on the full-width image of the pattern non-deformation fabric through a filter, wherein the filter includes a mean filter, a median filter, a low-pass filter, and a Gaussian filter in a spatial filter, or the filter includes a wavelet transform filter, a Fourier transform filter, and a cosine transform filter in a frequency filter, or the filter includes a morphological filter for denoising through morphological operations such as expansion, corrosion, or opening and closing operations; the central processing unit carries out edge detection on the full-width image after filtering and denoising processing through a sobel algorithm or a Roberts algorithm or a Prewitt algorithm or a Laplacian algorithm or a Canny algorithm to obtain the pattern region image of the pattern non-deformation fabric; or the central processing unit carries out brightness threshold segmentation processing on the full-width image after filtering and denoising processing through a fixed threshold segmentation method or a gray histogram-based threshold segmentation method or an adaptive threshold segmentation method or a maximum entropy threshold segmentation method or a maximum inter-class variance threshold segmentation method to obtain the pattern region image of the pattern non-deformation fabric.

4. The unsupervised learning based fabric online automatic matching method according to claim 1, characterized in that: In step S2, the central processing unit collects a full-width image of the running fabric to be detected by the industrial camera, and performs image processing on the full-width image of the fabric to be detected, including filtering and denoising processing on the full-width image of the fabric to be detected, edge detection or brightness threshold segmentation processing on the full-width image after filtering and denoising processing, to obtain a pattern area image of the fabric to be detected, wherein the width of the pattern area image of the fabric to be detected is C W , and the height is C H .

5. The unsupervised learning based fabric online automatic matching method according to claim 4, characterized in that: The central processing unit filters and denoises the full-width image of the fabric to be detected by a filter, which includes a mean filter, a median filter, a low-pass filter, a Gaussian filter in a spatial domain filter, or which includes a wavelet transform filter, a Fourier transform filter, a cosine transform filter in a frequency domain filter, or which includes a morphological filter for denoising by morphological operation of expansion, corrosion or opening-closing operation; the central processing unit performs edge detection on the full-width image after filtering and denoising by a sobel algorithm or a Roberts algorithm or a Prewitt algorithm or a Laplacian algorithm or a Canny algorithm, to obtain a pattern area image of the fabric to be detected, the width of the pattern area image of the fabric to be detected being C W , and the height being C H ; or the central processing unit performs luminance threshold segmentation processing on the full-width image after filtering and denoising by a fixed threshold segmentation method or a gray histogram-based threshold segmentation method or an adaptive threshold segmentation method or a maximum entropy threshold segmentation method or a maximum inter-class variance threshold segmentation method, to obtain a pattern area image of the fabric to be detected, the width of the pattern area image of the fabric to be detected being C W , and the height being C H .

6. The unsupervised learning based fabric online automatic filling method according to claim 1, characterized in that: The central processing unit includes a terminal device or a server, the terminal device includes a digital controller or an embedded control system or an industrial computer with a human-machine interface, and the server includes a cloud server.