Fabric size change rate intelligent visual measurement device and method

Abstract

The invention discloses an intelligent visual measurement device and method for fabric size change rate, the device is used for acquiring fabric images, the method comprises the following steps: S1, fabric image acquisition: respectively acquiring a pre-processing fabric image and a post-processing fabric image by using an image acquisition device; s2, preprocessing the fabric image; s3, mark point tracking and identification: a multi-feature fusion algorithm is utilized, color, shape and texture features are combined, a twin neural network architecture is fused, and mark point feature model training and detection are completed; marking point matching is carried out based on the relative position relation of the marking points, so that tracking and identification of the marking points are completed; and S4, size change rate calculation: performing Euclidean distance calculation on the obtained corresponding mark points before and after processing, and calculating the size change rate according to the initial distance of the mark points before processing and the distance of the mark points after processing. According to the invention, a machine vision technology is utilized, a rapid identification and positioning method is constructed for fabric image differences before and after processing, and the fabric size change rate detection efficiency is improved.

Description

Technical Field

[0001] The invention belongs to the field of fabric size change detection, and in particular relates to an intelligent visual measurement device and method for fabric size change rate. Background Art

[0002] Due to its interwoven or woven structure, flexible textile fabrics are affected by multiple factors such as mechanical external forces and detergents, causing the fabric shape to change and the size to shrink or expand compared to the initial state. Fabric shrinkage detection is a must-check item before garment processing, and it is important for evaluating the comfort, appearance and durability of clothing. At present, the fabric size change rate detection mainly uses manual marking and manual measurement to calculate the degree of fabric change before and after washing. The speed is slow and there are large differences in tests between different people. There is an urgent need for a fast, objective and intelligent detection method that can replace manual labor.

[0003] Among the patents that have been published, the invention patent with publication number CN117330349B discloses a shrinkage sampling device and method for textile production, which realizes the mechanized cutting of fabrics and improves the efficiency of sampling, but does not involve the measurement method of dimensional change rate. The appearance patent with publication number CN308327532S discloses a shrinkage tester. The utility model patent with publication number CN221039052U also discloses a high-precision fabric shrinkage analysis device, including a box, a marking component and a dry ironing detection component, which can realize direct marking of fabrics of the same experimental size. The utility model patent with publication number CN220399290U discloses an automatic fabric shrinkage measuring instrument, and a negative pressure fan, a leak-proof net, etc. are set on the fabric flat base to improve the fabric flattening effect. The invention patent with publication number CN118134418A also discloses a product intelligent mold splitting method and its product intelligent mold splitting system, and explains the product shrinkage module. The invention patent with publication number CN114739999A proposes a high-precision fabric shrinkage test method, which uses a square screen to fish out the washed fabric and calculates the size change rate only based on the fabric shape before and after shrinkage. None of the above patents propose an effective solution for the automatic determination of the size change rate. The invention patent with publication number CN117871832A proposes a textile shrinkage qualification rate detection method based on dual data model processing, which uses an empirical model to determine whether it is qualified rather than precise measurement. Similarly, the invention patent with publication number CN116008513A proposes a knitted fabric shrinkage prediction method, which uses a linear equation of the size change rate to predict the shrinkage rate of different fabrics. The invention patent with publication number CN114486898A proposes a fabric shrinkage test method and system for complex colors and textures, which realizes the positioning of the marking point through conventional image processing methods such as binarization and threshold segmentation, and uses specific luminous or fluorescent dye markings, which are sensitive to lighting conditions.

[0004] The methods described in the above-mentioned published patents and related papers fail to specifically explain how to automatically track, identify and locate fabric marker points, and fail to improve current detection. They mainly explain some basic steps for detecting dimensional change rates, and do not provide practical solutions to core technical problems. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent visual measurement device and method for fabric size change rate. In view of the defects in the prior art, machine vision technology is used to construct a method for rapid identification and positioning based on the difference in fabric images before and after processing, thereby improving the efficiency of fabric size change rate detection.

[0006] In order to solve the above technical problems, the following technical solutions are adopted:

[0007] An intelligent visual determination method for fabric size change rate, characterized by comprising the following steps:

[0008] S1: Fabric image acquisition: using an image acquisition device to respectively acquire the fabric image before and after processing;

[0009] S2: fabric image preprocessing: preprocessing the acquired fabric image before processing and the acquired fabric image after processing;

[0010] S3: Marker point tracking and recognition: Use multi-feature fusion algorithm, combine color, shape and texture features, integrate twin neural network architecture, complete marker point feature model training and detection; perform marker point matching based on the relative position relationship of marker points, thereby completing marker point tracking and recognition;

[0011] S4: Calculation of size change rate: Perform Euclidean distance calculation on the corresponding marking points before and after the processing, and calculate the size change rate based on the initial distance of the marking points before processing and the distance after processing.

[0012] After optimization, in step S1, three fabric samples of equal size are cut from different parts of the same fabric; marking points are marked on the surface of the untreated fabric sample, and the fabric image before treatment is collected by an image acquisition device; the fabric sample is washed or ironed, and after drying, it is balanced in a constant temperature and humidity environment, and the fabric image after treatment is collected by an image acquisition device.

[0013] After optimization, the preprocessing of S2 includes image enhancement, illumination balance and noise reduction processing, and the obtained fabric image before processing and the fabric image after processing are first subjected to image enhancement processing and illumination balance processing, and then subjected to noise reduction processing.

[0014] After optimization, the S3, marker point detection: using a multi-feature fusion algorithm, combined with color, shape and texture features, to complete high-precision marker point detection; including: a, color feature: preliminary detection of the marker point area through color threshold segmentation; b, shape feature: using Hough transform to detect marker points of circular or other specific shapes; c, texture feature: using local binary mode for texture analysis to confirm the marker point.

[0015] After optimization, the step S3, marking point matching: based on the shape context method, the shape context descriptor of the marking point is extracted; the matching verification is performed using the RANSAC algorithm to eliminate false matches and ensure the one-to-one correspondence of the marking points.

[0016] After optimization, the step S3, marker point tracking: use a twin neural network to perform real-time tracking of the marker points, the twin neural network consists of two convolutional neural networks with shared weights, inputs the template of the marker points and the cropped part of the new frame image, and outputs the marker point position with the highest matching degree.

[0017] After optimization, the marker point tracking includes the following specific steps:

[0018] a. Feature fusion: Integrate the visual saliency algorithm into the process of marker detection and tracking. Use the saliency detection algorithm based on spectrum residual to perform Fourier transform on the input image to obtain the spectrum map. Calculate the logarithmic amplitude spectrum of the spectrum map and smooth it through a Gaussian filter. The calculation formula is as follows:

[0019] R(u,v)=log(|F(u,v|)-log(|G(u,v)|)

[0020] Among them, R(u, v) is the spectrum residual, and G(u, v) is the smoothed amplitude spectrum;

[0021] The spectrum residual is converted into a saliency map through inverse Fourier transform; the saliency map is superimposed with the tracking result of the twin neural network;

[0022] b. Model training: Collect and annotate a large number of image pairs containing marked points. Each pair of images contains the fabric images before and after processing, as well as the location information of the marked points. Use a convolutional neural network with shared weights to extract image features. The feature vectors are used to calculate the similarity through a distance metric function. The loss function designed uses a contrast loss function, and the formula is as follows:

[0023]

[0024] Where b is the Boolean value of whether it is a marked point, D is the Euclidean distance between two feature vectors, and m is the set threshold;

[0025] c. Real-time tracking: Input the current image and the template of the marker point, use the twin neural network to extract the feature vectors of the current image and the feature point template; on this basis, calculate the similarity between the feature vectors of each possible position in the image and the feature vector of the template; use the Euclidean distance as the similarity metric, and finally select the position with the highest similarity as the new position of the marker point;

[0026] d. Template update strategy: After each tracking, the new position of the marker is updated as the template to maintain the robustness of the tracking; the sliding window strategy is used to update the template and retain some historical information;

[0027] e. Train the twin neural network and dynamically adjust the confidence of the network output position according to the strength of the saliency map; input the template of the marker point and the new frame image into the twin neural network and output the marker point position with the highest matching degree.

[0028] After optimization, the step S4 comprises:

[0029] (1) Euclidean distance calculation: Euclidean distance calculation is performed on the corresponding marked points before and after washing and care;

[0030] (2) Calculation of size change rate: Calculate the size change rate based on the initial distance before and after processing of the marking point. The calculation formula is:

[0031]

[0032] Among them, ΔL is the change in the distance between the marking points before and after processing, L 0 is the initial distance before processing;

[0033] (3) Calculation of mean and standard deviation: Calculate the mean and standard deviation of the three measurement results.

[0034] After optimization, the method also includes S6: report generation: visually displaying the position of the marking point and the size change rate on the image, and using a chart to display the statistical analysis results of the size change rate; automatically generating a test report, including images before and after fabric processing, the position of the marking point, and the size change rate analysis results; supporting printing and exporting of the report.

[0035] An intelligent visual measuring device for fabric dimensional change rate includes an image acquisition device, characterized in that: the image acquisition device includes a fabric stage, a camera bracket, a camera, a camera connecting line and a light source, the fabric stage is used to place fabric samples, the camera bracket is used to install the camera, the camera adopts a high-resolution digital camera of more than 5 million, and the camera has automatic focus and exposure adjustment functions; the camera connecting line connects the camera to a computer; the light source adopts an LED light source, which is installed on the camera bracket.

[0036] The above technical solution has the following beneficial effects:

[0037] (1) High-precision marker detection and tracking

[0038] The present invention adopts twin neural networks combined with visual saliency algorithms to achieve high-precision marker point detection and tracking. The twin neural network can accurately identify and locate marker points on the fabric through multi-feature fusion detection, and uses real-time tracking and template update strategies to ensure the continuity and stability of the marker points during the processing. The introduction of the visual saliency algorithm further improves the robustness of marker point detection, allowing the system to work stably under complex background and lighting conditions. This high-precision detection and tracking method effectively reduces errors and greatly improves the accuracy of fabric size change rate detection.

[0039] (2) Automatic dimensional change rate calculation

[0040] The present invention realizes the automatic calculation of fabric size change rate. Through the image acquisition device and algorithm, the marking points before and after the fabric processing are automatically detected and identified, and the Euclidean distance change of the marking points is calculated. Combined with the calibration and conversion of pixels and actual length units, the system can efficiently and accurately calculate the size change rate of the fabric. The automated process not only improves the detection efficiency and saves labor costs, but also ensures the consistency and reliability of the results. The final generated test report contains detailed size change rate analysis results and visual display, which is convenient for archiving and sharing of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with the accompanying drawings:

[0042] Figure 1 is a flow chart of the method of the present invention;

[0043] Figure 2 It is a structural schematic diagram of an image acquisition device;

[0044] Figure 3 It is a twin neural network architecture.

[0045] Attached Figure 2 The numbers in the figure are: fabric stage 1, camera bracket 2, camera 3, light source 4, camera connecting line 5. DETAILED DESCRIPTION

[0046] The present invention aims to provide an intelligent visual measurement device and method for fabric size change rate, which can measure the fabric size change rate before and after washing, drying, ironing and other treatments of textile fabrics. Figure 1 As shown, the technical solution of the present invention is described in detail below in conjunction with specific embodiments:

[0047] S1: Build an image acquisition device: The image acquisition device includes a fabric stage 1, a camera bracket 2, a camera 3, a camera connecting line 5 and a light source 4. Figure 2 As shown:

[0048] (1) Fabric stage 1: used to place fabric samples, using high-precision flat glass, acrylic plate or plate to ensure that the fabric is flat and does not deform during the collection process.

[0049] (2) Camera bracket 2: A metal bracket with adjustable height and angle is used to ensure that the camera 3 is always perpendicular to the fabric surface to avoid viewing angle deviation. The base of the camera bracket 2 is equipped with a shockproof pad to reduce the impact of environmental vibration on image quality.

[0050] (3) Camera 3: Use a high-resolution digital camera 3 with a resolution of 5 megapixels or more to ensure the clarity and detail of the captured images. Camera 3 has automatic focus and exposure adjustment functions to adapt to image acquisition under different lighting conditions.

[0051] (4) Camera connection cable 5: It is a USB data cable or an Ethernet cable. The camera 3 is connected to the computer via the camera connection cable 5 to facilitate real-time transmission and storage of images.

[0052] (5) Light source 4: An LED light source 4 with adjustable brightness is used to provide uniform lighting and reduce the impact of shadows and reflections on image quality. The light source 4 is installed above or to the side of the camera bracket 2 to ensure that the light evenly covers the entire fabric surface.

[0053] S2: Fabric image acquisition: using an image acquisition device to respectively acquire the fabric image before and after processing, specifically including:

[0054] (1) Fabric sample cutting: According to relevant standards or user-defined sample sizes, three fabric samples of equal size are cut from different parts of the same fabric. Ensure that the sample sizes are consistent to reduce the impact of human errors on the test results.

[0055] (2) Setting of marking points: On the flat surface of the untreated fabric sample, use a marker pen to mark the marking points around and in the middle of a rectangular position such as 35 cm × 35 cm, 25 cm × 25 cm, or 20 cm × 20 cm, ensuring that the position of the marking points is clear and fixed. Use an image acquisition device to capture the image of the fabric before treatment and record the initial position of the marking points.

[0056] (3) Fabric sample processing: Wash and iron the fabric samples according to user needs, and ensure that the processing process of each sample is consistent. After processing, the fabric samples need to go through drying steps such as drying or air-drying, and balance in a constant temperature and humidity environment for more than 24 hours.

[0057] (4) Post-processing image acquisition: Use an image acquisition device to capture the processed fabric image to ensure the clarity and consistency of the image.

[0058] S3: Sample image preprocessing: Preprocessing the acquired fabric image before and after processing, specifically including:

[0059] (1) Image enhancement: Use the histogram equalization method to improve the contrast of the image and make the markers more visible. Use the sharpening filter to enhance the image details and improve the edge clarity of the markers.

[0060] (2) Light balance: White balance adjustment is used to ensure uniform image lighting and reduce the impact of lighting changes on marker point detection.

[0061] (3) Noise reduction: Use Gaussian filter or median filter to remove random noise in the image to ensure the accuracy of marker point detection.

[0062] S4: Marker tracking and recognition: Using multi-feature fusion algorithm, combining color, shape and texture features, and fusing the twin neural network architecture (such as Figure 3 As shown in the figure), the marking point feature model training and detection are realized; the marking points are matched based on the relative position relationship of the marking points, so as to realize the tracking and identification of the marking points. The specific steps include:

[0063] (1) Marker point detection: A multi-feature fusion algorithm is used to combine color, shape and texture features to achieve high-precision marker point detection; including: a. Color feature: Preliminary detection of marker point area through color threshold segmentation; b. Shape feature: Use Hough transform to detect marker points of circular or other specific shapes; c. Texture feature: Use local binary pattern for texture analysis to further confirm marker points.

[0064] (2) Marker point matching: Based on the shape context method, the shape context descriptor of the marker point is extracted; the RANSAC algorithm is used for matching verification to eliminate false matches and ensure the one-to-one correspondence of the marker points.

[0065] (3) Marker point tracking: Use a twin neural network to track marker points in real time. The twin neural network consists of two convolutional neural networks with shared weights. It inputs the template of the marker point and the cropped part of the new frame image, and outputs the marker point position with the highest matching degree. The specific steps are as follows:

[0066] a. Feature fusion: Integrate the visual saliency algorithm into the process of marker detection and tracking to improve the detection and tracking accuracy of markers. Use the saliency detection algorithm based on spectrum residual to perform Fourier transform on the input image to obtain the spectrum map; calculate the logarithmic amplitude spectrum of the spectrum map and smooth it through a Gaussian filter. The calculation formula is as follows:

[0067] R(u,v)=log(|F(u,v|)-log(|G(u,v)|)

[0068] Among them, R(u, v) is the spectrum residual, and G(u, v) is the smoothed amplitude spectrum;

[0069] The spectrum residual is converted into a saliency map through inverse Fourier transform; the saliency map is superimposed with the tracking result of the twin neural network;

[0070] b. Model training: Collect and annotate a large number of image pairs containing markers. Each pair of images contains the fabric image before and after processing, as well as the location information of the markers. Use a convolutional neural network with shared weights to extract image features (such as Figure 3 As shown); the feature vectors are similarity calculated through a distance metric function (such as Euclidean distance), where the loss function designed uses a contrast loss function, and the formula is as follows:

[0071]

[0072] Where b is the Boolean value of whether it is a marked point, D is the Euclidean distance between two feature vectors, and m is the set threshold;

[0073] c. Real-time tracking: Input the current image and the template of the marker point, use the twin neural network to extract the feature vectors of the current image and the feature point template; on this basis, calculate the similarity between the feature vectors of each possible position in the image and the feature vector of the template; use the Euclidean distance as the similarity metric, and finally select the position with the highest similarity as the new position of the marker point;

[0074] d. Template update strategy: After each tracking, the new position of the marker is updated to the template to maintain the robustness of the tracking; the sliding window strategy is used to update the template, retaining some historical information to avoid cumulative errors caused by single-frame errors.

[0075] e. Train the twin neural network and dynamically adjust the confidence of the network output position according to the strength of the saliency map; input the template of the marker point and the new frame image into the twin neural network and output the marker point position with the highest matching degree.

[0076] S5: Dimensional change rate calculation:

[0077] (1) Euclidean distance calculation: Euclidean distance calculation is performed on the corresponding marked points before and after washing and care;

[0078] (2) Calculation of size change rate: Calculate the size change rate based on the initial distance before and after processing of the marking point. The calculation formula is:

[0079]

[0080] Among them, ΔL is the change in the distance between the marking points before and after processing, L 0 is the initial distance before processing;

[0081] (3) Calculation of mean and standard deviation: Calculate the mean and standard deviation of the three measurement results.

[0082] S6: Report Generation:

[0083] (1) Visual display: The position and size change rate of the marker points are visualized on the image, and different colors or marks are used to distinguish the marker points before and after processing.

[0084] (2) Statistical analysis: Use charts to display the statistical analysis results of the dimensional change rate, including indicators such as mean value and standard deviation.

[0085] (3) Report generation and export: Automatically generate test reports, including images of fabrics before and after treatment, locations of marking points, and analysis results of dimensional change rates.

[0086] Supports printing and exporting of reports for easy archiving and sharing.

[0087] The above are only specific embodiments of the present invention, but the technical features of the present invention are not limited thereto. Any simple changes, equivalent replacements or modifications made based on the present invention to solve basically the same technical problems and achieve basically the same technical effects are all included in the protection scope of the present invention.

Claims

1. An intelligent visual determination method for fabric size change rate, characterized in that The steps include: S1: Fabric image acquisition: using an image acquisition device to respectively acquire the fabric image before and after processing; S2: fabric image preprocessing: preprocessing the acquired fabric image before processing and the acquired fabric image after processing; S3: Marker point tracking and recognition: Use multi-feature fusion algorithm, combine color, shape and texture features, integrate twin neural network architecture, complete marker point feature model training and detection; perform marker point matching based on the relative position relationship of marker points, thereby completing marker point tracking and recognition; S4: Calculation of size change rate: Perform Euclidean distance calculation on the corresponding marking points before and after the processing, and calculate the size change rate based on the initial distance of the marking points before processing and the distance after processing.

2. The intelligent visual determination method of fabric dimensional change rate according to claim 1, characterized in that: In step S1, three fabric samples of equal size are cut from different parts of the same fabric; marking points are placed on the surface of the untreated fabric sample, and an image acquisition device is used to acquire an image of the fabric before the treatment; the fabric sample is washed or ironed, and after drying, it is balanced in a constant temperature and humidity environment, and an image acquisition device is used to acquire an image of the fabric after the treatment.

3. The intelligent visual determination method of fabric size change rate according to claim 1, characterized in that: The preprocessing of S2 includes image enhancement, illumination balance and noise reduction processing. The obtained fabric image before processing and the obtained fabric image after processing are first subjected to image enhancement processing and illumination balance processing, and then subjected to noise reduction processing.

4. The intelligent visual determination method of fabric size change rate according to claim 1, characterized in that: The S3, marker point detection: uses a multi-feature fusion algorithm to combine color, shape and texture features to complete high-precision marker point detection; including: a. Color feature: preliminary detection of the marker point area through color threshold segmentation; b. Shape feature: uses Hough transform to detect marker points of circular or other specific shapes; c. Texture feature: uses local binary mode to perform texture analysis to confirm the marker point.

5. The intelligent visual determination method of fabric dimensional change rate according to claim 1, characterized in that: The step S3, marking point matching: based on the shape context method, extracting the shape context descriptor of the marking point; using the RANSAC algorithm to perform matching verification, eliminate false matches, and ensure one-to-one correspondence of the marking points.

6. The intelligent visual determination method of fabric dimensional change rate according to claim 1, characterized in that: The step S3, marker point tracking: use a twin neural network to track the marker points in real time, the twin neural network consists of two convolutional neural networks with shared weights, inputs the template of the marker points and the cropped part of the new frame image, and outputs the marker point position with the highest matching degree.

7. The intelligent visual determination method of fabric dimensional change rate according to claim 6, characterized in that: The marker tracking specifically includes the following steps: a. Feature fusion: Integrate the visual saliency algorithm into the process of marker detection and tracking, use the saliency detection algorithm based on spectrum residual, perform Fourier transform on the input image, and obtain the spectrum map; Calculate the logarithmic amplitude spectrum of the spectrum graph and smooth it through a Gaussian filter. The calculation formula is as follows: R(u,v)=log(|F(u,v|)-log(|G(u,v)|) Among them, R(u, v) is the spectrum residual, and G(u, v) is the smoothed amplitude spectrum; The spectrum residual is converted into a saliency map through inverse Fourier transform; the saliency map is superimposed with the tracking result of the twin neural network; b. Model training: Collect and annotate a large number of image pairs containing marked points. Each pair of images contains the fabric images before and after processing, as well as the location information of the marked points. Use a convolutional neural network with shared weights to extract image features. The feature vectors are used to calculate the similarity through a distance metric function. The loss function designed uses a contrast loss function, and the formula is as follows: Where b is the Boolean value of whether it is a marked point, D is the Euclidean distance between two feature vectors, and m is the set threshold; c. Real-time tracking: Input the current image and the template of the marker point, use the twin neural network to extract the feature vectors of the current image and the feature point template; on this basis, calculate the similarity between the feature vectors of each possible position in the image and the feature vector of the template; use the Euclidean distance as the similarity metric, and finally select the position with the highest similarity as the new position of the marker point; d. Template update strategy: After each tracking, the new position of the marker is updated as the template to maintain the robustness of the tracking; the sliding window strategy is used to update the template and retain some historical information; e. Train the twin neural network and dynamically adjust the confidence of the network output position according to the strength of the saliency map; input the template of the marker point and the new frame image into the twin neural network and output the marker point position with the highest matching degree.

8. The intelligent visual measurement method for fabric dimensional change rate according to claim 1, characterized in that: The step S4 comprises: (1) Euclidean distance calculation: Euclidean distance calculation is performed on the corresponding marked points before and after washing and care; (2) Calculation of size change rate: Calculate the size change rate based on the initial distance before and after processing of the marking point. The calculation formula is: Among them, ΔL is the change in the distance between the marking points before and after the treatment, and L0 is the initial distance before the treatment; (3) Calculation of mean and standard deviation: Calculate the mean and standard deviation of the three measurement results.

9. The intelligent visual determination method of fabric dimensional change rate according to claim 1, characterized in that: The method further includes S6: report generation: visually displaying the position of the marking point and the size change rate on the image, and using a chart to display the statistical analysis results of the size change rate; automatically generating a test report, including the image before and after the fabric is processed, the position of the marking point, and the size change rate analysis results; Supports printing and exporting of reports.

10. An intelligent visual measuring device for fabric size change rate, comprising an image acquisition device, characterized in that: The image acquisition device includes a fabric stage, a camera bracket, a camera, a camera connecting line and a light source. The fabric stage is used to place fabric samples, the camera bracket is used to install the camera, the camera adopts a high-resolution digital camera of more than 5 million, and the camera has automatic focus and exposure adjustment functions; the camera connecting line connects the camera to a computer; the light source adopts an LED light source and is installed on the camera bracket.

Images