Woven fabric density intelligent detection method and system based on multispectral imaging

Abstract

The invention belongs to the technical field of textile detection, and discloses an intelligent woven fabric density detection method and system based on multispectral imaging, and the method comprises the steps: collecting multispectral fabric images through a multispectral imaging device, and carrying out the noise reduction and texture contrast enhancement processing of the multispectral fabric images in sequence through the combination of an image processing technology, obtaining a fabric enhanced image; constructing a deep convolutional network classifier to classify the fabric texture, and segmenting a yarn region in the fabric enhanced image based on a fabric texture classification result to obtain a segmented yarn binary mask image; and inputting the yarn arrangement direction into a sub-pixel density calculation model, and calculating a yarn density value through the sub-pixel density calculation model. In the yarn breakage repairing stage, the broken yarn is repaired according to the initial warp and weft direction meeting the orthogonal constraint, and the calculation precision of the yarn arrangement direction is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of textile detection, and particularly to an intelligent detection method and system for the density of woven fabrics based on multispectral imaging. Background Art

[0002] A woven fabric is a fabric formed by warp yarns and weft yarns interweaving with each other according to a certain rule. Its density, that is, the number of yarns per unit length, is an important indicator for measuring the tightness and performance of the fabric structure. The yarn density directly affects many physical properties of the fabric such as hand feeling, durability, air permeability, etc. Therefore, density detection is crucial for ensuring fabric quality, controlling production processes and optimizing product performance, especially playing a vital role in fabric production and quality control.

[0003] Existing fabric density detection methods have multiple problems. For example, although the traditional manual disassembly method can obtain relatively accurate detection results, this method is cumbersome and time-consuming to operate, not only with low efficiency, but also inevitably damages the sample, unable to meet the requirements in large-scale production; the optical microscopy method is often limited by lighting conditions when dealing with complex textures or dark fabrics, resulting in a decrease in accuracy and difficulty in accurately capturing the subtle structural changes of the yarn; at the same time, although single-spectral imaging provides a non-contact detection method, due to the light reflection characteristics of the fabric surface, especially the smooth and shiny yarn surface, it is easily interfered by reflection, thereby affecting the stability and accuracy of the detection.

[0004] Therefore, how to provide an intelligent detection method and system for the density of woven fabrics based on multispectral imaging is an urgent problem to be solved at present. Summary of the Invention

[0005] Embodiments of the present invention provide an intelligent detection method and system for the density of woven fabrics based on multispectral imaging to solve the problem in the prior art that it is difficult to accurately capture the subtle structural changes of the yarn.

[0006] To have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary part is not a general review, nor is it intended to identify key / important elements or describe the protection scope of these embodiments. Its sole purpose is to present some concepts in a simple form as a preface to the subsequent detailed description.

[0007] According to the first aspect of the embodiments of the present invention, an intelligent detection method for the density of woven fabrics based on multispectral imaging is provided.

[0008] In one embodiment, the intelligent detection method for the density of woven fabrics based on multispectral imaging includes: Collect multi - spectral fabric images using a multi - spectral imaging device, and successively perform noise reduction and texture contrast enhancement processing on the multi - spectral fabric images in combination with image processing techniques to obtain a fabric enhanced image; Construct a deep convolutional network classifier to classify fabric textures, and segment the yarn regions in the fabric enhanced image based on the fabric texture classification results to obtain a segmented binary mask image of the yarn; Use frequency - domain analysis techniques to identify the yarn arrangement direction in the binary mask image of the yarn, input the yarn arrangement direction into a sub - pixel - level density calculation model, and calculate the yarn density value through the sub - pixel - level density calculation model.

[0009] In one embodiment, the step of collecting multi - spectral fabric images using a multi - spectral imaging device, and successively performing noise reduction and texture contrast enhancement processing on the multi - spectral fabric images in combination with image processing techniques to obtain a fabric enhanced image includes: Integrate a three - band multi - spectral camera to collect multi - spectral fabric images. According to the brightness feedback of the multi - spectral fabric images, use an adaptive illumination compensation algorithm to adjust the brightness array in real time to eliminate ambient light interference and balance the multi - band illuminance difference; Use an improved bilateral filtering algorithm to perform joint denoising processing on the multi - spectral fabric images, dynamically balancing the filtering weights in the spatial domain and the intensity domain; Decompose the reflection component of the denoised multi - spectral fabric image based on the multi - scale Retinex algorithm, and fuse the reflection components to enhance the texture contrast of the denoised multi - spectral fabric image to obtain a fabric enhanced image.

[0010] In one embodiment, the step of decomposing the reflection component of the denoised multi - spectral fabric image based on the multi - scale Retinex algorithm, and fusing the reflection components to enhance the texture contrast of the denoised multi - spectral fabric image to obtain a fabric enhanced image includes: Convert the original multi - spectral fabric image to the CLH color space, calculate the difference map of the original multi - spectral fabric image in the chromaticity channel, and threshold the difference value through an adaptive threshold segmentation algorithm to obtain an initial label map; Figure 2 Value the difference value to obtain an initial label map; Fuse the maximum between - class variance thresholds of the visible light and near - infrared bands in the original multi - spectral fabric image to generate a multi - spectral binary map, fuse the initial label map with the multi - spectral binary map and eliminate isolated regions to obtain an optimized multi - spectral label map; Use the optimized multi - spectral label map to perform morphological dilation on the original multi - spectral fabric image, and capture the reflection components of microscopic yarns, mesoscopic weaving structures, and macroscopic backgrounds in the dilated regions; Fuse the reflection components of microscopic yarns, mesoscopic weaving structures, and macroscopic backgrounds based on weighted summation to form a fabric enhanced image with enhanced texture contrast.

[0011] In one embodiment, the constructed deep convolutional network classifier classifies fabric textures, and based on the fabric texture classification results, segments the yarn regions in the fabric enhanced image to obtain a segmented binary mask image of the yarns, including: Construct a deep convolutional network classifier, and train the deep convolutional network classifier based on a predefined training set, and use the trained deep convolutional network classifier to output probability vectors of various fabric textures; Use the encoder of the deep convolutional network classifier as the encoder of the improved U-Net model, and input the fabric enhanced image into the encoder, and output a feature map at the end of the encoder; Map the probability vector to a channel attention weight through the fully connected layer of the improved U-Net model, and multiply the channel attention weight and the feature map channel by channel to focus on the yarn regions of the feature map; Introduce a dilated spatial pyramid pooling module in the decoder stage of the improved U-Net model, and gradually restore the resolution of the yarn regions of the feature map to obtain a segmented binary mask image of the yarns.

[0012] In one embodiment, the dilated spatial pyramid pooling module includes a number of dilated convolutional layers with different dilation rates, which are used to capture multi-scale yarn structure features from micro to macro on the feature map of the same level; Among them, the dilated convolutional layer with a small dilation rate is used to focus on the width and edge details of a single yarn; The dilated convolutional layer with a medium dilation rate is used to perceive the local weaving structure of the warp and weft intersection points; The dilated convolutional layer with a large dilation rate is used to analyze the overall texture trend of the fabric.

[0013] In one embodiment, the frequency domain analysis technology is used to identify the yarn arrangement direction in the binary mask image of the yarns, and the yarn arrangement direction is input into the sub-pixel level density calculation model, and the yarn density value is calculated by the sub-pixel level density calculation model, including: Perform a fast Fourier transform on the binary mask image of the yarns to generate a spectrogram, and extract the main frequency components of the yarns based on the frequency distribution in the spectrogram; Set the Hough angle search range based on the main frequency components of the yarns, and use the Hough transform technology to detect the yarn arrangement direction, and at the same time suppress the interference of broken yarns during the detection process to obtain the warp and weft yarn arrangement directions; Construct a direction filter in the spectrogram according to the yarn arrangement direction, and use the direction filter to generate a separated warp and weft yarn spatial domain image; Perform a transform projection on the warp and weft yarn spatial domain image along the warp and weft yarn arrangement directions to generate a projection curve, and perform sub-pixel level peak detection on the projection curve by the spline interpolation method to obtain a sequence of yarn center coordinates; Input the sequence of yarn center coordinates into the sub-pixel level density calculation model to calculate the warp and weft yarn densities, and compare the calculation results of the warp and weft yarn densities with a preset range, and optimize the sub-pixel level density calculation model according to the comparison results.

[0014] In one embodiment, the expression of the direction filter is: ; In the formula, G ( u , v ; θ , f 0) represents the functional form of the direction filter; ( u , v ) represents the original frequency domain coordinates; θ represents the main direction angle of the direction filter; f 0 represents the center frequency of the filter; σ f represents the standard deviation of the frequency domain Gaussian function; represents the new coordinates obtained by rotating the original frequency domain coordinates around the origin by θ .

[0015] In one embodiment, suppressing the interference of broken yarns during the detection process to obtain the arrangement directions of warp and weft yarns includes: Perform morphological closing operation on the spectrogram to connect the broken gaps, generate a preliminary continuous yarn region, and extract the angle corresponding to the main lobe of energy in the preliminary continuous yarn region as a rough estimate reference for the yarn arrangement direction; Based on the rough estimate reference and combined with the average yarn length in the current spectrogram, set the accumulator threshold to filter out broken yarns smaller than the accumulator threshold; Cluster the preliminary continuous yarn region according to the angle to generate the initial warp and weft yarn directions, and perform orthogonality verification on the initial warp and weft yarn directions, and adjust the clustering center based on the verification results until the orthogonality constraint is satisfied; Repair the broken yarns in turn according to the initial warp and weft yarn directions that satisfy the orthogonality constraint, and evaluate the accuracy of the warp and weft yarn directions according to the repair results of the broken yarns to obtain the final yarn arrangement direction.

[0016] In one embodiment, performing transformation projection on the warp and weft yarn spatial domain image along the warp and weft yarn arrangement directions to generate a projection curve, and performing sub-pixel level peak detection on the projection curve by the spline interpolation method to obtain the sequence of yarn center coordinates includes: Perform directional projection on the warp and weft yarn spatial domain image along the warp and weft yarn arrangement directions respectively to obtain a one-dimensional projection curve, and generate dense interpolation points at the discrete sampling points of the one-dimensional projection curve at sub-pixel intervals; Dense interpolation points on the smoothed projection curve are fitted using a cubic spline function, and the first derivative of the cubic spline function is obtained to locate the critical points as candidate positions for the yarn center; Calculate the second derivative at the critical points and determine the type of extreme value. If the second derivative is less than or equal to zero, it indicates that the dense interpolation point is a local maximum point and corresponds to the yarn center position. Otherwise, it is a local minimum point and corresponds to the yarn gap background area; Only the sub-pixel peaks corresponding to the local maximum points are retained within the sub-pixel peak neighborhood, and the sub-pixel peaks are back-projected into the original image coordinate system according to the warp and weft yarn arrangement directions to generate a sequence of yarn center coordinates.

[0017] According to the second aspect of the embodiments of the present invention, an intelligent detection system for woven fabric density based on multi-spectral imaging is provided.

[0018] In one embodiment, the intelligent detection system for woven fabric density based on multi-spectral imaging includes: An image preprocessing module, configured to collect a multi-spectral fabric image using a multi-spectral imaging device, and sequentially perform noise reduction and texture contrast enhancement processing on the multi-spectral fabric image in combination with image processing techniques to obtain a fabric enhanced image; A yarn region analysis module, configured to construct a deep convolutional network classifier to classify the fabric texture, and segment the yarn region in the fabric enhanced image based on the fabric texture classification result to obtain a segmented binary mask image of the yarn; A yarn density calculation module, configured to use frequency domain analysis techniques to identify the yarn arrangement direction in the binary mask image of the yarn, input the yarn arrangement direction into a sub-pixel level density calculation model, and calculate the yarn density value through the sub-pixel level density calculation model.

[0019] According to the third aspect of the embodiments of the present invention, a computer device is provided.

[0020] In one embodiment, the computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0021] According to the fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided.

[0022] In one embodiment, a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0023] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: 1. The present invention decomposes and fuses the image reflection component based on the multi-scale Retinex algorithm, significantly enhancing the texture contrast of the fabric, making the microstructure clearer. By converting the image to the CLH color space and combining with adaptive threshold segmentation, the yarns and weaving structures are accurately extracted. The reflection components of the micro-yarns, meso-weaving structures, and macro backgrounds are fused, improving the image quality and density detection accuracy, especially suitable for complex textures and multi-band data.

[0024] 2. The present invention accurately outputs the probability vector of the fabric texture through a classifier and integrates the trained deep convolutional network classifier into the improved U-Net model to effectively extract image features. A dilated spatial pyramid pooling module is introduced in the decoder stage, and the resolution of the yarn region is gradually restored through multi-scale dilated convolutional layers to capture the multi-scale structural features of the yarns, greatly improving the texture segmentation accuracy and feature capture ability, ensuring the comprehensive capture of fabric features at multiple levels, and greatly improving the accuracy of fabric texture segmentation and multi-scale feature capture ability.

[0025] 3. The present invention ensures the accuracy of the warp and weft yarn directions through orthogonality verification. If the initial warp and weft yarn directions do not meet the orthogonal constraints, the clustering centers will be adjusted to ensure that the clustering results meet the orthogonal requirements, thereby improving the accuracy of the yarn arrangement direction. In the yarn breakage repair stage, the broken yarns are repaired according to the initial warp and weft yarn directions that meet the orthogonal constraints, effectively improving the calculation accuracy of the yarn arrangement direction, and further providing reliable data support for fabric density analysis.

[0026] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0028] Figure 1 is a flowchart of an intelligent detection method for the density of woven fabrics based on multi-spectral imaging shown according to an exemplary embodiment; Figure 2 is a schematic block diagram of the principle of an intelligent detection system for the density of woven fabrics based on multi-spectral imaging shown according to an exemplary embodiment; Figure 3 is a schematic structural diagram of a computer device shown according to an exemplary embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The following description and accompanying drawings fully illustrate the specific embodiments of this article so that those skilled in the art can practice them. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. The scope of the embodiments of this article includes the entire scope of the claims, as well as all available equivalents of the claims. Herein, the terms "first", "second", etc. are only used to distinguish one element from another, without requiring or implying any actual relationship or order between these elements. In fact, the first element can also be called the second element, and vice versa. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the structure, device or equipment including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also include elements inherent to such structure, device or equipment. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the structure, device or equipment including the elements. Each embodiment is described in a progressive manner herein, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referred to each other.

[0030] As used herein, the term "plurality" means two or more than two, unless otherwise specified.

[0031] In this document, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B.

[0032] In this article, the term "and / or" is a description of the association relationship between objects, indicating that three relationships may exist. For example, A and / or B means: A or B, or, A and B.

[0033] It should be understood that, although the various steps in the flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the figure may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0034] Each module in the device or system of the present application can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0035] Without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0036] Figure 1 An embodiment of the intelligent detection method for the density of woven fabrics based on multispectral imaging according to the present invention is shown.

[0037] In this alternative embodiment, the intelligent detection method for the density of woven fabrics based on multispectral imaging includes: Step S101: Use a multispectral imaging device to collect a multispectral fabric image, and sequentially perform noise reduction and texture contrast enhancement processing on the multispectral fabric image in combination with image processing technology to obtain a fabric enhanced image; Step S102: Construct a deep convolutional network classifier to classify the fabric texture, and segment the yarn area in the fabric enhanced image based on the fabric texture classification result to obtain a segmented yarn binary mask image; Step S103: Use frequency domain analysis technology to identify the yarn arrangement direction in the yarn binary mask image, input the yarn arrangement direction into the sub-pixel level density calculation model, and calculate the yarn density value through the sub-pixel level density calculation model.

[0038] In this alternative embodiment, the use of a multispectral imaging device to collect a multispectral fabric image, and sequentially perform noise reduction and texture contrast enhancement processing on the multispectral fabric image in combination with image processing technology to obtain a fabric enhanced image includes: Integrate a three-band multispectral camera to collect a multispectral fabric image. According to the brightness feedback of the multispectral fabric image, use an adaptive illumination compensation algorithm to adjust the brightness array in real time to eliminate ambient light interference and balance the multi-band illuminance difference.

[0039] It should be noted that the three-band multispectral camera includes multispectral cameras in three bands: visible light (400 - 700nm), near-infrared (850 - 950nm), and short-wave infrared (1000 - 1300nm). It cooperates with a polarized light source to eliminate specular reflection, and the illumination system uses an LED ring light source (color temperature 5500K) in cooperation with a polarizer group. Image processing uses an improved Canny edge detection algorithm, and set Gaussian filtering σ =1.2, and the double threshold ratio is 1:3. After testing, the detection error for black polyester fabrics is reduced from 8.2% of the traditional method to 1.1%. Among them, the adaptive illumination compensation algorithm includes:

[0040] 1. Real-time monitoring: Calculate the average brightness value of the ROI area (the central 80% range) using the multi-spectral fabric images captured by the camera. L avg ; 2. PID regulation: Dynamically adjust the LED drive current according to the target brightness L target = 0.4 (normalized value): I new = I prev + K p ( L target − L avg ) + K i ∑( L target − L avg )Δ t ; In the formula, I new represents the adjusted LED drive current; I prev represents the historical LED drive current; Δ t represents the time interval; K p represents the proportional term; K i represents the integral term.

[0041] 3. Multi-band independent control: Separate LED drive modules are added for the near-infrared and short-wave infrared bands to compensate for the penetration difference of different wavelengths.

[0042] Use the improved bilateral filtering algorithm to perform joint denoising on the multi-spectral fabric images, and dynamically balance the filtering weights in the spatial domain and intensity domain.

[0043] It should be noted that the improved bilateral filtering algorithm is a spatial-intensity dual-domain adaptive filtering algorithm, which specifically includes: 1. Modeling: Establish a mixed noise model (Gaussian noise + Poisson noise) for the multi-spectral image characteristics and estimate the noise variance ; 2. Dynamic parameter adjustment: ; In the formula, W ( i , j ) represents the weight function, which is used to measure two data pointsi and j the similarity between; ( x i, x j ) represents the spatial coordinates; ( I i , I j ) represents the intensity coordinates; σ d represents the standard deviation of the spatial domain; σ r represents the standard deviation of the spatial domain; σ n represents the noise variance term, where σ d = 3 (standard deviation of the intermediate domain), σ r = 0.1 (standard deviation of the intensity domain), and the noise variance term suppresses the color distortion in the high-noise region.

[0044] Decompose the reflection component of the denoised multi-spectral fabric image based on the multi-scale Retinex algorithm, and fuse the reflection components to enhance the texture contrast of the denoised multi-spectral fabric image, obtaining a fabric enhanced image.

[0045] It should be noted that the multi-scale Retinex algorithm (Multi-Scale Retinex, MSR) is an image enhancement algorithm aimed at enhancing the details and contrast of an image to improve the visual quality of the image. This algorithm is an extension of the Retinex algorithm, mainly by processing the image at multiple scales to remove the influence of illumination and shadows, highlighting the reflection information, thereby obtaining better image performance. The basic idea of the Retinex algorithm originates from the visual processing method of the human eye. When the human eye perceives color, it can automatically separate the reflected color of an object from the illumination change.

[0046] In this optional embodiment, the decomposing the reflection component of the denoised multi-spectral fabric image based on the multi-scale Retinex algorithm and fusing the reflection components to enhance the texture contrast of the denoised multi-spectral fabric image to obtain a fabric enhanced image includes: Convert the original multi-spectral fabric image to the CLH color space, calculate the difference map of the original multi-spectral fabric image in the chromaticity channel, and binarize the difference Figure 2 value through the adaptive threshold segmentation algorithm to obtain an initial label map; Fuse the maximum between-class variance threshold of the visible light and near-infrared bands in the original multi-spectral fabric image to generate a multi-spectral binary map, fuse the initial label map with the multi-spectral binary map and eliminate the isolated regions to obtain an optimized multi-spectral label map; Perform morphological dilation on the original multispectral fabric image using the optimized multispectral label map, and capture the reflection components of the microscopic yarns, mesoscopic weaving structures, and macroscopic background in the dilated regions; Fuse the reflection components of the microscopic yarns, mesoscopic weaving structures, and macroscopic background based on weighted summation to form a fabric enhanced image with enhanced texture contrast.

[0047] It should be noted that fusing the reflection components of the microscopic yarns, mesoscopic weaving structures, and macroscopic background based on weighted summation to form a fabric enhanced image with enhanced texture contrast includes: Step 1: Morphological region expansion guided by the label map: 1. Label map preprocessing: Input the optimized multispectral label map (binary image, white represents the yarn region, black is the background).

[0048] Use a 3×3 elliptical kernel to perform morphological dilation on the label map (expand the boundary of the white region by 1 - 2 pixels) to ensure that areas with broken yarns or discontinuous edges are filled, forming a continuous target operation area.

[0049] 2. Multi - scale reflection component extraction: Microscopic yarn layer: In the dilated region, apply a small - scale Gaussian filter (such as a 5×5 window) to the original image to smooth the noise on the yarn surface while retaining the fine texture of single fibers.

[0050] Mesoscopic weaving structure layer: In the same region, use a medium - scale Gaussian filter (such as a 15×15 window) to blur the details of single yarns and highlight the grid - like structure of the warp and weft intersections.

[0051] Macroscopic background layer: In the background region outside the label map, use a large - scale Gaussian filter (such as a 50×50 window) to eliminate local brightness differences and retain the overall illumination trend.

[0052] Step 2: Label - driven multi - scale reflection fusion: 1. Weight assignment strategy: Microscopic layer weight: In the yarn region marked by the label map, assign the highest weight (such as 0.6) to strengthen the yarn edges and fiber details.

[0053] Mesoscopic layer weight: At the edges and intersection points of the yarn region, assign a medium weight (such as 0.3) to enhance the visibility of the weaving structure.

[0054] Macroscopic layer weight: In the background region, set the weight to 0.8 to smooth the illumination interference; in the yarn region, the weight is reduced to 0.1 to avoid background information covering the texture.

[0055] 2. Pixel - level fusion operation: For each pixel, read the preset weight from the label map according to its belonging area (yarn / background): Pixels in the yarn area: Enhanced pixel value = Microscopic layer value × 0.6 + Mesoscopic layer value × 0.3 + Macroscopic layer value × 0.1; Pixels in the background area: Smoothed pixel value = Microscopic layer value × 0.1 + Mesoscopic layer value × 0.1 + Macroscopic layer value × 0.8; Perform the calculation of pixels in the yarn area and the calculation of pixels in the background area on the entire image, and directly output the fused enhanced image. For example: 1. Input data: Original multi-spectral image (512×512 pixels, including visible light and near-infrared bands).

[0056] Optimized label map (yarn binary mask generated by the classification-segmentation model).

[0057] 2. Operation records: Morphological dilation: 3×3 elliptical kernel, iterated 2 times → Expand the yarn area to cover the actual physical width.

[0058] Gaussian filter parameters: Microscopic layer: σ = 1.5 (5×5 kernel), capturing the fiber surface of 10 - 20μm.

[0059] Mesoscopic layer: σ = 4.5 (15×15 kernel), extracting 0.5 - 1mm knitting grid.

[0060] Macroscopic layer: σ = 15 (50×50 kernel), balancing the vignetting at the four corners of the image.

[0061] CLAHE parameters: Block size 32×32, contrast limit 2.0, stretching the gray distribution of the yarn.

[0062] 3. Output results: The contrast of the yarn texture is increased by more than 3 times, and the background uniformity (standard deviation) is decreased by 60%, meeting the requirements of defect detection.

[0063] In this alternative embodiment, the constructed deep convolutional network classifier classifies the fabric texture, and based on the fabric texture classification result, segments the yarn area in the fabric enhanced image to obtain the segmented yarn binary mask image, including: Construct a deep convolutional network classifier, train the deep convolutional network classifier based on a predefined training set, and use the trained deep convolutional network classifier to output the probability vectors of various fabric textures; Use the encoder of the deep convolutional network classifier as the encoder of the improved U-Net model, and input the fabric-enhanced image into the encoder to output a feature map at the end of the encoder; Map the probability vector to a channel attention weight through the fully connected layer of the improved U-Net model, and multiply the channel attention weight with the feature map channel by channel to focus on the yarn region of the feature map; Introduce a dilated spatial pyramid pooling module in the decoder stage of the improved U-Net model to gradually restore the resolution of the yarn region of the feature map and obtain a segmented binary mask image of the yarn.

[0064] In this alternative embodiment, the dilated spatial pyramid pooling module includes several dilated convolutional layers with different dilation rates for capturing multi-scale yarn structure features from micro to macro on the same-level feature map; Among them, the dilated convolutional layer with a small dilation rate is used to focus on the width and edge details of a single yarn; The dilated convolutional layer with a medium dilation rate is used to perceive the local weaving structure at the intersection of warp and weft yarns; The dilated convolutional layer with a large dilation rate is used to analyze the overall texture trend of the fabric.

[0065] It should be noted that the main improvement of the improved U-Net model is reflected in introducing the encoder part of the deep convolutional network classifier and adding a dilated spatial pyramid pooling module in the decoder stage to improve the fabric texture segmentation accuracy and multi-scale feature capture ability.

[0066] In the decoder stage, a dilated spatial pyramid pooling module is introduced to capture multi-scale yarn structure features from micro to macro with dilated convolutional layers of different dilation rates. The dilated convolutional layer with a small dilation rate mainly focuses on the width and edge details of a single yarn, improving the resolution of fine yarns; the dilated convolutional layer with a medium dilation rate perceives the local weaving structure at the intersection of warp and weft yarns, capturing the changes in local texture; the dilated convolutional layer with a large dilation rate helps to analyze the overall texture trend and extract more extensive pattern features. By gradually restoring the resolution of the feature map, this process can accurately restore the details of the yarn region and enhance the segmentation effect. Finally, after this series of improvements, the obtained segmented binary mask image of the yarn can accurately distinguish the yarn region, providing an accurate basis for subsequent fabric analysis.

[0067] In this alternative embodiment, the method for identifying the yarn arrangement direction in the yarn binary mask image using frequency domain analysis technology and inputting the yarn arrangement direction into the sub-pixel level density calculation model to calculate the yarn density value by the sub-pixel level density calculation model includes: Perform a fast Fourier transform on the yarn binary mask image to generate a frequency spectrum diagram, and extract the main frequency components of the yarn based on the frequency distribution in the frequency spectrum diagram; Set the Hough angle search range based on the main frequency components of the yarns, and use the Hough transform technology to detect the arrangement direction of the yarns. At the same time, suppress the interference of broken yarns during the detection process to obtain the arrangement directions of the warp and weft yarns. Construct a direction filter in the frequency spectrum diagram according to the yarn arrangement direction, and use the direction filter to generate the separated warp and weft yarn spatial domain images. Perform a transform projection on the warp and weft yarn spatial domain images along the warp and weft yarn arrangement directions to generate projection curves, and perform sub-pixel level peak detection on the projection curves through the spline interpolation method to obtain the yarn center coordinate sequences. Input the yarn center coordinate sequences into the sub-pixel level density calculation model to calculate the warp and weft yarn densities, and compare the calculation results of the warp and weft yarn densities with the preset range. Optimize the sub-pixel level density calculation model according to the comparison results.

[0068] It should be noted that the sub-pixel level density calculation model is a method for statistically calculating the warp and weft yarn densities of fabrics through high-precision coordinate data. First, it is necessary to obtain the sub-pixel level coordinates of the yarn center points, which is usually achieved through image segmentation and sub-pixel localization algorithms (such as Gaussian fitting or interpolation), with an accuracy of up to 0.1 pixel level to ensure that the coordinate data is fine enough.

[0069] Project the coordinates onto the reference axis according to the arrangement direction of the yarns (the warp yarns are usually vertically arranged, and the weft yarns are horizontally arranged), such as the X-axis coordinates of the warp yarns or the Y-axis coordinates of the weft yarns, and calculate the density by statistically calculating the average distance between adjacent yarns. In actual operation, due to possible local deformations or cross regions in the fabric, it is necessary to dynamically adjust the spacing calculation in the local area through weighted average or sliding window methods. For example, appropriately reduce the calculation window near the intersection points to avoid error accumulation.

[0070] Convert the average spacing to physical units (such as millimeters or centimeters) and take the reciprocal, and calculate the number of yarns per centimeter in combination with the image resolution (such as pixels / cm). The entire process needs to flexibly adjust parameters according to the fabric type (such as plain weave, twill). For example, for twill fabrics, angle correction may be required, and for high-density fabrics, noise resistance processing needs to be enhanced to ensure the stability and reliability of the output results.

[0071] In this alternative embodiment, the expression of the direction filter is: ; In the formula, G ( u , v ; θ , f 0) represents the functional form of the direction filter; ([[]] u , v ) represents the original frequency domain coordinates; θRepresents the main direction angle of the directional filter; f 0 represents the center frequency of the filter; σ f Represents the standard deviation of the Gaussian function in the frequency domain; Represents the new coordinates obtained by rotating the original frequency domain coordinates around the origin by θ ;

[0072] In this alternative embodiment, suppressing the interference of broken yarns during the detection process to obtain the warp and weft yarn arrangement directions includes: Performing morphological closing operation on the spectrogram to connect the broken gaps, generating a preliminary continuous yarn region, and extracting the angle corresponding to the main lobe of energy in the preliminary continuous yarn region as a rough estimate reference for the yarn arrangement direction; Based on the rough estimate reference and combined with the average yarn length in the current spectrogram, setting an accumulator threshold to filter out broken yarns smaller than the accumulator threshold; Clustering the preliminary continuous yarn region according to the angle to generate the initial warp and weft yarn directions, and performing orthogonality verification on the initial warp and weft yarn directions, and adjusting the clustering center based on the verification result until the orthogonality constraint is satisfied; Repairing the broken yarns in sequence according to the initial warp and weft yarn directions that satisfy the orthogonality constraint, and evaluating the accuracy of the warp and weft yarn directions according to the repair results of the broken yarns to obtain the final yarn arrangement direction.

[0073] It should be noted that repairing the broken yarns in sequence according to the initial warp and weft yarn directions that satisfy the orthogonality constraint, and evaluating the accuracy of the warp and weft yarn directions according to the repair results of the broken yarns to obtain the final yarn arrangement direction includes: Step 1: Adaptive Hough transform of the accumulator to suppress broken interference: 1. Improving the parameterization of the Hough transform: Direction constraint: Set the angle search range to θ 0 ± 15°, reducing the invalid detection range.

[0074] Accumulator adaptive threshold: Dynamic threshold calculation: According to the average yarn length (Lavg) in the current image, set the accumulator threshold as T = 0.6 × Lavg (for example, if Lavg = 50 pixels, then T = 30).

[0075] Broken filtering: Only accept line segments with an accumulated value ≥ T, filtering out short breaks (length < T).

[0076] 2. Separating the warp and weft yarn directions: The detected line segments are clustered according to the angle to generate two main directions θ 1 (warp yarn) and θ 2 = θ 1 + 90° (weft yarn).

[0077] Orthogonality verification: If | θ 1 - θ 2 - 90°| > 5°, readjust the clustering center until the orthogonality constraint is satisfied.

[0078] Step 2: Direction optimization and fracture area repair: 1. Morphological repair guided by direction: Warp repair: Apply a linear structuring element (length = 10 pixels, width = 1 pixel) along the θ 1 direction for closing operation to fill the warp fractures.

[0079] Weft repair: Similarly operate along the θ 2 direction to connect the fractured wefts.

[0080] 2. Frequency domain - spatial domain cross - verification: Convert the repaired image to the frequency domain and verify whether the spectral energy is concentrated in the θ 1 / θ 2 direction (energy proportion > 70%).

[0081] If not passed, return to Step 2 to adjust the accumulator threshold or the direction constraint range.

[0082] Step 3: Direction output and iteration termination condition: 1. Direction accuracy evaluation: Length consistency: Statistically calculate the variance (Var) of the line segment lengths in the θ 1 direction, and require Var < 20 pixel² (to ensure direction stability).

[0083] Fracture residue rate: Calculate the proportion of the fracture area in the repaired image, and set the threshold to < 2%.

[0084] 2. Result output and iteration: If the above indicators are all satisfied, output the final warp and weft directions θ 1 and θ 2, and generate separate binary masks for warp / weft yarns.

[0085] If any indicator exceeds the limit, return to Step 2, reduce the accumulator threshold (T = T × 0.9) or relax the orthogonality constraint (±2° increment), and re - detect.

[0086] In this alternative embodiment, the spatial domain image of the warp and weft yarns is transformed and projected along the arrangement direction of the warp and weft yarns to generate a projection curve, and sub - pixel peak detection is performed on the projection curve by the spline interpolation method, and the obtained sequence of yarn center coordinates includes: Perform directional projection on the warp and weft yarn spatial domain images respectively along the arrangement directions of the warp and weft yarns to obtain one-dimensional projection curves. At the discrete sampling points of the one-dimensional projection curves, dense interpolation points are generated at sub-pixel intervals. Use a cubic spline function to fit the dense interpolation points on the smoothed projection curve, and take the first derivative of the cubic spline function to locate the critical points as the candidate positions of the yarn centers. Calculate the second derivative at the critical points and determine the type of extreme value. If the second derivative is less than or equal to zero, it means that the dense interpolation point is a local maximum point and corresponds to the yarn center position. Otherwise, it is a local minimum point and corresponds to the yarn gap background area. Only retain the sub-pixel peaks corresponding to the local maximum points within the sub-pixel level peak neighborhood, and reverse the sub-pixel peaks to the original image coordinate system according to the arrangement directions of the warp and weft yarns to generate a sequence of yarn center coordinates.

[0087] It should be noted that only retaining the sub-pixel peaks corresponding to the local maximum points within the sub-pixel level peak neighborhood and reversing the sub-pixel peaks to the original image coordinate system according to the arrangement directions of the warp and weft yarns to generate a sequence of yarn center coordinates includes: Step 1. Generation of projection curves and sampling of sub-pixel interpolation points: Along the warp direction ( θ 1) and the weft direction ( θ 2), perform Radon transform directional projection on the separated warp / weft yarn spatial domain images respectively to obtain two one-dimensional projection curves. Each curve takes integer pixel positions as discrete sampling points and records the projection intensity values. Subsequently, dense interpolation points are generated at sub-pixel intervals (such as 0.1 pixel) between the discrete sampling points to expand the sampling resolution to the sub-pixel level, providing a data basis for subsequent accurate peak detection.

[0088] Step 2. Cubic spline fitting and critical point location: Based on the natural boundary conditions (the second derivative at both ends of the spline curve is zero), use a cubic spline function to fit the smoothed projection curve to construct a continuously differentiable spline model S ( x ). Take the first derivative of the spline function S ′( x ), and solve the equation S ′( x ) = 0 through a numerical iteration method (such as the Newton-Raphson method) to locate all critical points x pea That is, the possible candidate positions of the yarn centers.

[0089] Step 3. Discrimination of extreme value types and peak screening: Calculate the second derivative at the critical points S ′′(x peak ), determine the extreme value type according to its sign: If S ′′( x peak ) ≤ 0, it is determined as a local maximum point, corresponding to the yarn center position; If S ′′( x peak ) > 0, it is determined as a local minimum point, corresponding to the yarn gap background area.

[0090] Only retain the maximum points and filter out the false peaks with amplitudes lower than the dynamic threshold (such as 1.2 times the average projection intensity).

[0091] Step 4, Non-maximum suppression and coordinate mapping: Within the sub-pixel peak neighborhood (such as the range of ±1.5 pixels), adopt the non-maximum suppression strategy to only retain the peak with the largest amplitude. Map the filtered sub-pixel positions x peak According to the projection direction θ 1 / θ 2 back to the original image coordinate system to generate the warp center coordinate sequence {( x k , y k )} and the weft center coordinate sequence {( u m , v m )}, and complete the sub-pixel positioning of the yarn position.

[0092] Figure 2 Fig. shows an embodiment of the intelligent detection system for woven fabric density based on multi-spectral imaging of the present invention.

[0093] In this alternative embodiment, the intelligent detection system for woven fabric density based on multi-spectral imaging includes: An image preprocessing module 201, configured to collect a multi-spectral fabric image by using a multi-spectral imaging device, and sequentially perform noise reduction and texture contrast enhancement processing on the multi-spectral fabric image in combination with image processing techniques to obtain a fabric enhanced image; A yarn area analysis module 202, configured to construct a deep convolutional network classifier to classify the fabric texture, and segment the yarn area in the fabric enhanced image based on the fabric texture classification result to obtain a segmented binary mask image of the yarn; A yarn density calculation module 203, configured to use frequency domain analysis technology to identify the yarn arrangement direction in the binary mask image of the yarn, input the yarn arrangement direction into the sub-pixel level density calculation model, and calculate the yarn density value through the sub-pixel level density calculation model.

[0094] It should be noted that through multi-spectral collaborative imaging and adaptive algorithm optimization, the present invention achieves the following core advantages: High-precision detection: The detection accuracy meets the ISO 7211-6 standard, with a relative error < 1.5%, accurately quantifying yarn density and defects, and meeting the strict quality control requirements of the textile industry; Strong applicability: It supports non-destructive detection of dark fabrics (such as black velvet), highly reflective materials (metal fiber blends), and multi-component blended fabrics, breaking through the misdetection bottleneck caused by the light absorption / reflective characteristics of materials in traditional optical methods; Efficient processing: The full-process detection time for a single sample is < 15 seconds, supporting real-time quality inspection on the production line, with an efficiency improvement of more than 20 times compared to manual detection; Zero-damage operation: Based on non-contact optical imaging technology, it avoids stretching or scratching of the specimen in traditional mechanical detection, ensures the reusability of the sample, and reduces enterprise costs.

[0095] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as shown in Figure 3 the figure. The computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0096] Those skilled in the art can understand that Figure 3 the structure shown in is only a block diagram of some structures related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0097] In addition, the present invention also provides a computer device, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps in the above method embodiments.

[0098] In addition, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0099] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above various methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0100] The present invention is not limited to the structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. An intelligent detection method for the density of woven fabrics based on multispectral imaging, characterized in that, The method includes: Collecting a multi-spectral fabric image by using a multi-spectral imaging device, and successively performing noise reduction and texture contrast enhancement processing on the multi-spectral fabric image in combination with image processing techniques to obtain a fabric enhanced image; Constructing a deep convolutional network classifier to classify fabric textures, and segmenting the yarn regions in the fabric enhanced image based on the fabric texture classification result to obtain a segmented binary mask image of the yarns; Identifying the yarn arrangement direction in the binary mask image of the yarns by using frequency domain analysis techniques, inputting the yarn arrangement direction into a sub-pixel level density calculation model, and calculating the yarn density value through the sub-pixel level density calculation model.

2. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 1, wherein The collecting a multi-spectral fabric image by using a multi-spectral imaging device, and successively performing noise reduction and texture contrast enhancement processing on the multi-spectral fabric image in combination with image processing techniques to obtain a fabric enhanced image includes: Integrating a three-band multi-spectral camera to collect a multi-spectral fabric image, and in accordance with the brightness feedback of the multi-spectral fabric image, using an adaptive illumination compensation algorithm to adjust the brightness array in real time to eliminate ambient light interference and balance the multi-band illuminance difference; Performing joint denoising processing on the multi-spectral fabric image by using an improved bilateral filtering algorithm, and dynamically balancing the filtering weights in the spatial domain and the intensity domain; Decomposing the reflection component of the denoised multi-spectral fabric image based on the multi-scale Retinex algorithm, and fusing the reflection components to enhance the texture contrast of the denoised multi-spectral fabric image to obtain a fabric enhanced image.

3. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 2, wherein The decomposing the reflection component of the denoised multi-spectral fabric image based on the multi-scale Retinex algorithm, and fusing the reflection components to enhance the texture contrast of the denoised multi-spectral fabric image to obtain a fabric enhanced image includes: Converting the original multi-spectral fabric image to the CLH color space, calculating the difference map of the original multi-spectral fabric image in the chromaticity channel, and binarizing the difference map through an adaptive threshold segmentation algorithm to obtain an initial label map; Fusing the maximum between-class variance thresholds of the visible light and near-infrared bands in the original multi-spectral fabric image to generate a multi-spectral binary map, fusing the initial label map and the multi-spectral binary map and eliminating isolated regions to obtain an optimized multi-spectral label map; Performing morphological dilation on the original multi-spectral fabric image by using the optimized multi-spectral label map, and capturing the reflection components of microscopic yarns, mesoscopic weaving structures and macroscopic backgrounds in the dilated regions; Fusing the reflection components of microscopic yarns, mesoscopic weaving structures and macroscopic backgrounds in a weighted summation manner to form a fabric enhanced image with enhanced texture contrast.

4. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 1, characterized in that The constructing a deep convolutional network classifier to classify fabric textures, and segmenting the yarn regions in the fabric enhanced image based on the fabric texture classification result to obtain a segmented binary mask image of the yarns includes: Constructing a deep convolutional network classifier, training the deep convolutional network classifier based on a predefined training set, and using the trained deep convolutional network classifier to output the probability vectors of various fabric textures; Taking the encoder of the deep convolutional network classifier as the encoder of the improved U-Net model, inputting the fabric enhanced image into the encoder, and outputting a feature map at the end of the encoder; The probability vector is mapped into channel attention weights through the fully connected layer of the improved U-Net model, and the channel attention weights are multiplied with the feature map channel by channel to focus on the yarn region of the feature map; At the decoder stage of the improved U-Net model, an atrous spatial pyramid pooling module is introduced to gradually restore the resolution of the yarn region of the feature map, and a segmented binary mask image of the yarn is obtained.

5. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 4, characterized in that, The atrous spatial pyramid pooling module includes several atrous convolutional layers with different dilation rates, which are used to capture multi-scale yarn structure features from micro to macro on the feature map at the same level; Among them, the atrous convolutional layer with a small dilation rate is used to focus on the width and edge details of a single yarn; The atrous convolutional layer with a medium dilation rate is used to perceive the local weaving structure of the warp and weft intersection points; The atrous convolutional layer with a large dilation rate is used to analyze the overall texture trend of the fabric.

6. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 1, characterized in that, The method for identifying the yarn arrangement direction in the binary mask image of the yarn by using frequency domain analysis technology and inputting the yarn arrangement direction into the sub-pixel level density calculation model to calculate the yarn density value by the sub-pixel level density calculation model includes: Performing a fast Fourier transform on the binary mask image of the yarn to generate a spectrogram, and extracting the main frequency components of the yarn based on the frequency distribution in the spectrogram; Setting the Hough angle search range based on the main frequency components of the yarn, and using the Hough transform technology to detect the yarn arrangement direction, and suppressing the interference of broken yarns during the detection process to obtain the warp and weft yarn arrangement directions; Constructing a direction filter in the spectrogram according to the yarn arrangement direction, and using the direction filter to generate separated warp and weft yarn spatial domain images; Performing a transform projection on the warp and weft yarn spatial domain images along the warp and weft yarn arrangement directions to generate projection curves, and performing sub-pixel level peak detection on the projection curves by using spline interpolation to obtain a sequence of yarn center coordinates; Inputting the sequence of yarn center coordinates into the sub-pixel level density calculation model to calculate the warp and weft yarn densities, comparing the calculation results of the warp and weft yarn densities with a preset range, and optimizing the sub-pixel level density calculation model according to the comparison results.

7. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 6, characterized in that, The expression of the direction filter is: ； In the formula, G ( u , v ; θ , f 0) represents the functional form of the directional filter; ([[]] u , v ) represents the original frequency domain coordinates; θ represents the main direction angle of the directional filter; f 0 represents the center frequency of the filter; σ f represents the standard deviation of the frequency domain Gaussian function; represents the new coordinates obtained by rotating the original frequency domain coordinates by θ around the origin.

8. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 6, characterized in that, The method for suppressing the interference of broken yarns during the detection process to obtain the warp and weft yarn arrangement directions includes: Performing a morphological closing operation on the spectrogram to connect the broken gaps, generating a preliminary continuous yarn region, and extracting the angle corresponding to the main lobe of energy in the preliminary continuous yarn region as a rough estimate reference for the yarn arrangement direction; Based on the rough estimate reference and combined with the average length of the yarns in the current spectrogram, setting an accumulator threshold to filter out broken yarns smaller than the accumulator threshold; Clustering the preliminary continuous yarn region according to the angle to generate initial warp and weft yarn directions, and performing orthogonality verification on the initial warp and weft yarn directions, and adjusting the clustering center based on the verification results until the orthogonality constraint is satisfied; Repairing the broken yarns in sequence according to the initial warp and weft yarn directions that satisfy the orthogonality constraint, and evaluating the accuracy of the warp and weft yarn directions according to the repair results of the broken yarns to obtain the final yarn arrangement direction.

9. The intelligent detection method for the density of woven fabrics based on multispectral imaging according to claim 8, wherein, The method for performing a transform projection on the warp and weft yarn spatial domain images along the warp and weft yarn arrangement directions to generate projection curves, and performing sub-pixel level peak detection on the projection curves by using spline interpolation to obtain a sequence of yarn center coordinates includes: Perform directional projection on the warp and weft yarn spatial domain images along the directions of the warp and weft yarn arrangements respectively to obtain one-dimensional projection curves. At the discrete sampling points of the one-dimensional projection curves, generate dense interpolation points at sub-pixel intervals; Use a cubic spline function to fit the dense interpolation points on the smoothed projection curve, and calculate the first derivative of the cubic spline function to locate the critical points as the candidate positions of the yarn centers; Calculate the second derivative at the critical points and determine the type of extremum. If the second derivative is less than or equal to zero, it indicates that the dense interpolation point is a local maximum point and corresponds to the yarn center position. Otherwise, it is a local minimum point and corresponds to the yarn gap background area; Only retain the sub-pixel peaks corresponding to the local maximum points within the sub-pixel level peak neighborhood, and reverse the sub-pixel peaks to the original image coordinate system according to the warp and weft yarn arrangement directions to generate a sequence of yarn center coordinates.

10. An intelligent detection system for the density of woven fabrics based on multispectral imaging, characterized in that, The system includes: An image preprocessing module for collecting multi-spectral fabric images using a multi-spectral imaging device, and sequentially performing noise reduction and texture contrast enhancement processing on the multi-spectral fabric images in combination with image processing techniques to obtain enhanced fabric images; A yarn region analysis module for constructing a deep convolutional network classifier to classify the fabric texture, and segmenting the yarn regions in the enhanced fabric image based on the fabric texture classification results to obtain a segmented binary mask image of the yarns; A yarn density calculation module for using frequency domain analysis techniques to identify the yarn arrangement directions in the binary mask image of the yarns, inputting the yarn arrangement directions into a sub-pixel level density calculation model, and calculating the yarn density values through the sub-pixel level density calculation model.

Images