Textile fabric with controllable orientation pattern and method for detecting the same

Abstract

This invention discloses a textile fabric with controllable orientation patterns and its detection method. By acquiring polarized image sequences, a rotating linearly polarized light source and an orthogonal polarization camera are used to capture the fabric's reflection images at different polarization angles, thus comprehensively recording the directional reflection characteristics of the pile. This distinguishes between normal brightness changes caused by the geometric direction of the pile and abnormal responses of real defects. Based on this, in defect area identification and segmentation, pixel clusters with low modulation depth and dominant polarization angles deviating from the surrounding trend are detected, effectively isolating real defect areas and avoiding interference from directional color differences in the pile. Finally, in defect classification and output, defect type identification and result output are performed. This scheme fundamentally solves the problem of false alarms and missed alarms caused by directional color differences in the pile. By separating normal optical changes from defects through polarization response characteristics, it significantly improves the accuracy and reliability of automatic quality inspection of piled textile fabrics.

Description

Technical Field

[0001] This invention relates to the field of textile fabric production technology, specifically to a textile fabric with a controllable orientation pattern and its testing method. Background Technology

[0002] As described in the published patent CN117291925B, "A Textile Surface Defect Detection System Based on Image Features," fabrics produced on textile production lines may have defects during the production process, affecting product quality. To ensure the quality of fabric products, inspection is necessary. Existing technologies typically utilize structuring elements with fixed values ​​for morphological operations.

[0003] As described in the published patent CN116309324A, "A Method for Detecting Defects in Textiles," textiles are a traditional manufacturing industry, and improving the development level of the textile industry is beneficial to my country's economic development. Defect detection can improve the quality of textiles and help enterprises build a good reputation. Defects in textiles can cause economic losses to textile manufacturers. Due to subjective factors, manual inspection is prone to false detections and has low accuracy. It requires highly trained workers to complete defect detection, which increases the economic and labor costs for textile enterprises and is unsuitable for large-scale textile production. Introducing a computer-based textile defect detection system can solve these problems, reducing costs while ensuring a high success rate. Therefore, it is necessary to research stable and efficient methods for detecting textile defects.

[0004] In the field of textile production and quality inspection, fabrics with pile (such as velvet, corduroy, blankets, etc.) are highly favored due to their unique feel and appearance. However, there is a long-standing technical problem that has not been effectively solved in the final automated visual inspection process for these fabrics.

[0005] The root of this problem lies in the presence of numerous randomly oriented fibers on the fabric surface. When light shines on it, these fibers, due to their different orientations, reflect light anisotropically, resulting in noticeable, random color differences or variations in brightness as perceived by the human eye or camera sensors. This optical phenomenon caused by physical geometry is known as "fiber tropism." It is important to emphasize that this color difference is a normal physical characteristic of the fabric itself and not a product quality defect (such as stains or weaving flaws).

[0006] However, when using machine vision for automated defect detection, this ubiquitous and random "fuzzy directional color difference" poses a serious obstacle. Traditional defect detection algorithms typically set thresholds for anomaly detection based on a single dimension or simple combination of color, texture, or brightness. In this case, a dark or light patch with strong contrast to its surrounding area, caused by the fuzz pointing in a specific direction, is easily misjudged by the algorithm as a stain or blemish (i.e., a "false alarm"); conversely, a real defect with low contrast may be lost in the complex background of fuzzy color difference and go undetected (i.e., a "false negative"). The fundamental reason is that traditional techniques cannot fundamentally distinguish between "normal optical changes caused by geometric direction" and "abnormal optical changes caused by contaminants or structural defects."

[0007] To alleviate this problem, existing technologies have attempted methods such as multi-angle illumination, high-resolution imaging, and even spectral analysis. However, multi-angle illumination can only partially reduce shadows and cannot quantify the directional properties of the hair; high-resolution imaging magnifies hair details but also exacerbates the complexity of background texture; while spectral analysis can distinguish colors, it is powerless for hair of the same color but opposite orientation. None of these methods address the essence of the problem—the directional reflective properties of the hair.

[0008] Therefore, there is an urgent need in this field for a new detection method that can fundamentally and effectively distinguish between "color difference caused by different directions of pile tilting" and "real defects", thereby significantly improving the accuracy and reliability of automatic quality inspection of fabrics with random pile. Summary of the Invention

[0009] To overcome the shortcomings mentioned above, the present invention aims to provide a technical solution that can solve the above problems.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A method for testing textile fabrics includes the following steps:

[0012] S100: Polarization image sequence acquisition step: Illuminate the textile fabric under test with a linearly polarized light source rotating around the optical axis, and acquire multiple fabric images at different polarization angles of the light source under orthogonal polarization camera configuration to form a polarization image sequence.

[0013] S200: Image registration and pixel alignment step, registering all images in the polarization image sequence to achieve pixel-level alignment;

[0014] S300: Pixel-level polarization response feature extraction step. For each aligned pixel in the sequence, its modulation depth and dominant polarization angle are calculated based on the change in brightness value in images at different polarization angles.

[0015] S400: Defect area identification and segmentation step, by identifying pixel clusters whose modulation depth is lower than the first preset threshold and whose dominant polarization angle deviates from the dominant polarization angle trend of the surrounding area, the real defect area is segmented.

[0016] S500: Defect classification and output steps, classify the segmented defect areas, and output the detection results.

[0017] As a further aspect of the present invention: the S100 polarization image sequence acquisition step specifically includes:

[0018] S110: Provides an illumination unit, which includes a linearly polarized light source and a rotation controller that drives the polarizer of the light source to rotate in steps around the optical axis.

[0019] S120: An imaging unit is provided, which includes a camera and an analyzer fixedly mounted in front of the camera lens, wherein the polarization direction of the analyzer is orthogonal to the initial polarization direction of the linearly polarized light source.

[0020] S130: The rotation controller controls the polarizer of the linearly polarized light source to rotate in equal angular steps within a range of 0° to 180°, and at each step angle position, the imaging unit acquires a frame of fabric image, thereby obtaining a polarized image sequence containing N images, where N is an integer greater than or equal to 6.

[0021] As a further aspect of the present invention, the S200 image registration and pixel alignment step specifically includes:

[0022] S210: Select a reference image from the polarization image sequence;

[0023] S220: Identify multiple common feature reference points between the reference image and other images in the sequence that are insensitive to changes in polarization angle;

[0024] S230: Based on the feature reference points, calculate the geometric transformation matrix from each image to be registered to the reference image;

[0025] S240: Based on the geometric transformation matrix, perform geometric transformation and pixel resampling on each image to be registered, so that the entire polarization image sequence achieves sub-pixel level alignment accuracy.

[0026] As a further aspect of the present invention: the S300 pixel-level polarization response feature extraction step specifically includes:

[0027] S310: For each pixel position in the registered polarized image sequence, extract its brightness value at different polarization angles to form a discrete brightness sequence.

[0028] S320: Based on the discrete brightness sequence, calculate the modulation depth of the pixel, where the modulation depth is the difference between the maximum and minimum values ​​of the brightness sequence, or twice the amplitude value obtained after fitting the brightness sequence with a sine curve.

[0029] S330: Based on the discrete brightness sequence, calculate the dominant polarization angle of the pixel. The dominant polarization angle is the polarization light source angle corresponding to the brightness sequence reaching its maximum value, or it is the angle obtained by converting the phase angle after fitting the brightness sequence with a sine curve.

[0030] S340: Traverse all pixel locations to generate a modulation depth feature map and a dominant polarization angle feature map for the entire image.

[0031] As a further aspect of the present invention, the S400 defect area identification and segmentation step specifically includes:

[0032] S410: Modulation depth threshold segmentation sub-step, extracting all pixels in the modulation depth feature map whose modulation depth value is lower than the first preset threshold to form the first candidate defect region;

[0033] S420: Dominant polarization angle consistency verification sub-step: For each pixel in the first candidate defect region, calculate the difference between its dominant polarization angle and the statistical value of the dominant polarization angle of the pixels in the surrounding neighborhood, and mark the pixels with a difference value greater than the second preset threshold as the second candidate defect region.

[0034] S430: The logic fusion and region generation sub-step determines the pixels that belong to both the first candidate defect region and the second candidate defect region as real defect pixels, and clusters the spatially adjacent real defect pixels to generate the final defect region segmentation map.

[0035] As a further aspect of the present invention: the S500 defect classification and output step specifically includes:

[0036] S510: Multimodal feature extraction sub-step: For each defect region segmented in S400, extract a combined feature vector containing polarization invariant features, polarization modulation features, and polarization angle field features from the polarization image sequence, modulation depth feature map, and dominant polarization angle feature map.

[0037] S520: Defect classification sub-step, matching the combined feature vector with a predefined defect type feature rule library, or inputting it into a pre-trained defect classification model, thereby determining the category of the defect region;

[0038] S530: Structured result output sub-step, generating and outputting an inspection report, the inspection report including a visual image marked with classified defect areas and their categories, and a structured data list containing information on each defect category, location, and area.

[0039] A textile fabric with a controllable orientation pattern, the fabric being prepared based on the method described above, the preparation of the fabric comprising the following steps:

[0040] S700: The step of preparing the base layer and the directional adhesive layer involves coating a layer of thermoplastic polymer adhesive on the surface of the base layer to form an uncured adhesive layer, and applying a patterned field force template to the adhesive layer, wherein different regions of the template have different field force directions.

[0041] S800: Electrostatic flocking and directional curing steps, electrostatic flocking is performed while the patterned field force template is applied, followed by partial curing to make the flock deflect in the preset direction of the template under the action of the field force, and then complete curing to permanently lock the flock orientation.

[0042] S900: Post-processing and finishing steps, cleaning and finishing the cured fabric to obtain a finished fabric with a preset pile orientation pattern.

[0043] As a further aspect of the present invention: the patterned field force template in step S700 is a patterned electric field template, which is composed of an array of microelectrodes with different spatial orientations embedded on an insulating substrate; the arrangement pattern of the microelectrode array corresponds to the final desired fabric visual pattern.

[0044] As a further aspect of the present invention: the electrostatic flocking in step S800 specifically involves: while applying an electric field to the patterned electric field template, charging the short fiber flocking and vertically implanting it into the uncured adhesive layer;

[0045] The partial curing in step S800 specifically involves: under the condition of maintaining the electric field, a first-stage low-temperature curing is adopted, and the curing temperature is controlled at the critical temperature range between the thermoplastic polymer adhesive and the viscous flow state and the high elastic state, so that the fibers can be deflected under the action of electric field torque, and at the same time, their roots are initially anchored.

[0046] The complete curing in step S800 specifically involves: after removing the electric field template, performing a second stage of high-temperature curing, with the curing temperature higher than the flow temperature of the adhesive, to fully cross-link or harden it, and permanently fix the orientation of the fluff.

[0047] As a further aspect of the present invention: the finishing process in step S900 includes at least a dyeing process, which utilizes the differences in dye adsorption and color development of different micro-regions with different velvet orientations to enhance the visual contrast effect of the preset velvet orientation pattern.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] By acquiring polarized image sequences, this method utilizes a rotating linearly polarized light source and an orthogonal polarization camera to capture reflection images of the fabric at different polarization angles, thus comprehensively recording the directional reflection characteristics of the pile. Subsequently, image registration and pixel alignment ensure spatial consistency of the image sequence, laying the foundation for pixel-level analysis. In pixel-level polarization response feature extraction, the modulation depth and dominant polarization angle of each pixel are calculated to quantify its polarization response, thereby distinguishing between normal brightness variations caused by the geometric direction of the pile and abnormal responses of real defects. Based on this, in defect area identification and segmentation, pixel clusters with low modulation depth and dominant polarization angles deviating from the surrounding trend are detected to effectively isolate real defect areas and avoid interference from directional color differences in the pile. Finally, in defect classification and output, defect type identification and result output are performed. This scheme fundamentally solves the problem of false alarms and missed alarms caused by directional color differences in the pile, separating normal optical changes from defects through polarization response features, significantly improving the accuracy and reliability of automatic quality inspection of piled textile fabrics. Attached Figure Description

[0050] Figure 1 This is a three-dimensional structural view of the detection method in this invention;

[0051] Figure 2 This is a schematic diagram of the optical path of the detection method in this invention;

[0052] Figure 3 This is a flowchart of steps S100-S500 in this invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see Figure 1-3 A method for testing textile fabrics includes the following steps:

[0055] S100: Polarization image sequence acquisition step: Illuminate the textile fabric under test with a linearly polarized light source rotating around the optical axis, and acquire multiple fabric images at different polarization angles of the light source under orthogonal polarization camera configuration to form a polarization image sequence.

[0056] S200: Image registration and pixel alignment step, registering all images in the polarization image sequence to achieve pixel-level alignment;

[0057] S300: Pixel-level polarization response feature extraction step. For each aligned pixel in the sequence, its modulation depth and dominant polarization angle are calculated based on the change in brightness value in images at different polarization angles.

[0058] First, in fundamentally distinguishing between interference and defects, this scheme achieves a fundamental deconstruction of the optical signal by extracting two physical features of each pixel: modulation depth and dominant polarization angle. Modulation depth reflects the material's ability to modulate polarized light at that point. Real stains, oil stains, or structural defects usually change the optical properties of the fabric surface, resulting in a significant reduction in its modulation ability (i.e., low modulation depth). The dominant polarization angle directly encodes the tilting direction of the pile at that point. Therefore, the algorithm no longer searches for simple differences in brightness, but rather for pixels with "abnormal modulation behavior" (suspected defects) and "direction that is incompatible with the surrounding pile" (excluding normal directional color difference). This dual criterion fundamentally distinguishes between normal areas that "appear dark due to different directions" and defective areas that "appear dark due to inherent problems," thereby eliminating the main source of false alarms—"pile directional color difference"—in principle and highlighting real defects with low contrast but abnormal modulation behavior, significantly reducing the false alarm rate.

[0059] Secondly, in terms of improving the robustness and stability of detection, traditional methods are greatly affected by factors such as ambient light and slight color differences in the fabric. However, this method is based on a relatively stable polarization reflection physical model, which is not sensitive to changes in absolute light intensity. More importantly, it does not require pre-learning or defining the complex patterns of "normal" pile texture, because the directional distribution of random pile itself is reflected in the continuous change of the dominant polarization angle. By identifying abnormal clusters that "deviate from the surrounding trend", the algorithm can adaptively cope with various random texture backgrounds, making the detection process have good generalization ability for different types of pile fabrics (such as velvet and corduroy), without having to spend a lot of time re-adjusting the threshold for each new fabric.

[0060] S400: Defect area identification and segmentation step, by identifying pixel clusters whose modulation depth is lower than the first preset threshold and whose dominant polarization angle deviates from the dominant polarization angle trend of the surrounding area, the real defect area is segmented.

[0061] S500: Defect classification and output steps, classify the segmented defect areas, and output the detection results;

[0062] Meanwhile, in achieving accurate pixel-level segmentation, this scheme benefits from the preceding image registration and pixel-level feature extraction steps. Since the sequence images have already achieved pixel-level alignment, all subsequent calculations are performed on the time series of each independent pixel. This ensures the accuracy of feature extraction, making the final identified defect boundary location extremely accurate, reaching the pixel level. This high-precision segmentation result provides higher quality and purer regional feature data for subsequent defect classification (such as distinguishing between longitudinal defects, latitudinal defects, stains, etc.), thereby indirectly improving the accuracy of the final classification.

[0063] By acquiring polarized image sequences (S100), the reflective images of the fabric at different polarization angles are captured using a rotating linearly polarized light source and an orthogonal polarization camera, thus comprehensively recording the directional reflective characteristics of the pile. Subsequently, image registration and pixel alignment (S200) ensure the spatial consistency of the sequence images, laying the foundation for pixel-level analysis. In pixel-level polarization response feature extraction (S300), the modulation depth and dominant polarization angle of each pixel are calculated to quantify its polarization response, thereby distinguishing between normal brightness changes caused by the geometric direction of the pile and abnormal responses of real defects. Based on this, in defect area identification and segmentation (S400), pixel clusters with low modulation depth and dominant polarization angles deviating from the surrounding trend are detected to effectively isolate real defect areas and avoid interference from directional color differences in the pile. Finally, in defect classification and output (S500), defect type identification and result output are performed. This scheme solves the problem of false alarms and missed alarms caused by directional color differences in the pile in principle. By separating normal optical changes from defects through polarization response features, it significantly improves the accuracy and reliability of automatic quality inspection of piled textile fabrics.

[0064] In this embodiment of the invention, the S100 polarization image sequence acquisition step specifically includes:

[0065] S110: Provides an illumination unit, which includes a linearly polarized light source and a rotation controller that drives the polarizer of the light source to rotate in steps around the optical axis.

[0066] S120: An imaging unit is provided, which includes a camera and an analyzer fixedly mounted in front of the camera lens, wherein the polarization direction of the analyzer is orthogonal to the initial polarization direction of the linearly polarized light source.

[0067] S130: The polarizer of the linearly polarized light source is controlled to rotate in equal angular steps within a range of 0° to 180° by the rotation controller, and at each step angle position, the imaging unit acquires a frame of fabric image, thereby obtaining a polarized image sequence containing N images, where N is an integer greater than or equal to 6.

[0068] The orthogonal polarization configuration (S120) constitutes a physical filtering system. After the linearly polarized light source illuminates the fabric, its reflected light contains two main components: scattered light whose polarization direction has changed due to the modulation of the pile surface geometry (i.e., inverted direction), and direct specular reflection light whose polarization direction remains basically unchanged according to Fresnel's law. The fixedly installed analyzer, which is orthogonal to the initial polarization direction of the light source, acts like a "valve" to effectively block most of the strong specular reflection light without direction information, while allowing the scattered light whose polarization direction has been "twisted" by the pile to pass through. This mechanism significantly suppresses highlight overexposure and greatly enhances the contrast and signal-to-noise ratio of the surface micro-geometric information determined by the directionality of the pile, providing a clean input signal for subsequent feature extraction.

[0069] The design of rotating at equal angles (0° to 180°) and acquiring N≥6 images (S130) ensures that the periodic response curve of each pixel as a function of polarization angle (i.e., the Mueller ellipse) can be completely and uniformly sampled. This not only theoretically meets the minimum amount of data required for accurate calculation of the two core physical features, modulation depth and dominant polarization angle (according to the sampling theorem), but more importantly, sufficient sample points (N≥6) can effectively smooth the influence of random noise. Curve fitting greatly improves the calculation accuracy and robustness of feature parameters. In contrast, too few sampling points (such as only four images at 0°, 45°, 90°, and 135°) will introduce calculation errors due to insufficient sampling and have a worse ability to resist noise. Therefore, this design directly determines the accuracy of subsequent defect identification and segmentation.

[0070] By employing an actively rotating linearly polarized light source (S110) combined with programmed step acquisition, an active and closed sensing environment is constructed. This system does not rely on fluctuating ambient light but actively emits "probes" through its own controllable light source, thereby completely eliminating the interference of ambient light fluctuations on the stability of the detection results. At the same time, the standardized process of equal-angle stepping ensures that data collected from different batches and at different times have high consistency and comparability, laying the foundation for achieving stable and repeatable industrial automated detection.

[0071] In this embodiment of the invention, the S200 image registration and pixel alignment step specifically includes:

[0072] S210: Select a reference image from the polarization image sequence;

[0073] S220: Identify multiple common feature reference points between the reference image and other images in the sequence that are insensitive to changes in polarization angle;

[0074] S230: Based on the feature reference points, calculate the geometric transformation matrix from each image to be registered to the reference image;

[0075] S240: Based on the geometric transformation matrix, perform geometric transformation and pixel resampling on each image to be registered, so that the entire polarization image sequence achieves sub-pixel level alignment accuracy.

[0076] The brightness variation of the same physical point under different polarization angles is analyzed. However, during the physical acquisition process, although the equipment is fixed, there may still be slight mechanical vibrations, thermal expansion and contraction, or slight displacement of the camera / sample stage. These factors can cause sub-pixel or even several-pixel deviations in the pixel position of the same object between different images in the sequence. If registration is not performed, the "modulation depth" and "dominant polarization angle" calculated later will be mixed information of misaligned pixels, and its feature clarity and accuracy will be severely reduced, just like trying to compare different points on multiple misaligned targets. This step ensures that in the subsequent analysis, for any pixel coordinate (such as (x, y)) on the image, the brightness value under all polarization angles strictly corresponds to the same micro-region on the fabric, thereby ensuring that the polarization response curve truly reflects the physical properties of the point.

[0077] Step S220 explicitly states that it identifies "multiple feature reference points that are insensitive to changes in polarization angle." This is a crucial and ingenious design because the brightness of the fabric surface changes drastically with the polarization angle in a polarization sequence. If traditional feature points that depend on brightness or contrast (such as SIFT, ORB, etc.) are used, the description of these feature points in different images of the sequence will be extremely unstable, or even appear and disappear intermittently, leading to registration failure or insufficient accuracy. This method ensures that these reference points are consistent and repeatable in all images by finding spatial structural features that are not affected by the polarization angle (e.g., nodes of the fabric weave structure, edges of fixed defects, or intersections of specific textures), thus providing a reliable basis for calculating stable and accurate geometric transformation matrices.

[0078] The "sub-pixel level" alignment accuracy pursued in step S240 is key to distinguishing between tiny real defects and registration error artifacts. If the registration accuracy is only at the pixel level, then in the edge area of ​​the defect, because the information of multiple physical points is averaged into one pixel, the calculated "modulation depth" and "dominant polarization angle" will become blurred. This will not only blur the boundary of the real defect, making the segmentation inaccurate, but may also generate pixels with abnormal feature values ​​at these boundaries, which will be misjudged as defects (i.e., false alarms). After achieving sub-pixel level alignment, the feature value of each pixel represents purer and more accurate physical point information. This allows the system to more clearly define the boundary between defects and normal areas, significantly improving the detection capability of tiny, low-contrast defects (reducing false alarms), while avoiding false defects introduced by poor registration (reducing false alarms).

[0079] This automated registration process effectively compensates for the minute deformations and displacements that may occur in the imaging system during long-term operation, reducing the stringent requirements for the absolute stability of the mechanical structure. This enables the detection scheme to adapt to complex environments such as slight vibrations that may exist in industrial sites, enhancing the practicality and repeatability of the method and laying the foundation for industrialization.

[0080] In this embodiment of the invention, the S300 pixel-level polarization response feature extraction step specifically includes:

[0081] S310: For each pixel position in the registered polarized image sequence, extract its brightness value at different polarization angles to form a discrete brightness sequence.

[0082] S320: Based on the discrete brightness sequence, calculate the modulation depth of the pixel, where the modulation depth is the difference between the maximum and minimum values ​​of the brightness sequence, or twice the amplitude value obtained after fitting the brightness sequence with a sine curve.

[0083] S330: Based on the discrete brightness sequence, calculate the dominant polarization angle of the pixel. The dominant polarization angle is the polarization light source angle corresponding to the brightness sequence reaching its maximum value, or it is the angle obtained by converting the phase angle after fitting the brightness sequence with a sine curve.

[0084] S340: Traverse all pixel positions to generate a modulation depth feature map and a dominant polarization angle feature map for the entire image;

[0085] In step S320, the modulation depth is not a simple brightness value, but rather a quantification of the material's ability to modulate polarized light. A healthy, structurally complete velvet dot will have its surface micro-geometry effectively modulating polarized light, causing the brightness to change strongly and periodically with the rotation of the light source, thus exhibiting a high modulation depth. Conversely, an area soaked in oil (altering the surface reflection characteristics), an area with a fabric defect (lacking the normal velvet structure), or a foreign object attachment point will have a significantly reduced ability to modulate light, resulting in a smaller fluctuation range in its brightness sequence, i.e., exhibiting a low modulation depth. Therefore, the modulation depth feature map effectively filters out visual interference caused by absolute brightness differences and directly highlights those areas with "abnormal optical behavior."

[0086] In step S330, the dominant polarization angle directly maps the microscopic tilting direction of the fluff at each pixel. On normal fabric, although the tilting direction of the fluff is random, its direction is continuous and gradually changing in a local small area. This will be represented as a smooth and continuously changing vector field in the dominant polarization angle feature map. A real defect (such as an abnormal fluff tilting direction caused by a scratch) will break this continuous trend and form a boundary with a significant abrupt change in the dominant polarization angle compared to the surrounding area. This allows the algorithm to detect defects that are abnormal in physical structure orientation even if the brightness or color is the same as the surrounding area.

[0087] The direct calculation based on extrema (maximum value - minimum value) and the fitting calculation based on sine curves offer flexibility and robustness. The direct calculation method is fast and suitable for scenarios with high signal-to-noise ratios. The sine curve fitting method is more advanced and powerful because it is based on an ideal polarization physics model. Through fitting, it can smooth out random noise in the original brightness sequence, thereby obtaining more accurate and stable amplitude (for calculating modulation depth) and phase (for calculating dominant polarization angle) estimates. This fitting method is particularly suitable for detecting low-contrast defects or in noisy industrial environments. It can effectively prevent feature calculation errors caused by individual noise points.

[0088] The resulting modulation depth feature map and dominant polarization angle feature map (S340) constitute a complementary and information-redundant feature set, which enables subsequent defect segmentation algorithms to perform complex logical judgments:

[0089] If a region shows a low value on the modulation depth map (abnormal optical behavior) and exhibits significant discontinuity with its surrounding regions on its dominant polarization angle map (abnormal physical structure orientation), then it has a very high confidence level in being classified as a genuine defect. Conversely, a region that appears dark in a normal brightness image but shows normal behavior on the modulation depth map (indicating its darkening is solely due to the orientation of its fuzz, but its optical modulation behavior is healthy) and whose dominant polarization angle map aligns with its surrounding regions, can be safely excluded as a defect. This effectively solves the false alarm problem caused by "fuzz directional chromatic aberration." This dual-feature mechanism enables the system to detect various types of defects. For example, oil stains may primarily manifest as a decrease in modulation depth with little change in direction; while a scratch may simultaneously cause a decrease in modulation depth and a drastic change in the dominant polarization angle. By condensing the information of the entire image sequence into two global feature maps, the computational burden of subsequent defect recognition and segmentation algorithms is greatly simplified. The algorithm no longer needs to repeatedly access and process the entire massive image sequence, but only needs to operate on these two feature maps containing core physical information, significantly improving processing efficiency and making real-time online detection possible.

[0090] In this embodiment of the invention, the S400 defect area identification and segmentation step specifically includes:

[0091] S410: Modulation depth threshold segmentation sub-step, extracting all pixels in the modulation depth feature map whose modulation depth value is lower than the first preset threshold to form the first candidate defect region;

[0092] S420: Dominant polarization angle consistency verification sub-step: For each pixel in the first candidate defect region, calculate the difference between its dominant polarization angle and the statistical value of the dominant polarization angle of the pixels in the surrounding neighborhood, and mark the pixels with a difference value greater than the second preset threshold as the second candidate defect region.

[0093] S430: The logic fusion and region generation sub-step determines the pixels that belong to both the first candidate defect region and the second candidate defect region as real defect pixels, and clusters the spatially adjacent real defect pixels to generate the final defect region segmentation map.

[0094] The modulation depth threshold segmentation first performs a "screening" process. It is based on the physical principle that "any real defect will, to some extent, disrupt the normal optical modulation capability of the fabric surface." This step is designed to be highly sensitive, and its "first preset threshold" can be set relatively loosely to ensure that all possible abnormal points can be captured, including those small potential defects with low contrast. However, this inevitably introduces a lot of "noise," which refers to normal areas that are not defects but happen to have a low modulation depth due to various reasons (such as dark fuzz, shadows, or noise).

[0095] The dominant polarization angle consistency check plays the role of a "precision filter". It is based on another independent physical principle: "real defects will disrupt the spatial continuity of the local fiber orientation field". For a normal area that appears dark (low modulation depth) due to the fiber falling in a certain direction, its dominant polarization angle is consistent with or continuously changes with the direction of the surrounding fiber. Therefore, it can be excluded by this check. Conversely, a real stain or structural damage will have a significant break or abrupt change in its orientation field compared with the surrounding healthy fiber orientation field. Step S420 accurately captures this abrupt change by calculating the difference between the pixel and the surrounding orientation statistics.

[0096] The final decision criterion is AND logic—a pixel must belong to both the first and second candidate regions to be considered a true defect. This design ensures that only regions exhibiting both "optical behavior anomalies" (low modulation depth) and "geometric structure anomalies" (inconsistent orientation) are ultimately confirmed. This is akin to having two independent experts conduct a joint review: one checks for "color / brightness anomalies," and the other checks for "texture / orientation anomalies." Both experts must vote in favor for a region to be considered a defect. This significantly improves the specificity of the decision at the algorithm level, reducing "fuzzy directional color difference," the primary source of false alarms, to a low level. It also achieves high sensitivity detection of low-contrast defects in complex backgrounds. This method effectively prevents false alarms. Traditional single-threshold methods often have to increase the threshold to control false alarms when dealing with complex velvety textures. This directly leads to many weak, low-contrast real defects being submerged in the background. The dual criteria of this method provide higher fault tolerance. Since the modulation depth threshold in the first step can be set lower to ensure sensitivity, even if the modulation depth of a defect does not drop drastically, as long as it can be captured in the first step and its orientation field anomaly features are obvious enough, it can still be identified and finally segmented in the second step. This means that the system does not need to sacrifice sensitivity for a low false alarm rate. It can discover those "hidden" defects that are almost invisible in ordinary images but whose physical properties have changed.

[0097] The clustering operation in S430 groups spatially adjacent pixels identified as defects together to form connected and complete defect regions. Since the preceding steps are based on pixel-level judgments using precise physical features, the boundaries of the generated defect regions closely match the physical boundaries of the actual defects, achieving pixel-level or sub-pixel-level positioning accuracy. This high-quality "defect region segmentation map" is crucial for the subsequent defect classification and output steps in S500. The classification algorithm (such as a convolutional neural network or shape analyzer) will receive clean defect image patches with background interference largely eliminated. This significantly improves the accuracy and reliability of classification because the classifier no longer needs to learn and resist complex fuzzy background variations and can focus more on the shape, size, and texture features of the defects themselves.

[0098] In this embodiment of the invention, the S500 defect classification and output step specifically includes:

[0099] S510: Multimodal feature extraction sub-step: For each defect region segmented in S400, extract a combined feature vector containing polarization invariant features, polarization modulation features, and polarization angle field features from the polarization image sequence, modulation depth feature map, and dominant polarization angle feature map.

[0100] S520: Defect classification sub-step, matching the combined feature vector with a predefined defect type feature rule library, or inputting it into a pre-trained defect classification model, thereby determining the category of the defect region;

[0101] S530: Structured result output sub-step, generating and outputting an inspection report, the inspection report including a visual image marked with classified defect areas and their categories, and a structured data list containing information on each defect category, location, and area;

[0102] Step S510 does not extract features from a single image, but rather extracts polarization-invariant features (such as stable colors or textures at certain polarization angles), polarization modulation features (such as the statistical distribution of modulation depth), and polarization angle field features (such as the directional consistency of the interior and edges of defects) jointly from three complementary data sources: polarization image sequence, modulation depth feature map, and dominant polarization angle feature map.

[0103] Traditional RGB images may be difficult to distinguish between "oil stains" and "water stains" because their colors and brightness may be similar. However, in the polarization multi-modal feature space, oil stains (which usually change the surface chemistry and strongly inhibit polarization modulation) may exhibit extremely low modulation depth characteristics, while the modulation depth of water stains (which may mainly change the light path but not completely destroy the villus structure) may not decrease as significantly. However, their polarization angle field characteristics may呈现特定模式due to the change in the倒伏方向 of the villi caused by the surface tension of water; The combined feature vector provides these key physical discrimination bases that cannot be directly perceived by the human eye for the classifier;

[0104] Even for the same type of defect (such as "broken warp"), its appearance may be different under different fabrics or different lighting conditions, but the pattern of changes in its physical properties is relatively stable (for example, it will all cause a linear area with a sharp drop in modulation depth and a broken direction field). By learning and matching these deeper physical feature patterns, the classification system is more robust to lighting changes and fabric color changes, and its generalization ability is much better than that of classifiers based on appearance features;

[0105] Step S520 provides two classification paths: matching with a pre-defined defect type feature rule base, or inputting it into a pre-trained defect classification model (such as a machine learning or deep learning model). It is applicable to known defect types with clear physical mechanisms. Its advantages are high transparency and strong interpretability. Engineers can directly set rules based on physical knowledge (for example: "IF the average modulation depth < T1 AND the aspect ratio of the area > T2 THEN classify as'scratch'"), which is very beneficial for initial verification and rapid deployment in specific scenarios; It is applicable to complex, diverse or difficult-to-describe defects with simple rules. Through training (such as using support vector machines or convolutional neural networks), the model can automatically learn and discover the complex decision boundaries of different defects in the multi-modal feature space, with stronger processing capabilities, especially擅长区分那些特征交织在一起的、难以言喻的瑕疵类型; This dual-path design gives the system great flexibility and scalability. Users can choose the most suitable classification strategy according to their own data accumulation, technical reserves and specific needs, or use both in combination;

[0106] The output of step S530 is not a simple "qualified / unqualified" signal, but a complete detection report containing a visual image and a list of structured data. The visual image (marking the areas and categories of classified defects) provides the most intuitive interpretation basis for on-site operators without professional knowledge, facilitating rapid confirmation and intervention, and improving the human-machine interaction efficiency;

[0107] Note: In the translation of the part marked "呈现特定模式" and "倒伏方向", the specific expressions may need to be adjusted according to the actual context and accurate physical terms. Here, approximate translations are used for the time. And for some special tags like , they are kept as they are according to the requirement.Structured data lists (containing defect categories, precise location coordinates, pixel areas, etc.) are the core of achieving intelligent manufacturing and quality traceability. This data can: drive automated actuators, such as linking marking machines to mark defect points, or control robotic arms for sorting; perform statistical analysis to generate quality reports (such as the number of defects per 100 meters, the proportion of each type of defect, and heat maps of defect location distribution), helping managers to identify production bottlenecks and track process problems (for example, a certain defect always appears in a specific location, which may point to a fault in a certain loom); and be stored in a database to achieve full lifecycle quality traceability for each roll of fabric.

[0108] A textile fabric with a controllable orientation pattern, the fabric being prepared based on the above-described method, comprising the following steps:

[0109] S700: The step of preparing the base layer and the directional adhesive layer involves coating a layer of thermoplastic polymer adhesive on the surface of the base layer to form an uncured adhesive layer, and applying a patterned field force template to the adhesive layer, wherein different regions of the template have different field force directions.

[0110] S800: Electrostatic flocking and directional curing steps, electrostatic flocking is performed while the patterned field force template is applied, followed by partial curing to make the flock deflect in the preset direction of the template under the action of the field force, and then complete curing to permanently lock the flock orientation.

[0111] S900: Post-processing and finishing steps, cleaning and finishing the cured fabric to obtain a finished fabric with a preset pile orientation pattern.

[0112] In step S600, a base layer and an oriented adhesive layer are prepared. A textile fabric is provided as the base layer, and a layer of thermoplastic polymer adhesive is coated on the surface of the base layer to form an uncured adhesive layer. A pre-designed patterned electric field template or magnetic field template is applied at close range to the surface of the uncured adhesive layer. The template is designed according to the final desired fabric visual pattern, and different regions of the template have different electric field directions or magnetic field directions.

[0113] In step 700, electrostatic flocking and directional curing are performed. While applying a field force to the patterned template, the electrostatic flocking process is carried out, so that short fiber fluffs are vertically implanted into the uncured adhesive layer. While maintaining the function of the patterned template, the flocked fabric undergoes a first-stage curing process. The curing temperature is such that the adhesive layer is partially cross-linked but still retains a certain degree of plasticity. In this stage, under the action of the field force (such as electric field torque) applied by the template, the roots of the implanted fluffs are fixed, but the upper part is deflected, so that the tilting direction is aligned with the field force direction of the micro-region.

[0114] In step 800, the patterned template is removed, and the second stage of complete high-temperature curing is carried out to harden the adhesive layer completely, permanently locking the orientation of the pile. The cured oriented pile fabric is cleaned to remove loose fibers that are not firmly bonded. As needed, finishing processes such as dyeing, waterproofing, or antistatic are carried out. Because the pile direction is consistent, the dyeing effect of different micro areas will produce regular color differences, forming a unique visual pattern.

[0115] Traditional flocked fabrics are limited to a single color and a velvety texture in terms of visual expression. However, this method uses a patterned field force template to pre-draw different field force directions on the adhesive layer, much like printing. During the flocking and curing process, the flock fibers in these areas will collectively tilt towards the pre-set direction of the template, just like magnetized iron filings. Since the direction of the flock fibers directly determines their angle of light reflection, flock areas with different orientations will present completely different brightness or color. This allows complex visual art effects such as logic, logos, gradient shadows, and hidden patterns to be created on a single-color fabric without the use of dyes or printing, solely through physical structure. When the observer views the fabric from different angles, the brightness of the pattern will change accordingly, creating a dynamic visual experience.

[0116] The pattern of the field force template can be designed and changed at will, which means that the pile orientation pattern is not limited by the number of colors in traditional printing, and can realize an infinite variety of designs from macro to micro, from geometric to organic.

[0117] The orientation of the fibers not only affects optical properties but also directly determines the directionality of the fabric's physical properties; this method achieves spatial programming of the fabric's surface properties by controlling the field force pattern.

[0118] The capillary force is strongest along the direction of the pile. By designing specific orientation patterns, sweat or liquid can be guided to be transported and diffused quickly on the fabric surface along a preset path, thereby designing sportswear and functional fabrics with efficient one-way moisture wicking, local quick drying or intelligent sweat management functions.

[0119] The fabric feels smooth when touched with the grain, but there is more resistance when touching against the grain. By programming the direction of the pile in different areas, you can create localized tactile feedback on a piece of fabric. For example, you can design specific directions at the collar and cuffs of a garment to increase friction and prevent slipping, or provide a smooth feel on the main body.

[0120] Oriented fibers can control the passage of light and heat like blinds. By designing patterns, they can enhance the absorption of sunlight when needed (with fibers standing upright or facing the light source) or reflect light to achieve cooling when needed, providing a new way to develop smart thermal management textiles.

[0121] The two-step curing strategy (S800) of "partial curing-orientation-full curing" first performs partial curing, giving the adhesive layer sufficient viscosity to temporarily fix the ideal orientation of the flock under the action of electric field force, while maintaining a certain degree of plasticity to allow the flock to make a final precise deflection. The subsequent full curing is like "welding", permanently locking this ideal orientation state to form a stable three-dimensional structure. This step-by-step curing process ensures that the flock can fully and accurately respond to the field force direction of the template, avoiding orientation ambiguity caused by external force or relaxation before full curing, thus ensuring the clear boundaries and high fidelity of the final pattern. Compared with traditional flocking, the oriented flock prepared by this method has its roots firmly anchored in the fully cured adhesive layer. Its orientation is not obtained by post-processing heat pressing, but comes from its inherent three-dimensional structure. Therefore, it has excellent wash resistance, abrasion resistance and fatigue resistance, and the pattern is durable.

[0122] In this embodiment of the invention, the patterned field force template in step S700 is a patterned electric field template, which is composed of a microelectrode array with different spatial orientations embedded on an insulating substrate; the arrangement pattern of the microelectrode array corresponds to the final desired fabric visual pattern.

[0123] This template is essentially an "electric field mask," with an array of microelectrodes embedded on an insulating substrate, each with different spatial orientations. Each microelectrode or combination of microelectrodes acts like an independently addressable "electric field pixel." When energized, each "pixel" generates a local electric field with a consistent orientation and height in the adhesive layer region directly above it.

[0124] Unlike traditional monolithic electrodes that generate a uniform electric field, this microarray structure allows for spatial programming of the electric field direction. This means that different electric field directions can be independently set at different locations on the fabric. For example, in the area where a logo needs to be formed, the electrodes are all oriented upwards; in the background area, the electrodes can be designed to be random or at specific angles. This enables digital, pixel-level control of the pile orientation, allowing for the perfect reproduction of every detail of complex patterns, including clear boundaries, smooth gradient transitions, and minute features—something that no mechanical guiding or airflow method can achieve.

[0125] Traditional methods can only apply a uniform electric field direction to the entire fabric, while this template achieves a qualitative change from "macroscopic uniform control" to "microscopic differentiated control", laying the physical foundation for creating unprecedented visual effects;

[0126] This template uses electrostatic force for non-contact directional control. During flight, the charged flocking fibers are affected by the electric field force generated by the microelectrodes below them, thereby changing their flight attitude and finally implanting into the adhesive layer at a specific angle. Because it is an electric field force, there is no physical contact between the template and the wet adhesive layer and flocking material. This completely avoids the problems of scratching, sticking, and contamination that may be caused by mechanical templates, making it particularly suitable for the production of high-quality fabrics with high surface requirements. The electric field force can be precisely controlled by voltage to ensure that the force applied to each area is uniform. This avoids the local unevenness that may exist in mechanical pressure or airflow, thus ensuring a high degree of consistency and repeatability of the final pattern effect.

[0127] In theory, the template can be made into a curved surface that matches the shape of the fabric, so as to perform directional flocking on a three-dimensional object. At the same time, if the microelectrode array is dynamically programmable (such as a digital micromirror device), there is no need to make a physical template. The flocking pattern can be switched instantly by changing the software program, which makes it possible to realize flexible and intelligent manufacturing of small batches and multiple varieties.

[0128] The arrangement pattern of the microelectrode array directly corresponds to the final required fabric visual pattern, which establishes a direct path from "digital design" to "physical product". Any vector or bitmap completed by the designer on the computer can be converted into corresponding microelectrode orientation layout data and used directly to manufacture the template. This greatly liberates design creativity, and any pattern that can be expressed by directional field can be realized.

[0129] As mentioned earlier, the direction of the pile determines the functional anisotropy. Therefore, this template can not only control the visual pattern, but also accurately "draw" the functional areas. For example, the pile direction pointing outward can be designed in the underarm area of ​​the garment to promote sweat evaporation, while maintaining visual aesthetics in other areas. This makes it possible to integrate functional design and aesthetic design in the same manufacturing step.

[0130] In this embodiment of the invention, the electrostatic flocking in step S800 specifically involves: while applying an electric field to the patterned electric field template, charging the short fiber flocking and vertically implanting it into the uncured adhesive layer.

[0131] The partial curing in step S800 specifically involves: under the condition of maintaining the electric field, a first-stage low-temperature curing is adopted, and the curing temperature is controlled at the critical temperature range between the thermoplastic polymer adhesive and the viscous flow state and the high elastic state, so that the fibers can be deflected under the action of electric field torque, and at the same time, their roots are initially anchored.

[0132] The complete curing in step S800 specifically involves: after removing the electric field template, performing a second stage of high-temperature curing, with the curing temperature higher than the flow temperature of the adhesive, to fully cross-link or harden it and permanently fix the orientation of the fluff.

[0133] Traditional electrostatic flocking methods, which involve immediate curing after flocking, cause the flock fibers to deviate from their intended direction the instant the electric field is removed, due to elasticity, collisions, or external disturbances. The breakthrough of this invention lies in "partial curing while maintaining the electric field." The temperature during the partial curing stage is precisely controlled to keep the adhesive in a critical range between a viscous flow state and a highly elastic state. At this point, the macroscopic fluidity of the adhesive disappears (preventing the flock fibers from sinking or drifting), but the microscopic molecular chain segments still have sufficient mobility. This allows the flock fibers to overcome viscous resistance and slowly and precisely rotate to the optimal orientation that is completely consistent with the direction of the local electric field, much like a compass turning in viscous oil, under the continuous electric field torque. This is a key step of "final fine-tuning in a fixed environment," ensuring that the orientation of each flock fiber is faithful to the template design.

[0134] If even the most basic curing is performed after the electric field is removed, the fibers will collectively loosen their orientation due to their own elasticity and the rebound of the compressed substrate, resulting in blurred patterns and reduced contrast. This method performs preliminary anchoring while maintaining the electric field, which is equivalent to "freezing" the fibers instantly when they are in the "ideal posture", permanently locking their most perfect orientation state, thereby achieving unprecedented pattern sharpness and high fidelity.

[0135] This process employs a two-step curing strategy that separates "preliminary anchoring" and "final reinforcement." In the partial curing (low temperature) stage, the adhesive enters a highly elastic state, and the roots of the pile are initially wrapped and anchored. In the subsequent complete curing (high temperature) stage, the adhesive is fully cross-linked or hardened, forming a robust three-dimensional network structure that completely fixes the roots of the pile like steel bars in concrete. This step-by-step curing results in a much stronger interfacial bonding force than one-step curing, giving the pile excellent pull-out resistance and enabling it to withstand repeated friction, washing, and other physical tests. Since the orientation of the pile is determined before the molecular chains are completely frozen and subsequently fixed by a strong cross-linking network, its directionality is an inherent and irreversible physical property of the fabric. This is completely different from the temporary orientation obtained through finishing processes (such as ironing), ensuring that the product's visual pattern and functionality remain stable throughout its entire life cycle and will not fail due to use and washing.

[0136] By partially overlapping the "orientation" and "curing" processes in time and precisely controlling the physical conditions (temperature) of their transition, this process achieves a wide and easily controllable process window. It cleverly resolves the contradiction between "requiring the adhesive layer to be 'soft' enough to allow the flock to turn" and "requiring the adhesive layer to be 'hard' enough to fix the flock." The partial curing stage achieves a special state of "soft yet firm," allowing orientation while preventing disorder. The well-defined temperature control points (critical temperature, flow temperature) make this process easy to implement on existing flocking production lines by modifying the temperature zone. The parameters are controllable and have good repeatability, providing a practical technical path for the large-scale, stable production of high-quality oriented pattern flocking products.

[0137] In this embodiment of the invention, the finishing process in step S900 includes at least a dyeing process, which utilizes the difference in dye adsorption and color development degree of different micro-regions with different pile orientations to enhance the visual contrast effect of the preset pile orientation pattern.

[0138] Different orientations of the micro-pile regions exhibit systematic differences in their physical structure. For example, upright or nearly upright fibers have a larger effective specific surface area, and their fiber axes are more directly exposed to the dye bath; while flat fibers are the opposite. This results in fibers with different orientations naturally adsorbing different amounts of dye molecules in the same dye bath, forming differences in color depth or hue. The color differences produced by this method do not come from external pigment coverage (such as printing), but rather from the differentiated manifestation of the fiber's own intrinsic ability to adsorb dye. This makes the final pattern color transition extremely natural, possessing a three-dimensional feel and high-end texture similar to "embossing" or "light and shadow carving".

[0139] Before dyeing, the orientation pattern formed in the previous steps may only be vaguely discernible under specific angles of light. After this differentiated dyeing process, the contrast and visibility of the pattern are enhanced, making it clearly visible from any angle. This allows even very fine pattern details to be perfectly presented and preserved, greatly enhancing the aesthetic value and artistic expression of the product. This technology achieves "structural coloring," that is, completing the overall coloring and pattern display simultaneously through a unified dyeing process.

[0140] Traditionally, achieving similar multi-color or gradient effects requires complex and time-consuming multi-color printing processes, expensive jacquard weaving, or subsequent special finishing processes such as burnout and discharge dyeing. This method achieves the same effect with just one dyeing step, eliminating multiple printing, steaming, and washing processes, significantly shortening the production cycle, reducing energy and water consumption, and avoiding the use of large amounts of printing paste, printing backing fabric, and related equipment debugging and maintenance costs. Furthermore, due to the shortened process flow, the defect rate also decreases, resulting in a significant reduction in overall production costs.

[0141] The color difference is based on the physical morphology of the fiber and the dyeing chemistry, rather than a surface coating. Since the color is formed by the chemical bonding between the dye and the fiber, its rubbing resistance, washing fastness and color fastness are completely consistent with ordinary dyed fabrics. It does not have the problems of poor adhesion, stiff hand feel or reduced breathability that are common with printed patterns, thus ensuring wearing comfort. The pattern is an inherent physical property of the fabric and will not crack, peel or become blurred due to long-term use and washing like some printed patterns. The color of the pattern has the same lifespan as the fabric, ensuring the durability of the product's visual effect. By simplifying the process, this method reduces the consumption of water, energy and chemicals, while also reducing the discharge of wastewater (containing a large amount of unfixed dyes and auxiliaries) commonly found in printing processes, which is in line with the industry trend of green manufacturing and sustainable development.

[0142] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for testing textile fabrics, characterized in that, Includes the following steps: S100: Polarization image sequence acquisition step: Illuminate the textile fabric under test with a linearly polarized light source rotating around the optical axis, and acquire multiple fabric images at different polarization angles of the light source under orthogonal polarization camera configuration to form a polarization image sequence. S200: Image registration and pixel alignment step, registering all images in the polarization image sequence to achieve pixel-level alignment; S300: Pixel-level polarization response feature extraction step. For each aligned pixel in the sequence, its modulation depth and dominant polarization angle are calculated based on the change in brightness value in images at different polarization angles. S400: Defect area identification and segmentation step, by identifying pixel clusters whose modulation depth is lower than the first preset threshold and whose dominant polarization angle deviates from the dominant polarization angle trend of the surrounding area, the real defect area is segmented. S500: Defect classification and output steps, classify the segmented defect areas, and output the detection results; The S400 defect area identification and segmentation steps specifically include: S410: Modulation depth threshold segmentation sub-step, extracting all pixels in the modulation depth feature map whose modulation depth value is lower than the first preset threshold to form the first candidate defect region; S420: Dominant polarization angle consistency verification sub-step: For each pixel in the first candidate defect region, calculate the difference between its dominant polarization angle and the statistical value of the dominant polarization angle of the pixels in the surrounding neighborhood, and mark the pixels with a difference value greater than the second preset threshold as the second candidate defect region. S430: The logic fusion and region generation sub-step determines the pixels that belong to both the first candidate defect region and the second candidate defect region as real defect pixels, and clusters the spatially adjacent real defect pixels to generate the final defect region segmentation map.

2. The method for testing textile fabrics according to claim 1, characterized in that, The S100 polarization image sequence acquisition steps specifically include: S110: Provides an illumination unit, which includes a linearly polarized light source and a rotation controller that drives the polarizer of the light source to rotate in steps around the optical axis. S120: An imaging unit is provided, which includes a camera and an analyzer fixedly mounted in front of the camera lens, wherein the polarization direction of the analyzer is orthogonal to the initial polarization direction of the linearly polarized light source. S130: The rotation controller controls the polarizer of the linearly polarized light source to rotate in equal angular steps within a range of 0° to 180°, and at each step angle position, the imaging unit acquires a frame of fabric image, thereby obtaining a polarized image sequence containing N images, where N is an integer greater than or equal to 6.

3. The method for testing textile fabrics according to claim 2, characterized in that, The S200 image registration and pixel alignment steps specifically include: S210: Select a reference image from the polarization image sequence; S220: Identify multiple common feature reference points between the reference image and other images in the sequence that are insensitive to changes in polarization angle; S230: Based on the feature reference points, calculate the geometric transformation matrix from each image to be registered to the reference image; S240: Based on the geometric transformation matrix, perform geometric transformation and pixel resampling on each image to be registered, so that the entire polarization image sequence achieves sub-pixel level alignment accuracy.

4. The method for testing textile fabrics according to claim 3, characterized in that, The S300 pixel-level polarization response feature extraction step specifically includes: S310: For each pixel position in the registered polarized image sequence, extract its brightness value at different polarization angles to form a discrete brightness sequence. S320: Based on the discrete brightness sequence, calculate the modulation depth of the pixel, where the modulation depth is the difference between the maximum and minimum values ​​of the brightness sequence, or twice the amplitude value obtained after fitting the brightness sequence with a sine curve. S330: Based on the discrete brightness sequence, calculate the dominant polarization angle of the pixel. The dominant polarization angle is the polarization light source angle corresponding to the brightness sequence reaching its maximum value, or it is the angle obtained by converting the phase angle after fitting the brightness sequence with a sine curve. S340: Traverse all pixel locations to generate a modulation depth feature map and a dominant polarization angle feature map for the entire image.

5. The method for testing textile fabrics according to claim 4, characterized in that, The S500 defect classification and output steps specifically include: S510: Multimodal feature extraction sub-step: For each defect region segmented in S400, extract a combined feature vector containing polarization invariant features, polarization modulation features, and polarization angle field features from the polarization image sequence, modulation depth feature map, and dominant polarization angle feature map. S520: Defect classification sub-step, matching the combined feature vector with a predefined defect type feature rule library, or inputting it into a pre-trained defect classification model, thereby determining the category of the defect region; S530: Structured result output sub-step, generating and outputting an inspection report, the inspection report including a visual image marked with classified defect areas and their categories, and a structured data list containing information on each defect category, location, and area.

6. A textile fabric with a controllable orientation pattern, characterized in that, The fabric is prepared based on the method according to any one of claims 1 to 5, and the preparation of the fabric includes the following steps: S700: The step of preparing the base layer and the directional adhesive layer involves coating a layer of thermoplastic polymer adhesive on the surface of the base layer to form an uncured adhesive layer, and applying a patterned field force template to the adhesive layer, wherein different regions of the template have different field force directions. S800: Electrostatic flocking and directional curing steps, electrostatic flocking is performed while the patterned field force template is applied, followed by partial curing to make the flock deflect in the preset direction of the template under the action of the field force, and then complete curing to permanently lock the flock orientation. S900: Post-processing and finishing steps, cleaning and finishing the cured fabric to obtain a finished fabric with a preset pile orientation pattern.

7. The textile fabric according to claim 6, characterized in that, The patterned field force template in step S700 is a patterned electric field template, which is composed of an array of microelectrodes with different spatial orientations embedded on an insulating substrate; the arrangement pattern of the microelectrode array corresponds to the final desired fabric visual pattern.

8. The textile fabric according to claim 7, characterized in that, The electrostatic flocking in step S800 specifically involves: while applying an electric field to the patterned electric field template, charging the short fiber flocking and vertically implanting it into the uncured adhesive layer. The partial curing in step S800 specifically involves: under the condition of maintaining the electric field, a first-stage low-temperature curing is adopted, and the curing temperature is controlled at the critical temperature range between the thermoplastic polymer adhesive and the viscous flow state and the high elastic state, so that the fibers can be deflected under the action of electric field torque, and at the same time, their roots are initially anchored. The complete curing in step S800 specifically involves: after removing the electric field template, performing a second stage of high-temperature curing, with the curing temperature higher than the flow temperature of the adhesive, to fully cross-link or harden it, and permanently fix the orientation of the fluff.

9. The textile fabric according to claim 8, characterized in that, The finishing process in step S900 includes at least a dyeing process, which utilizes the differences in dye adsorption and color development of different micro-regions with different pile orientations to enhance the visual contrast effect of the preset pile orientation pattern.

Images