A method and system for detecting defects in textile production based on image processing

Abstract

This application discloses a method and system for detecting defects in textile production based on image processing, belonging to the field of textile defect detection. This application generates point cloud data by collecting surface deformation stripes and acquires penetrating reflection signals containing sound wave propagation time using a multi-frequency ultrasonic probe. After denoising the point cloud data, candidate areas of surface deformation are located based on curvature gradient and normal vector deflection angle; internal abnormal regions are determined by extracting reflected wave amplitude and time delay parameters through ultrasonic frequency band separation. An iterative nearest-point algorithm is used to match surface point cloud coordinates with ultrasonic propagation time to establish a feature correlation between normal vector deflection angle and reflected wave amplitude. Multimodal defect distribution maps are generated by dynamically assigning fusion weights to surface features and internal features based on the rate of change of curvature gradient and amplitude attenuation rate. Geometric shape and reflection intensity parameters are extracted and matched with a preset defect template to achieve type and location determination, making it applicable to the detection of defects in thick textiles with multi-layer composite material structures.

Description

Technical Field

[0001] This application belongs to the field of textile defect detection, and in particular relates to a method and system for detecting textile production defects based on image processing. Background Technology

[0002] Defect detection in textile production is a crucial step in ensuring product quality. With the development of intelligent manufacturing technology, automated inspection technology is gradually replacing traditional manual visual inspection methods. Efficient and accurate defect detection methods can significantly reduce production costs and improve product yield, which is of great significance for the intelligent upgrading of the textile industry.

[0003] Existing detection technologies mostly employ surface defect identification methods based on visible light image analysis. These methods acquire images of the fabric surface using high-resolution cameras and utilize image processing algorithms to identify defects such as texture anomalies, stains, or tears. These methods rely on optical imaging technology and have been widely applied in the detection of surface defects in textiles.

[0004] However, existing methods for detecting manufacturing defects in thick textiles with multi-layered composite structures suffer from limitations because visible light cannot effectively characterize the internal physical properties of the material. Consequently, they cannot accurately detect defects such as internal fiber breakage, interlayer separation, and foreign object embedding. Therefore, existing methods for detecting manufacturing defects in textiles with multi-layered composite structures are unsuitable for this purpose, resulting in poor accuracy. Summary of the Invention

[0005] This application provides a method, system, device, and computer storage medium for detecting defects in textile production based on image processing, which is applicable to the detection of defects in thick textiles with multi-layer composite material structures.

[0006] In a first aspect, this application provides an image processing-based method for detecting defects in textile production, applied to multilayer composite textiles, the method comprising:

[0007] The surface point cloud data and ultrasonic structural image data of the target textile are acquired. The surface point cloud data is generated by acquiring the deformation stripe image of the target textile surface. The ultrasonic structural image data includes sound wave transmission time data and is generated by receiving the penetrating sound wave reflection signal emitted by a multi-frequency ultrasonic probe array.

[0008] Noise point filtering based on density distribution is performed on surface point cloud data, and surface curvature gradient and normal vector deflection angle are extracted to obtain surface deformation candidate regions. Frequency band separation processing is performed on ultrasonic structural image data, and reflected wave amplitude and signal delay values ​​are extracted to obtain internal abnormal signal regions.

[0009] Spatial matching is performed between the point cloud coordinates of the candidate surface deformation region and the acoustic wave propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established by iterative nearest point matching. The normal vector deflection angle and the reflected wave amplitude form a feature pair.

[0010] Anisotropic diffusion filtering and wavelet thresholding denoising are applied to surface morphology features and internal defect features respectively. Fusion weights are assigned based on the ratio of the surface curvature gradient change rate to the attenuation rate of the reflected wave amplitude. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map.

[0011] Based on the multimodal defect distribution map, geometric shape parameters and reflection intensity parameters are extracted, and then matched with the geometric shape parameters and reflection intensity parameters with the preset defect feature template to generate detection results including defect type and defect location annotations.

[0012] In one feasible implementation, anisotropic diffusion filtering and wavelet thresholding denoising are applied to the surface morphology features and internal defect features respectively. Fusion weights are assigned based on the ratio of the surface curvature gradient change rate to the reflected wave amplitude attenuation rate. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map, including:

[0013] Anisotropic diffusion filtering and wavelet thresholding denoising are applied to surface morphology features and internal defect features respectively; and initial fusion weights are assigned based on the ratio of the surface curvature gradient change rate to the attenuation rate of the reflected wave amplitude.

[0014] The defect depth information is determined based on the acoustic wave transmission time data corresponding to the internal defect features, and a weight function with the defect depth information as the exponential factor is constructed. The fusion weight of the surface morphology features is negatively correlated with the defect depth information, while the fusion weight of the internal defect features is positively correlated with the defect depth information.

[0015] The target fusion weight is determined based on the weight function and the initial fusion weight. The target fusion weight includes the curvature gradient weight coefficient and the reflection energy weight coefficient.

[0016] Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused with target fusion weights to obtain a multimodal defect distribution map.

[0017] In one feasible implementation, based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused using target fusion weights to obtain a multimodal defect distribution map, including:

[0018] The feature value of each point in the filtered surface morphology feature is weighted and superimposed with the feature value of the same spatial position in the internal defect feature corresponding to the association mapping relationship to generate a fused feature value. The weight of the surface morphology feature is the curvature gradient weight coefficient, and the weight of the internal defect feature is the reflection energy weight coefficient.

[0019] Based on the spatial location correspondence in the association mapping relationship, the fused feature values ​​are mapped to the three-dimensional spatial location corresponding to the point cloud coordinates of the surface deformation candidate region;

[0020] In three-dimensional space, continuous regions with fused feature values ​​exceeding a preset defect determination threshold are marked as defect regions, generating a multimodal defect distribution map that includes defect region location information.

[0021] In one feasible implementation, noise point filtering based on density distribution is performed on surface point cloud data, and surface curvature gradient and normal vector deflection angle are extracted to obtain candidate regions for surface deformation. Frequency band separation processing is performed on ultrasonic structural image data, and reflected wave amplitude and signal delay values ​​are extracted to obtain internal anomalous signal regions, including:

[0022] Calculate the density distribution statistics of each point based on the number of points in the neighborhood of each point in the surface point cloud data, remove points whose density distribution statistics are lower than the preset density threshold, and generate target point cloud data.

[0023] In the target point cloud data, the surface curvature gradient value is obtained by calculating the three-dimensional coordinate difference between each point and its neighboring points. At the same time, the normal vector deflection angle value is calculated based on the angle between the normal vector direction of each point and the average normal vector direction of the adjacent area.

[0024] In the target point cloud data, continuous point cloud regions whose surface curvature gradient values ​​are greater than a preset curvature gradient threshold and whose normal vector deflection angle values ​​are greater than a preset deflection angle threshold are marked as surface deformation candidate regions.

[0025] The ultrasonic structural image data is decomposed into reflected wave signals in multiple independent frequency bands. The maximum amplitude of the reflected wave signal in each frequency band is calculated, and the sound wave propagation time data corresponding to the maximum amplitude is recorded.

[0026] In the ultrasound structural image data, the region where the maximum amplitude exceeds the preset reflected wave threshold and the difference between the corresponding sound wave propagation time data and the preset standard propagation time data exceeds the preset difference threshold is marked as an internal abnormal signal region.

[0027] In one feasible implementation, before establishing the association mapping between surface morphology features and internal defect features through iterative nearest-point matching, the method includes:

[0028] Based on the surface curvature gradient of the candidate region of surface deformation, the curvature abrupt change point is extracted to obtain the feature key point;

[0029] The random sampling consensus algorithm is used to match the reflected wave amplitude distribution of key feature points with the internal abnormal signal region to generate a spatial transformation matrix. The spatial transformation matrix is ​​used to spatially match the point cloud coordinates of the surface deformation candidate region with the acoustic wave propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established by iterative nearest point matching.

[0030] In one feasible implementation, the point cloud coordinates of the surface deformation candidate region are spatially matched with the acoustic propagation time data of the internal anomalous signal region. An iterative nearest-point matching is used to establish a correlation mapping between surface morphology features and internal defect features. The normal vector deflection angle and the reflected wave amplitude form a feature pair, including:

[0031] The normal vector deflection angle value of each point in the candidate region of surface deformation is weighted and summed with the reflected wave amplitude value at the corresponding spatial location in the internal anomalous signal region to generate the feature correlation value of each point;

[0032] Based on the spatial transformation matrix, the point cloud coordinates of the candidate surface deformation region are transformed to the coordinate system of the ultrasonic structural image data to generate transformed point cloud coordinates;

[0033] Based on the time data of the spatial location corresponding to the transformed point cloud coordinates and the sound wave propagation time data, the spatial distance deviation between the transformed point cloud coordinates and the time data is calculated. By adjusting the translation and rotation parameters of the spatial transformation matrix, the sum of the product of the feature correlation value and the spatial distance deviation is maximized.

[0034] Based on the adjusted spatial transformation matrix, a mapping relationship is established between the point cloud coordinates of each candidate surface deformation region and the acoustic wave propagation time data of the internal anomalous signal region, thus generating an associated mapping relationship.

[0035] In one feasible implementation, based on a multimodal defect distribution map, geometric shape parameters and reflection intensity parameters are extracted, and these parameters are matched with a preset defect feature template to generate detection results including defect type and defect location annotations, including:

[0036] Extract the set of boundary point coordinates for each defect region from the multimodal defect distribution map, and calculate the geometric shape parameters of the defect region based on the set of boundary point coordinates. The geometric shape parameters include the area of ​​the maximum bounding rectangle and the concavity and convexity of the boundary contour.

[0037] In the multimodal defect distribution map, the average reflection intensity value and the reflection intensity change gradient value are calculated based on the reflection intensity value of each point in the defect area to generate reflection intensity parameters;

[0038] The area of ​​the largest bounding rectangle and the concavity and convexity of the boundary contour in the geometric shape parameters, and the average reflection intensity value and the reflection intensity change gradient value in the reflection intensity parameters are compared with the corresponding parameter ranges in the preset defect feature template to generate the type matching degree of each defect region.

[0039] For defect areas where the type matching degree exceeds the preset matching threshold, the defect type label is determined, and the set of boundary point coordinates corresponding to the defect area is converted into three-dimensional spatial position coordinates to generate a detection result that includes defect type and three-dimensional position information.

[0040] Secondly, this application provides an image processing-based textile manufacturing defect detection system, applied to multilayer composite textiles, the system comprising:

[0041] The acquisition module is used to acquire surface point cloud data and ultrasonic structural image data of the target textile. The surface point cloud data is generated by acquiring the deformation stripe image of the target textile surface. The ultrasonic structural image data includes sound wave transmission time data and is generated by receiving the penetrating sound wave reflection signal emitted by the multi-frequency ultrasonic probe array.

[0042] The extraction module is used to filter noise points based on density distribution in surface point cloud data, extract surface curvature gradient and normal vector deflection angle to obtain surface deformation candidate regions, perform frequency band separation processing on ultrasonic structural image data, and extract reflected wave amplitude and signal delay values ​​to obtain internal abnormal signal regions.

[0043] A module is established to spatially match the point cloud coordinates of the candidate surface deformation region with the acoustic propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established through iterative nearest point matching. The normal vector deflection angle and the reflected wave amplitude form a feature pair.

[0044] The fusion module is used to perform anisotropic diffusion filtering and wavelet threshold denoising on surface morphology features and internal defect features respectively, and to allocate fusion weights according to the ratio of the surface curvature gradient change rate to the reflection wave amplitude attenuation rate. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map.

[0045] The generation module is used to extract geometric shape parameters and reflection intensity parameters based on the multimodal defect distribution map, and match the geometric shape parameters and reflection intensity parameters with the preset defect feature template to generate detection results including defect type and defect location annotations.

[0046] Thirdly, this application provides an electronic device, the device comprising: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the image processing-based textile production defect detection method as described in any embodiment of the first aspect.

[0047] Fourthly, this application provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the image processing-based textile production defect detection method as described in any embodiment of the first aspect.

[0048] This application discloses a method, system, device, and computer storage medium for detecting defects in textile production based on image processing. By fusing surface 3D point cloud data and ultrasonic structural image data, it simultaneously acquires the surface morphology features and internal structural information of textiles, overcoming the limitations of traditional visible light detection in perceiving internal defects. Based on density distribution noise filtering and frequency band separation processing, it optimizes the signal-to-noise ratio of surface and internal data respectively, improving the accuracy of defect candidate region extraction. Through spatial matching, it establishes a correlation mapping relationship between surface and internal features, and combines anisotropic diffusion filtering and wavelet threshold denoising to effectively suppress multi-source data coupling noise. Finally, it utilizes multimodal feature fusion and parameterized matching mechanisms to achieve collaborative identification of complex defects such as surface depressions, fiber breaks, and interlayer separation. This application can improve the detection accuracy and reliability of internal defects in thick textiles and solve the technical problem that existing technologies cannot effectively detect internal defects in multilayer composite textiles.

[0049] Furthermore, anisotropic diffusion filtering is used to optimize surface morphology features and suppress texture interference noise. Wavelet thresholding is employed to enhance the defect characterization capability of ultrasonic reflection signals. Based on a dynamic weight allocation mechanism of surface curvature gradient change rate and reflected wave amplitude attenuation rate, an exponential weight function is constructed by combining defect depth information. This causes the surface feature weight to decrease with increasing defect depth, while the internal feature weight increases with increasing defect depth, achieving adaptive feature fusion for defects of different depths. Finally, through accurate reconstruction of multimodal defect distribution maps, the distinction between deep fiber fractures and shallow delamination in thick textiles is significantly improved, solving the problem of low accuracy in detecting internal defects in multilayer composite materials in existing technologies. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1This is a schematic flowchart of a textile production defect detection method based on image processing provided in one embodiment of this application;

[0052] Figure 2 This is a flowchart illustrating a method for determining a multimodal defect distribution map according to an embodiment of this application;

[0053] Figure 3 This is a flowchart illustrating a method for generating textile production defect detection results according to an embodiment of this application;

[0054] Figure 4 This is a schematic diagram of the structure of a textile production defect detection system based on image processing provided in one embodiment of this application;

[0055] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0056] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0058] Existing methods for detecting manufacturing defects in thick textiles with multilayered composite structures suffer from limitations because visible light cannot effectively characterize the internal physical properties of the material. Consequently, they cannot accurately detect defects such as internal fiber breakage, interlayer separation, and foreign object embedding. Therefore, existing methods for detecting manufacturing defects in textiles with multilayered composite structures are unsuitable for this purpose, resulting in poor defect detection accuracy.

[0059] To address the problems of existing technologies, embodiments of this application provide a method, system, device, and computer storage medium for detecting textile production defects based on image processing. The method for detecting textile production defects based on image processing provided in this application embodiment will be described first below.

[0060] Figure 1 A schematic flowchart of an image processing-based textile production defect detection method according to an embodiment of this application is shown. Figure 1 As shown, steps S110 to S150 are included.

[0061] An image processing-based defect detection method for textile production is applied to multilayer composite textiles. The method includes:

[0062] S110: Acquire surface point cloud data and ultrasonic structural image data of the target textile. The surface point cloud data is generated by acquiring the deformation stripe image of the target textile surface. The ultrasonic structural image data includes sound wave transmission time data and is generated by receiving the penetrating sound wave reflection signal emitted by a multi-frequency ultrasonic probe array.

[0063] Surface point cloud data refers to the set of three-dimensional coordinates of a textile surface obtained through structured light 3D scanning technology. The data is generated by projecting coded grating fringes, acquiring deformed fringe images, and then reconstructing them using a binocular vision system, containing surface topography information with millimeter-level precision. Ultrasonic structural image data refers to the detection data generated by emitting penetrating sound waves through a multi-frequency ultrasonic probe array and receiving the reflected signals. It includes information on the time of sound wave propagation within the material and the distribution of reflected energy, used to characterize the internal structural features of textiles.

[0064] The target textile refers to a thick textile with a multi-layered composite structure, including a fiber reinforcement layer and an adhesive layer. Both surface deformation and internal defects need to be detected simultaneously. Deformation stripe images are image sequences containing surface deformation information captured by a binocular camera after a structured light projection system projects specifically coded grating stripes onto the textile surface. The stripes undergo geometric distortion due to surface unevenness. Acoustic wave propagation time data refers to the time difference between when ultrasonic waves encounter a defect interface and are reflected by the probe, used to calculate the defect depth and location. A multi-frequency ultrasonic probe array is a detection device composed of multiple ultrasonic probes with different operating frequencies. Low-frequency waves are used to detect deep structures, while high-frequency waves are used to capture shallow, minute defects. The array is arranged to cover the detection area.

[0065] First, a structured light projection system projects coded grating stripes onto the surface of the target textile. A binocular camera simultaneously captures images of the stripe distortion caused by surface deformation. High-precision surface point cloud data is reconstructed using a stereo vision algorithm, where each point contains three-dimensional coordinates and a normal vector. Simultaneously, a multi-frequency ultrasonic probe array moves along a preset path along the detection area of ​​the target textile, sequentially emitting penetrating sound waves of different frequencies. It receives reflection signals from the interfaces of different material layers within the textile, records the sound wave propagation time and reflected wave amplitude, and generates cross-sectional ultrasonic image data. The surface point cloud data and the ultrasonic structural image data are synchronized using a time synchronization device to ensure spatiotemporal consistency during acquisition.

[0066] For example, in a multilayer composite textile production line, after the target textile enters the inspection station, a structured light scanning module projects sinusoidal grating fringes at a fixed frequency. A binocular camera continuously captures images of deformed fringes with a microsecond time difference. A phase unwrapping algorithm generates surface point cloud data, with a point cloud density of several points per square millimeter. Simultaneously, an ultrasonic probe array mounted at the end of a robotic arm scans the textile surface in a serpentine path. Each probe emits sound waves of different frequencies at alternating frequencies. After reflection from the fiber layer and adhesive layer, the sound waves are recorded by a receiver with the reflection time and amplitude, generating ultrasonic image data containing multilayer reflection signals. The surface point cloud data and ultrasonic structural image data are stored synchronously with timestamps and transmitted to subsequent processing modules.

[0067] S120: Filter noise points based on density distribution on surface point cloud data, extract surface curvature gradient and normal vector deflection angle to obtain surface deformation candidate regions, perform frequency band separation processing on ultrasonic structural image data, and extract reflected wave amplitude and signal delay values ​​to obtain internal abnormal signal regions.

[0068] Surface deformation candidate regions refer to continuous point cloud regions that are preliminarily identified as potential areas with surface defects by screening for curvature gradients and normal vector deflection angles exceeding preset thresholds. Internal anomalous signal regions refer to areas in ultrasound images where the reflected wave amplitude exceeds the normal range and the time delay data deviates from the standard propagation time, representing potential defect locations where internal structural anomalies may exist.

[0069] Surface curvature gradient refers to the rate of change of three-dimensional coordinates between each point in a point cloud and its neighbors, reflecting the severity of surface unevenness. It is used to identify localized bulges or depressions on the surface of textiles caused by internal defects. Normal vector deflection angle represents the angle between the normal vector direction of a point in the point cloud and the average normal vector direction of the adjacent region. It is used to characterize abrupt changes in surface orientation, such as abnormal normal vector deflection caused by wrinkles or tears. Reflected wave amplitude refers to the signal energy intensity received after ultrasonic waves propagate within the textile and are reflected from a defect interface. The amplitude reflects the degree of acoustic impedance difference between the defect interface and the surrounding material. Signal delay value refers to the time difference between the ultrasonic wave's emission and the reception of the reflected signal, used to calculate the depth and location of the defect.

[0070] In the surface point cloud data processing stage, noise filtering is first performed based on density distribution statistics. Specifically, for each point in the point cloud, the number of points within its neighborhood radius is calculated as a density index. If this density value is lower than a preset threshold, the point is considered a noise point and is removed, retaining the region with higher density as the valid point cloud. Subsequently, in the valid point cloud, the local surface is fitted and the curvature gradient value is obtained by calculating the difference between the three-dimensional coordinates of each point and its neighboring points; simultaneously, the normal vector deflection angle is obtained by comparing the angle between the normal vector of the point and the average normal vector of the neighborhood. Continuous regions where both the curvature gradient and the deflection angle exceed the threshold are marked as candidate regions for surface deformation. These regions may correspond to deformation defects such as surface wrinkles, depressions, or protrusions.

[0071] In the ultrasonic structural image data processing stage, the original signal is first separated into frequency bands. The reflected wave signal is decomposed into multiple independent frequency bands using Fourier transform, each corresponding to a different depth detection sensitivity. For each frequency band, the maximum amplitude of the reflected wave is extracted, and the corresponding time delay data is recorded. Furthermore, regions with amplitudes exceeding the normal reflection intensity threshold and time delays significantly different from the standard propagation time are marked as internal anomalous signal regions. These regions may correspond to internal defects such as fiber breakage, interlaminar separation, or foreign object embedding.

[0072] S130: Spatial matching is performed between the point cloud coordinates of the candidate surface deformation region and the acoustic propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established through iterative nearest point matching. The normal vector deflection angle and the reflected wave amplitude form a feature pair.

[0073] Iterative nearest-point matching is a 3D point cloud registration algorithm that iteratively optimizes the spatial transformation parameters between two point clouds, including translation and rotation, to minimize the distance between corresponding points in the source and target point clouds, thereby achieving precise alignment of the spatial coordinate systems. The association mapping relationship refers to the geometric correspondence between the point cloud coordinates of candidate surface deformation regions and the acoustic wave propagation time data of internal anomalous signal regions established through spatial matching.

[0074] In the spatial matching stage, the point cloud coordinates of the surface deformation candidate region are first aligned with the acoustic wave propagation time data of the ultrasonic structural image data. Since there are spatial pose differences between the acquisition devices of the surface point cloud data and the ultrasonic data, an iterative nearest-point matching algorithm is used to calculate the optimal spatial transformation matrix between them. Specifically, the point cloud of the surface deformation candidate region is used as the source point cloud, and the three-dimensional spatial position corresponding to the acoustic wave propagation time of the ultrasonic data is used as the target point cloud. The translation vector and rotation matrix are iteratively calculated to minimize the Euclidean distance between corresponding points in the source and target point clouds.

[0075] When establishing the correlation mapping relationship, for each point in the candidate surface deformation region, the corresponding sound wave propagation time data is searched in the ultrasonic data according to the transformed coordinates, and the normal vector deflection angle and reflected wave amplitude at that location are recorded to form a feature pair. This ensures the physical correlation between surface deformation features and internal defect features. For example, a large normal vector deflection angle in a surface depression region may correspond to an abnormal reflected wave amplitude caused by internal interlayer separation.

[0076] S140: Anisotropic diffusion filtering and wavelet threshold denoising are applied to the surface morphology features and internal defect features respectively. The fusion weights are assigned according to the ratio of the surface curvature gradient change rate to the reflection wave amplitude attenuation rate. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map.

[0077] Anisotropic diffusion filtering is an image smoothing algorithm that preserves edge features. It suppresses noise while maintaining the geometric integrity of surface morphology by controlling the diffusion direction. Wavelet thresholding denoising removes random interference from ultrasonic reflection signals by decomposing the signal into different frequency components using wavelet transform and then thresholding the high-frequency noise coefficients. Fusion weights are feature fusion coefficients assigned to surface morphology and internal defect features based on the dynamic proportional relationship between the rate of change of surface curvature gradient and the attenuation rate of reflected wave amplitude. Multimodal defect distribution maps are two-dimensional or three-dimensional defect spatial distribution models generated by fusing surface morphology and internal structural features, containing multi-dimensional information such as the spatial location, geometric shape, and reflection energy intensity of defects.

[0078] In the feature processing stage, anisotropic diffusion filtering is first applied to surface morphology features, such as normal vector deflection angle and curvature gradient. This algorithm, based on partial differential equations, adjusts the diffusion intensity along the surface curvature direction, smoothing isolated noise points while preserving the edge sharpness of deformation features such as wrinkles and depressions. For internal defect features, such as reflected wave amplitude and time delay data, wavelet thresholding is used for noise reduction. Discrete wavelet transform is used to decompose the ultrasonic signal into approximation coefficients and detail coefficients. Soft thresholding is applied to the high-frequency detail coefficients to suppress noise, and the reconstructed signal retains the main energy components of the defect reflected wave.

[0079] In the fusion weight allocation phase, the ratio of the surface curvature gradient change rate (the magnitude of curvature change per unit distance) to the reflected wave amplitude attenuation rate (the proportion of energy loss per unit depth) is calculated. This ratio reflects the energy correlation between surface deformation and internal defects; for example, when curvature changes drastically but reflected wave attenuation is slow, surface features have a higher weight. Initial fusion weights are dynamically allocated based on this ratio to ensure that the contributions of surface and internal features in the fusion process match their physical correlation.

[0080] Finally, based on the spatial alignment of the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are superimposed according to weights. For example, the surface curvature gradient feature value and the reflected wave amplitude value at the corresponding position are weighted and summed to generate a fused feature value, which is finally mapped to three-dimensional space to form a multimodal defect distribution map.

[0081] S150: Based on the multimodal defect distribution map, extract geometric shape parameters and reflection intensity parameters, and match the geometric shape parameters and reflection intensity parameters with the preset defect feature template to generate detection results including defect type and defect location annotations.

[0082] Geometric parameters refer to the morphological features extracted from the defect region boundary, including the area of ​​the maximum bounding rectangle and the concavity / convexity of the boundary contour, used to characterize the size and shape complexity of the defect. Reflectance intensity parameters include the average reflection intensity value and the reflection intensity gradient value, i.e., the average reflected wave amplitude value and the rate of change of reflected energy in space across all points within the defect region. The defect feature template is a pre-defined database containing geometric and reflection parameter ranges for typical defect types. Each defect type corresponds to specific shape features and reflection intensity distribution patterns; typical defect types include interlaminar separation, fiber breakage, and foreign object embedding. Defect type labeling is the classification of defect physical attributes determined based on parameter matching results. For example, interlaminar separation manifests as a large, flat area accompanied by a low reflection intensity gradient, while fiber breakage presents as a narrow, elongated shape with abrupt changes in high reflection intensity. Defect location labeling describes the specific orientation of the defect within the textile using three-dimensional spatial coordinates, including planar position and depth information.

[0083] In the defect parameter extraction stage, the boundary points of the defect region are first identified from the multimodal defect distribution map. A region growing algorithm is used to traverse continuous regions where the fused feature values ​​exceed a threshold, extracting the set of boundary point coordinates. Based on the boundary point coordinates, the area of ​​the maximum circumscribed rectangle is calculated, and the smallest rectangle area that can completely enclose the defect region is found using a rotating caliper algorithm. The concavity / convexity of the boundary contour is obtained by calculating the cumulative angle between the normal vectors of adjacent points on the boundary curve; a larger value indicates a more irregular contour. The reflection intensity parameter is calculated by statistically analyzing the reflected wave amplitude values ​​of each point within the defect region. The average reflection intensity value is the arithmetic mean of the amplitude values ​​of all points within the region, reflecting the overall energy reflection level of the defect. The gradient value of the reflection intensity change is obtained by calculating the average absolute value of the difference between the amplitude values ​​of adjacent points, characterizing the uniformity of the internal structure of the defect. In the template matching stage, the extracted geometric shape parameters and reflection intensity parameters are compared with the parameter ranges in the preset defect feature template for similarity calculation. A weighted Euclidean distance algorithm is used to assign weights to different parameters; for example, concavity / convexity has a higher discriminative effect on interlayer separation. The similarity score between the current defect parameters and the template parameters is calculated. When the similarity exceeds a preset threshold, the defect type label corresponding to the template is assigned to the region. Finally, based on the three-dimensional coordinates of the boundary points of the defect region, they can be converted into positional information in the textile production coordinate system, generating an inspection report containing the defect type label and three-dimensional coordinates.

[0084] This embodiment integrates surface 3D point cloud data and ultrasonic structural image data to simultaneously acquire surface morphology features and internal structural information of textiles, overcoming the limitations of traditional visible light detection in perceiving internal defects. Based on density distribution noise filtering and frequency band separation processing, the signal-to-noise ratio of surface and internal data is optimized respectively, improving the accuracy of defect candidate region extraction. A correlation mapping relationship between surface and internal features is established through spatial matching, and combined with anisotropic diffusion filtering and wavelet threshold denoising, multi-source data coupling noise is effectively suppressed. Finally, multimodal feature fusion and parameterized matching mechanisms are used to achieve collaborative identification of complex defects such as surface depressions, fiber breaks, and interlayer separation. This application can improve the detection accuracy and reliability of internal defects in thick textiles, solving the technical problem that existing technologies cannot effectively detect internal defects in multilayer composite textiles.

[0085] Figure 2 A flowchart illustrating a method for determining a multimodal defect distribution map according to an embodiment of this application is shown. Figure 1 As shown, steps S210 to S240 are included.

[0086] In one feasible implementation, step S140 involves performing anisotropic diffusion filtering and wavelet thresholding denoising on the surface morphology features and internal defect features respectively, assigning fusion weights based on the ratio of the surface curvature gradient change rate to the attenuation rate of the reflected wave amplitude, and fusing the filtered surface morphology features and the denoised internal defect features into a multimodal defect distribution map based on the correlation mapping relationship, including:

[0087] S210: Anisotropic diffusion filtering and wavelet thresholding are applied to the surface morphology features and internal defect features respectively; and initial fusion weights are assigned based on the ratio of the surface curvature gradient change rate to the attenuation rate of the reflected wave amplitude.

[0088] First, anisotropic diffusion filtering is applied to surface topography features, including curvature gradients and normal vector deflection angles. This is based on the Perona-Malik model, adjusting the diffusion coefficient according to the local curvature direction to smooth isolated noise while preserving the edge sharpness of deformation features such as wrinkles and depressions. For internal defect features, including reflected wave amplitude and time delay data, wavelet thresholding denoising is performed. This involves decomposing the signal into multi-scale components using the Daubechies wavelet basis, applying soft thresholding to high-frequency noise components, and retaining the main energy components of the defect-reflected waves.

[0089] Next, the ratio of the surface curvature gradient change rate to the reflected wave amplitude attenuation rate is calculated, that is, the ratio of the maximum change in curvature value per unit distance to the energy attenuation ratio of the amplitude per unit depth. This ratio is converted into initial weights through normalization. For example, when the curvature change rate is significantly higher than the amplitude attenuation rate, the initial weights of the surface morphology features are set to a higher value.

[0090] S220: Determine the defect depth information based on the acoustic wave transmission time data corresponding to the internal defect features, and construct a weight function with the defect depth information as the exponential factor. The fusion weight of the surface morphology features is negatively correlated with the defect depth information, and the fusion weight of the internal defect features is positively correlated with the defect depth information.

[0091] Defect depth information refers to the vertical distance between the defect location and the textile surface, calculated using sound wave propagation time data. It is determined by the product of the sound wave propagation speed and the time difference, reflecting the depth of the defect buried inside the material.

[0092] S230: Determine the target fusion weight based on the weight function and the initial fusion weight. The target fusion weight includes the curvature gradient weight coefficient and the reflection energy weight coefficient.

[0093] Based on the ultrasonic transmission time data, the defect depth information is calculated, and an exponential weighting function is constructed, as shown in formula (1).

[0094] (1)

[0095] in, The depth of the defect is indicated by the ultrasonic transmission time; Indicates the attenuation coefficient; The initial weights represent the surface morphology features; The target fusion weights represent the surface topography features, i.e., the curvature gradient weight coefficients. The target fusion weight represents the internal defect characteristics, i.e., the reflection energy weight coefficient.

[0096] S240: Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused with target fusion weights to obtain a multimodal defect distribution map.

[0097] Based on spatial alignment of correlation mapping, the filtered surface curvature gradient feature value and the denoised reflected wave amplitude value are superimposed according to their respective target fusion weights to obtain the fused feature value. The fused feature value is then mapped to three-dimensional space to generate a multimodal defect distribution map containing location and energy intensity.

[0098] For example, in the inspection of thick multilayer composite textiles, surface morphology features, after anisotropic diffusion filtering, retain the curvature abrupt edges of the wrinkled areas while smoothing isolated noise points; internal defect features, after wavelet denoising, suppress fiber layer scattering noise and highlight the main peak of the reflected wave from interlayer separation defects. For interlayer separation defects with a depth of 5 mm, the initial fusion weight is calculated as the ratio of curvature gradient change rate to amplitude attenuation rate of 0.8. Combined with the exponential weight function and attenuation coefficient k=0.1, the surface weight coefficient is 0.48, and the internal weight coefficient is 1.31. In the spatially aligned region, the surface concavity feature value of 0.75 and the internal reflected wave amplitude of 1.2 are fused according to the weights to obtain 0.48×0.75 + 1.31×1.2 = 1.92, which exceeds the defect judgment threshold of 1.5, and is marked as an interlayer separation defect region. The fused multimodal defect distribution map simultaneously includes surface deformation and internal energy reflection features, providing a data basis for subsequent defect classification.

[0099] In one feasible implementation, step S240: Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused using target fusion weights to obtain a multimodal defect distribution map, including:

[0100] The fused feature value is generated by weighting and superimposing the feature values ​​of each point in the filtered surface morphology features with the feature values ​​of the same spatial location in the internal defect features corresponding to the association mapping relationship. The weight of the surface morphology features is the curvature gradient weight coefficient, and the weight of the internal defect features is the reflection energy weight coefficient.

[0101] Based on the spatial location correspondence table of the association mapping relationship, each point in the candidate surface deformation region is traversed. The curvature gradient value of the current point is extracted from the filtered surface morphology features. The reflected wave amplitude value at the corresponding location in the ultrasonic structural image data is found according to the association mapping relationship. The target fusion weight, i.e., the curvature gradient weight coefficient, is then used. With reflection energy weighting coefficient Perform weighted calculation: Fusion feature value = Curvature gradient value × + Reflected wave amplitude × This process ensures the physical correlation between surface and internal features; for example, when a high curvature gradient is accompanied by a high reflection wave amplitude, the fused feature value is significantly improved.

[0102] Based on the spatial location correspondence in the association mapping relationship, the fused feature values ​​are mapped to the three-dimensional spatial location corresponding to the point cloud coordinates of the surface deformation candidate region.

[0103] Three-dimensional spatial location mapping refers to assigning three-dimensional coordinates to the point cloud of surface deformation candidate regions by applying coordinate transformation rules in the correlation mapping relationship to the fused feature values, thus forming defect feature distribution data with spatial location information. First, according to the spatial transformation matrix of the correlation mapping relationship, the point cloud coordinates of the surface deformation candidate regions are transformed to the coordinate system of the ultrasonic structural image data. Then, the fused feature values ​​are matched with the ultrasonic data timestamps according to the point cloud index, ensuring that each fused feature value corresponds to a unique three-dimensional coordinate. Finally, an interpolation algorithm is used to fill in any missing defect locations that may arise from the coordinate system transformation, generating a complete three-dimensional spatial feature distribution.

[0104] In three-dimensional space, continuous regions with fused feature values ​​exceeding a preset defect determination threshold are marked as defect regions, generating a multimodal defect distribution map that includes defect region location information.

[0105] The defect determination threshold is a critical value set based on the statistical distribution of fused feature values ​​from historical defect samples, used to distinguish between normal and defective regions. First, all fused feature values ​​in 3D space are traversed, and points exceeding the preset threshold are filtered out. Then, a region growing algorithm is used to merge spatially adjacent points exceeding the threshold, forming continuous defective regions. Finally, based on the geometric center coordinates and boundary point set of the defective regions, a 2D projection or 3D point cloud model containing location information is generated.

[0106] For example, firstly, the curvature gradient feature value of the candidate surface deformation region after anisotropic diffusion filtering (e.g., a gradient value of 0.85 at a certain point) is weighted and superimposed with the denoised reflected wave amplitude value at the corresponding spatial location (e.g., an amplitude of 1.2), according to a curvature gradient weighting coefficient of 0.48 and a reflection energy weighting coefficient of 1.31, resulting in a fused feature value of 1.92. The fused feature value is mapped to the corresponding point cloud position in the three-dimensional coordinate system through the association mapping relationship established by the iterative nearest-point algorithm. When the fused feature value of a continuous region exceeds a preset threshold of 1.5 (e.g., an average value of 1.75 within a 3×3 neighborhood), the region is marked as a defect region. The finally generated multimodal defect distribution map is converted from a discrete point cloud into a continuous three-dimensional model using a spatial interpolation algorithm.

[0107] In one feasible implementation, step S120 involves: filtering noise points from the surface point cloud data based on density distribution, extracting the surface curvature gradient and normal vector deflection angle to obtain candidate regions for surface deformation, performing frequency band separation processing on the ultrasonic structural image data, and extracting the reflected wave amplitude and signal delay values ​​to obtain internal anomalous signal regions, including:

[0108] The density distribution statistics of each point are calculated based on the number of points in the neighborhood of each point in the surface point cloud data. Points with density distribution statistics below a preset density threshold are removed from the surface point cloud data to generate the target point cloud data.

[0109] A neighborhood search algorithm, such as the K-nearest neighbor algorithm, is used to traverse each point in the surface point cloud data, calculating the number of points within its neighborhood radius as a density distribution statistic. If the density statistic of a point is lower than a preset density threshold, it is identified as a noise point and removed. The preset density threshold is set based on the average density of point clouds on normal textile surfaces in historical data, for example, by subtracting the standard deviation from the mean density distribution of multiple defect-free samples. The final generated target point cloud data retains only continuous regions with high density, eliminating the interference of isolated noise points on subsequent feature extraction. For example, if the average point cloud spacing is 0.5 mm, the neighborhood radius is set to three times the average spacing of 1.5 mm, and the threshold is set to require that the neighborhood contain at least a certain number of points. The final generated target point cloud retains the geometric features of continuous surface regions.

[0110] In the target point cloud data, the surface curvature gradient value is obtained by calculating the three-dimensional coordinate difference between each point and its neighboring points. At the same time, the normal vector deflection angle value is calculated based on the angle between the normal vector direction of each point and the average normal vector direction of the adjacent area.

[0111] For each point in the target point cloud data, several neighboring points within its neighborhood are selected. A local surface is fitted using the least squares method, and the principal curvature value at that point is calculated. The curvature gradient value is defined as the maximum rate of change of principal curvature between neighboring points. The calculation process for the normal vector deflection angle is as follows: first, the average normal vector of all points in the neighborhood is calculated; then, the angle between the current point's normal vector and the average normal vector is calculated using the vector dot product formula. The curvature gradient value and the normal vector deflection angle value together characterize the geometric features of surface deformation.

[0112] In the target point cloud data, continuous point cloud regions whose surface curvature gradient values ​​are greater than a preset curvature gradient threshold and whose normal vector deflection angle values ​​are greater than a preset deflection angle threshold are marked as surface deformation candidate regions.

[0113] For each point in the target point cloud data, if its curvature gradient value exceeds a preset curvature gradient threshold and its normal vector deflection angle value exceeds a preset deflection angle threshold, it is marked as a candidate point. A region growing algorithm is used to merge spatially adjacent candidate points to form a continuous region. The preset curvature gradient threshold is determined by adding a certain margin to the maximum value of the normal curvature distribution of a normal surface, while the preset deflection angle threshold is set based on the magnitude of the normal vector mutation of a typical defect. The finally generated surface deformation candidate regions provide high-confidence surface defect locations for subsequent correlation analysis.

[0114] The ultrasonic structural image data is decomposed into reflected wave signals in multiple independent frequency bands. The maximum amplitude of the reflected wave signal in each frequency band is calculated, and the sound wave propagation time data corresponding to the maximum amplitude is recorded.

[0115] Fast Fourier Transform (FFT) is performed on ultrasonic structural image data to decompose it into multiple independent frequency bands, such as low-frequency, mid-frequency, and high-frequency bands. For the reflected wave signal of each frequency band, the maximum value of its amplitude versus time curve is extracted as the maximum amplitude of that frequency band, and the corresponding sound wave propagation time is recorded. The low-frequency band is used to detect deep structures, while the high-frequency band captures shallow details. Multi-band separation enhances the detection capability of defects at different depths.

[0116] In the ultrasound structural image data, the region where the maximum amplitude exceeds the preset reflected wave threshold and the difference between the corresponding sound wave propagation time data and the preset standard propagation time data exceeds the preset difference threshold is marked as an internal abnormal signal region.

[0117] The process iterates through each detection location in the ultrasonic structural image data. If the maximum amplitude at a location exceeds a preset reflection threshold, and the difference between its sound wave propagation time and the standard propagation time exceeds a preset difference threshold, it is marked as an anomaly. The preset reflection threshold is determined by statistically analyzing the reflection amplitude distribution of defect-free samples, while the preset difference threshold is set based on the additional propagation delay caused by the sound wave at the defect interface. Finally, adjacent anomalies are merged to form an internal anomalous signal region, characterizing internal defects such as fiber breakage or interlaminar separation.

[0118] For example, in the detection scenario of multilayer composite textiles, surface point cloud data is first acquired through structured light scanning. Isolated noise points are filtered out using neighborhood density statistics, retaining only valid point clouds. Then, the curvature gradient and normal vector deflection angle of each point are calculated, and continuous regions with abrupt curvature changes and abnormal normal vector directions are selected as candidate regions for surface deformation. Simultaneously, the ultrasonic structural image data is frequency-band separated, and the maximum reflection amplitude and corresponding time data for each frequency band are extracted. Internal anomalous signal regions are then marked using amplitude and time difference thresholds. For instance, if a region is detected with a surface curvature gradient value of 0.12 (exceeding the threshold of 0.1), a normal vector deflection angle of 15 degrees (exceeding the threshold of 10 degrees), and a corresponding low-frequency amplitude of 8 dB (exceeding the threshold of 6 dB) and a time difference of 2 μs (exceeding the threshold of 1.5 μs), it is marked as a candidate region for surface deformation and an internal anomalous signal region, respectively.

[0119] In one feasible implementation, before establishing the association mapping between surface morphology features and internal defect features through iterative nearest-point matching, the method includes:

[0120] Based on the surface curvature gradient of the candidate region of surface deformation, the curvature abrupt change point is extracted to obtain the feature key point.

[0121] Feature keypoints refer to the point cloud locations in the candidate region of surface deformation where the curvature gradient value changes significantly, representing the local geometric features of the surface deformation, such as wrinkle edges and depression boundaries. In the point cloud data of the candidate region of surface deformation, the curvature gradient value of each point is traversed, and the extreme value of the curvature gradient in the local neighborhood is calculated using a sliding window. Specifically, for each point, the rate of change of the curvature gradient within its neighborhood, such as the radius neighborhood or K-nearest neighbors, is calculated. If the curvature gradient value of a point is the maximum value in the neighborhood and exceeds a preset abrupt change threshold, it is marked as a curvature abrupt change point. The preset abrupt change threshold is set based on the curvature gradient distribution of typical defect areas in historical data. Finally, a region growing algorithm is used to merge spatially adjacent curvature abrupt change points to form a continuous set of feature keypoints.

[0122] The random sampling consensus algorithm is used to match the reflected wave amplitude distribution of key feature points with the internal abnormal signal region to generate a spatial transformation matrix. The spatial transformation matrix is ​​used to spatially match the point cloud coordinates of the surface deformation candidate region with the acoustic wave propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established by iterative nearest point matching.

[0123] Random Sampling Consensus (RANSAC) is a robust parameter estimation method that estimates mathematical model parameters from noisy data through random sampling and iterative optimization, making it suitable for matching scenarios with outliers. It uses the 3D coordinates of key feature points as the source point set and the spatial locations corresponding to the reflected wave amplitude distribution within the internal anomalous signal region as the target point set. Several pairs of matching points are randomly selected, such as three pairs of non-coplanar points, and an initial spatial transformation matrix including rotation and translation vectors is calculated. The RANSAC algorithm is iteratively optimized to calculate the distance error between the transformed source and target points. The number of matching points (inliers) that meet the error threshold is counted, and the transformation matrix with the most inliers is retained as the optimal solution. The parameters of the spatial transformation matrix are optimized using the least squares method. The final transformation matrix is ​​used to transform the surface point cloud coordinates to the coordinate system of the ultrasonic data, providing initial alignment parameters for iterative nearest-point matching.

[0124] For example, in the inspection of multilayer composite textiles, curvature abrupt change points are first extracted from candidate surface deformation regions as key feature points. For instance, if the curvature gradient value of a certain folded region exhibits a local peak in its neighborhood, point clouds with curvature gradients exceeding a threshold and continuously distributed are selected to form a set of key feature points. Subsequently, the key feature points are matched with the reflected wave amplitude distribution of internal abnormal signal regions: assuming that the reflected wave amplitude corresponding to a certain interlayer separation defect in the ultrasonic data exhibits a high amplitude distribution at a specific spatial location, three sets of key feature points are randomly selected using the RANSAC algorithm to perform initial matching with the high amplitude points. After calculating the transformation matrix, the proportion of internal points is verified. If the proportion of internal points exceeds a preset threshold, for example, if the proportion of points with matching errors within a certain range exceeds a certain percentage, the transformation matrix is ​​accepted as the initial alignment parameter. The final generated transformation matrix is ​​used for spatial alignment of the surface point cloud and the ultrasonic data, providing a high-precision initial value for subsequent iterative nearest-point matching, ensuring the accuracy of the correlation mapping relationship between surface morphology features and internal defect features.

[0125] In one feasible implementation, step S130: spatially matching the point cloud coordinates of the candidate surface deformation region with the acoustic wave propagation time data of the internal abnormal signal region, establishing the correlation mapping relationship between surface morphology features and internal defect features through iterative nearest-point matching, and forming a feature pair between the normal vector deflection angle and the reflected wave amplitude, including:

[0126] The normal vector deflection angle value of each point in the candidate region of surface deformation is weighted and summed with the reflected wave amplitude value of the corresponding spatial location in the internal abnormal signal region to generate the feature correlation value of each point.

[0127] For each point in the candidate surface deformation region, the reflected wave amplitude at the corresponding spatial location in the ultrasonic data is found based on the correlation mapping relationship. For each point, the normal vector deflection angle and the reflected wave amplitude are weighted and summed according to preset weight coefficients. The feature correlation value = normal vector deflection angle × weight coefficient 1 + reflected wave amplitude × weight coefficient 2. The weight coefficients are allocated based on the contribution of surface and internal features in historical data, for example, by statistically determining the parameter correlation of typical defect regions, such as a normal vector weight of 0.6 and an amplitude weight of 0.4. The final generated feature correlation value is mapped to three-dimensional space to form a preliminary defect correlation distribution.

[0128] Based on the spatial transformation matrix, the point cloud coordinates of the candidate surface deformation region are transformed to the coordinate system of the ultrasonic structural image data to generate transformed point cloud coordinates.

[0129] Based on the initial spatial transformation matrix (containing the rotation matrix R and translation vector T) generated by the RANSAC algorithm, a geometric transformation is performed on the point cloud coordinates of the surface deformation candidate regions. Specifically, for each point cloud coordinate P, the transformed coordinate P' is calculated, P' = R × P + T. The coordinate system transformation is completed by traversing all point cloud coordinates, generating transformed point cloud coordinates aligned with the ultrasonic data space. This transformation ensures a one-to-one correspondence between the physical locations of the surface deformation candidate regions and the internal anomalous signal regions.

[0130] Based on the time data of the spatial location corresponding to the transformed point cloud coordinates and the sound wave propagation time data, the spatial distance deviation between the transformed point cloud coordinates and the time data is calculated. By adjusting the translation and rotation parameters of the spatial transformation matrix, the sum of the product of the feature correlation value and the spatial distance deviation is maximized.

[0131] The Iterative Nearest Point (ICP) algorithm is used to optimize the parameters of the spatial transformation matrix. First, the distance deviation is calculated by finding the spatial location of the nearest neighbor in the ultrasound data for each transformed point's cloud coordinates and calculating the Euclidean distance as the deviation value. Next, an objective function is constructed, using the sum of the products of feature association values ​​and distance deviations as the optimization objective, maximizing this value to simultaneously satisfy feature association strength and spatial alignment accuracy. Finally, parameter optimization is performed by adjusting the rotation matrix R and translation vector T using gradient descent or singular value decomposition (SVD), iteratively updating the transformation matrix until the objective function converges.

[0132] Based on the adjusted spatial transformation matrix, a mapping relationship is established between the point cloud coordinates of each candidate surface deformation region and the acoustic wave propagation time data of the internal anomalous signal region, thus generating an associated mapping relationship.

[0133] Under the optimized spatial transformation matrix, the point cloud coordinates of each candidate region of surface deformation are precisely mapped to the sound wave propagation time data of the corresponding spatial location in the ultrasonic data. The transformed position of each point cloud coordinate and its associated sound wave propagation time and reflected wave amplitude are recorded to generate a mapping table containing spatial location, surface morphology features (normal vector deflection angle), and internal defect features (reflected wave amplitude).

[0134] For example, when detecting defects in multilayer composite textiles, the normal vector deflection angle (e.g., 15 degrees) of the surface wrinkle region is first weighted and fused with the high-reflection wave amplitude (e.g., 8 dB) in the corresponding ultrasonic data through feature correlation value calculation to generate a high correlation value region. For example, the normal vector deflection angle can be 15 degrees, and the high-reflection wave amplitude can be 8 dB. Subsequently, the surface point cloud coordinates are transformed to the ultrasonic coordinate system based on the initial spatial transformation matrix, and the deviation between the transformed coordinates and the ultrasonic data position is calculated. The transformation matrix is ​​iteratively adjusted through the ICP algorithm to maximize the sum of the products of the feature correlation value and the spatial deviation. For example, the sum of the products is increased by a certain proportion after optimization, and finally, an accurate mapping between the surface wrinkle points and the internal high-reflection regions is established. The adjusted transformation matrix ensures a one-to-one correspondence between the surface point cloud coordinates and the sound wave propagation time data. For example, after transformation, the coordinates of a certain point cloud coincide with the position in the ultrasonic data with a time difference of 2 μs, forming an correlation mapping relationship.

[0135] Figure 3 A flowchart illustrating a method for generating textile manufacturing defect detection results according to an embodiment of this application is shown. Figure 1 As shown, steps S310 to S350 are included.

[0136] In one feasible implementation, step S150: Based on the multimodal defect distribution map, extract geometric shape parameters and reflection intensity parameters, and match the geometric shape parameters and reflection intensity parameters with a preset defect feature template to generate a detection result including defect type and defect location annotation, including:

[0137] S310: Extract the set of boundary point coordinates for each defect region from the multimodal defect distribution map, and calculate the geometric shape parameters of the defect region based on the set of boundary point coordinates. The geometric shape parameters include the area of ​​the maximum circumscribed rectangle and the concavity and convexity of the boundary contour.

[0138] A region growing algorithm is used to traverse continuous regions in the multimodal defect distribution map where the fused feature value exceeds a preset threshold, extracting the boundary point coordinates of the defect regions. Based on the boundary point coordinates, a rotating caliper algorithm is used to find the minimum bounding rectangle, and its area is calculated as the area of ​​the maximum bounding rectangle. The calculation process for the concavity and convexity of the boundary contour is as follows: traverse the contour point sequence, calculate the absolute value of the angle between the normal vectors of adjacent points and accumulate them; the ratio of the accumulated value to the contour length is the concavity and convexity.

[0139] S320: In the multimodal defect distribution map, calculate the average reflection intensity value and the reflection intensity change gradient value based on the reflection intensity value of each point in the defect area, and generate the reflection intensity parameter.

[0140] For each point within the defect region, the reflected wave amplitude is counted and the average value is calculated as the average reflection intensity value. At the same time, for each point, the maximum difference in reflected wave amplitude within its neighborhood, such as the eight-neighborhood, is calculated, and the average of the maximum differences among all points is taken as the reflection intensity change gradient value.

[0141] S330: The area of ​​the largest bounding rectangle and the concavity and convexity of the boundary contour in the geometric shape parameters, and the average reflection intensity value and the reflection intensity change gradient value in the reflection intensity parameters are compared with the corresponding parameter ranges in the preset defect feature template to generate the type matching degree of each defect region.

[0142] For each defect region, its geometric shape parameters and reflection intensity parameters are calculated, namely, the area of ​​the maximum bounding rectangle, concavity / convexity, average reflection intensity, and gradient variation, and the weighted Euclidean distance between these parameters and the corresponding parameters for each defect type in the template. Weights are assigned based on the parameter discrimination; for example, concavity / convexity has a higher weight for interlayer separation. Finally, the distances are converted into matching scores through normalization.

[0143] S340: Based on the defect areas where the type matching degree exceeds the preset matching threshold, determine the defect type label, convert the set of boundary point coordinates corresponding to the defect area into three-dimensional spatial position coordinates, and generate a detection result that includes defect type and three-dimensional position information.

[0144] For each defect area, if its matching degree exceeds a preset threshold, the defect type with the highest matching degree is selected as the label. The defect type can be interlayer separation, fiber breakage, foreign object embedding, etc. Based on the spatial transformation relationship between the multimodal defect distribution map and the textile production coordinate system, for example, by calibrating parameters or transformation matrices, the three-dimensional coordinates of the defect boundary points are converted into the actual production location, generating an inspection report that includes type labels, planar coordinates, and depth.

[0145] For example, when detecting interlaminar separation defects in multilayer composite textiles, the boundary points of the defect region are first extracted from the multimodal defect distribution map. Assuming a region has a maximum circumscribed rectangle area of ​​several square millimeters, its boundary contour convexity exceeds a preset threshold, and its average reflection intensity is low with a small gradient of reflection intensity variation, the region's geometric and reflection parameters match the parameter range of the interlaminar separation template most closely through similarity calculation. For example, if the matching degree exceeds a certain proportion, it is marked as an interlaminar separation type. Subsequently, the coordinates of the boundary points of this region are converted into three-dimensional positions in the textile production coordinate system using a spatial transformation matrix, for example, planar coordinates of several millimeters and depth of several millimeters, generating an inspection report. For fiber breakage defects, if they exhibit a narrow and elongated shape (i.e., a small maximum circumscribed rectangle area, low boundary convexity, and a significant gradient of reflection intensity variation), the fiber breakage template is matched, and the corresponding position is marked.

[0146] Based on the same concept, this application provides an image processing-based textile production defect detection system, which is described below in conjunction with... Figure 4 The image processing-based textile production defect detection system provided in this application will be described in detail.

[0147] Figure 4 This is a structural block diagram of an image processing-based textile production defect detection system, as shown in an embodiment of this application.

[0148] like Figure 4 As shown, this image processing-based textile manufacturing defect detection system is applied to multi-layer composite textiles. The system includes:

[0149] The acquisition module 410 is used to acquire surface point cloud data and ultrasonic structural image data of the target textile. The surface point cloud data is generated by acquiring the deformation stripe image of the target textile surface. The ultrasonic structural image data includes sound wave transmission time data and is generated by receiving the penetrating sound wave reflection signal emitted by the multi-frequency ultrasonic probe array.

[0150] The extraction module 420 is used to filter noise points based on density distribution in surface point cloud data, extract surface curvature gradient and normal vector deflection angle to obtain surface deformation candidate regions, perform frequency band separation processing on ultrasonic structural image data, and extract reflected wave amplitude and signal delay values ​​to obtain internal abnormal signal regions.

[0151] Module 430 is established to spatially match the point cloud coordinates of the candidate surface deformation region with the acoustic wave propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established through iterative nearest point matching. The normal vector deflection angle and the reflected wave amplitude form a feature pair.

[0152] The fusion module 440 is used to perform anisotropic diffusion filtering and wavelet threshold denoising on surface morphology features and internal defect features respectively, and to allocate fusion weights according to the ratio of the surface curvature gradient change rate to the reflection wave amplitude attenuation rate. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map.

[0153] The generation module 450 is used to extract geometric shape parameters and reflection intensity parameters based on the multimodal defect distribution map, and match the geometric shape parameters and reflection intensity parameters with the preset defect feature template to generate detection results including defect type and defect location annotation.

[0154] In one embodiment, the fusion module 440 is specifically used to perform anisotropic diffusion filtering and wavelet thresholding denoising on surface morphology features and internal defect features, respectively; and to allocate initial fusion weights according to the ratio of the surface curvature gradient change rate to the reflection wave amplitude attenuation rate; to determine defect depth information based on the acoustic wave transmission time data corresponding to the internal defect features, and to construct a weight function with the defect depth information as an exponential factor, wherein the fusion weight of surface morphology features is negatively correlated with the defect depth information, and the fusion weight of internal defect features is positively correlated with the defect depth information; to determine the target fusion weight based on the weight function and the initial fusion weight, the target fusion weight including the curvature gradient weight coefficient and the reflection energy weight coefficient; and to fuse the filtered surface morphology features and the denoised internal defect features with the target fusion weight based on the correlation mapping relationship to obtain a multimodal defect distribution map.

[0155] In one embodiment, the fusion module 440 is specifically used to generate a fused feature value by weighting and superimposing the feature values ​​of each point in the filtered surface morphology features with the feature values ​​of the same spatial position in the internal defect features corresponding to the association mapping relationship. The weight of the surface morphology features is the curvature gradient weight coefficient, and the weight of the internal defect features is the reflection energy weight coefficient. According to the spatial position correspondence in the association mapping relationship, the fused feature value is mapped to the three-dimensional spatial position corresponding to the point cloud coordinates of the surface deformation candidate region. In the three-dimensional spatial position, the continuous region where the fused feature value exceeds the preset defect judgment threshold is marked as a defect region, and a multimodal defect distribution map including defect region location information is generated.

[0156] In one embodiment, the extraction module 420 is specifically used to calculate the density distribution statistics of each point based on the number of points in the neighborhood of each point in the surface point cloud data, remove points in the surface point cloud data whose density distribution statistics are lower than a preset density threshold, and generate target point cloud data; in the target point cloud data, the surface curvature gradient value is obtained by calculating the three-dimensional coordinate difference between each point and its neighboring points, and the normal vector deflection angle value is calculated based on the angle between the normal vector direction of each point and the average normal vector direction of the adjacent region; continuous point cloud regions in the target point cloud data whose surface curvature gradient value is greater than a preset curvature gradient threshold and whose normal vector deflection angle value is greater than a preset deflection angle threshold are marked as candidate regions for surface deformation; the ultrasonic structural image data is decomposed into reflected wave signals of multiple independent frequency bands, the maximum amplitude of the reflected wave signal in each frequency band is calculated, and the sound wave propagation time data corresponding to the maximum amplitude is recorded; regions in the ultrasonic structural image data whose maximum amplitude exceeds a preset reflected wave threshold and whose difference between the corresponding sound wave propagation time data and the preset standard propagation time data exceeds a preset difference threshold are marked as internal abnormal signal regions.

[0157] In one embodiment, the establishment module 430 is specifically used to extract curvature abrupt change points to obtain key feature points based on the surface curvature gradient of the surface deformation candidate region before establishing the association mapping relationship between surface morphology features and internal defect features through iterative nearest point matching; and to match the key feature points with the reflected wave amplitude distribution of the internal abnormal signal region through a random sampling consensus algorithm to generate a spatial transformation matrix. The spatial transformation matrix is ​​used to spatially match the point cloud coordinates of the surface deformation candidate region with the acoustic wave propagation time data of the internal abnormal signal region, and to establish the association mapping relationship between surface morphology features and internal defect features through iterative nearest point matching.

[0158] In one embodiment, the module 430 is specifically used to perform a weighted summation of the normal vector deflection angle value of each point in the surface deformation candidate region and the reflected wave amplitude value of the corresponding spatial position in the internal abnormal signal region to generate a feature association value for each point; based on the spatial transformation matrix, the point cloud coordinates of the surface deformation candidate region are transformed to the coordinate system of the ultrasonic structural image data to generate transformed point cloud coordinates; according to the time data of the spatial position corresponding to the transformed point cloud coordinates and the sound wave propagation time data, the spatial distance deviation between the transformed point cloud coordinates and the time data is calculated, and the sum of the product of the feature association value and the spatial distance deviation is maximized by adjusting the translation and rotation parameters of the spatial transformation matrix; based on the adjusted spatial transformation matrix, a mapping relationship is established between each point cloud coordinate of the surface deformation candidate region and the sound wave propagation time data of the internal abnormal signal region to generate an association mapping relationship.

[0159] In one embodiment, the generation module 450 is specifically used to extract the set of boundary point coordinates for each defect region from the multimodal defect distribution map, calculate the geometric shape parameters of the defect region based on the set of boundary point coordinates, the geometric shape parameters including the area of ​​the maximum bounding rectangle and the concavity / convexity of the boundary contour; in the multimodal defect distribution map, calculate the average reflection intensity value and the reflection intensity change gradient value based on the reflection intensity value of each point in the defect region to generate reflection intensity parameters; calculate the similarity between the area of ​​the maximum bounding rectangle and the concavity / convexity of the boundary contour in the geometric shape parameters, and the average reflection intensity value and the reflection intensity change gradient value in the reflection intensity parameters, respectively, with the corresponding parameter ranges in the preset defect feature template to generate the type matching degree for each defect region; determine the defect type label for defect regions whose type matching degree exceeds the preset matching threshold, and convert the set of boundary point coordinates corresponding to the defect region into three-dimensional spatial position coordinates to generate a detection result including defect type and three-dimensional position information.

[0160] Figure 4 Each module in the system shown has an implementation Figures 1 to 3 The functions of each step in the process and their corresponding technical effects are described in detail here for the sake of brevity.

[0161] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application is shown.

[0162] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0163] Specifically, the processor 510 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0164] Memory 520 may include mass storage for data or instructions. For example, and not limitingly, memory 520 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 520 may include removable or non-removable (or fixed) media. Where appropriate, memory 520 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 520 is non-volatile solid-state memory.

[0165] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this disclosure.

[0166] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any of the image processing-based textile production defect detection methods in the above embodiments.

[0167] In one example, the electronic device may also include a communication interface 530 and a bus 540. Wherein, such as Figure 5 As shown, the processor 510, memory 520, and communication interface 530 are connected through bus 540 and complete communication with each other.

[0168] The communication interface 530 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0169] Bus 540 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 540 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0170] This electronic device can execute the image processing-based textile production defect detection method described in the embodiments of this application, thereby achieving a combination of Figures 1 to 3 The image processing-based method for detecting defects in textile production is described.

[0171] Furthermore, in conjunction with the image processing-based textile production defect detection method in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the image processing-based textile production defect detection methods in the above embodiments.

[0172] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0173] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0174] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0175] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0176] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for detecting defects in textile production based on image processing, applied to multilayer composite textiles, characterized in that, include: The surface point cloud data and ultrasonic structural image data of the target textile are acquired. The surface point cloud data is generated by acquiring the deformation stripe image of the surface of the target textile. The ultrasonic structural image data includes sound wave propagation time data and is generated by receiving the penetrating sound wave reflection signal emitted by a multi-frequency ultrasonic probe array. The surface point cloud data is subjected to noise point filtering based on density distribution, and the surface curvature gradient and normal vector deflection angle are extracted to obtain the surface deformation candidate region. The ultrasonic structure image data is subjected to frequency band separation processing, and the reflected wave amplitude and signal delay value are extracted to obtain the internal abnormal signal region. Spatial matching is performed between the point cloud coordinates of the candidate surface deformation region and the acoustic wave propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established by iterative nearest point matching. The normal vector deflection angle and the reflected wave amplitude form a feature pair. Anisotropic diffusion filtering and wavelet threshold denoising are applied to the surface morphology features and the internal defect features respectively. Fusion weights are assigned according to the ratio of the surface curvature gradient change rate to the reflection wave amplitude attenuation rate. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map. Based on the multimodal defect distribution map, geometric shape parameters and reflection intensity parameters are extracted, and the geometric shape parameters and reflection intensity parameters are matched with a preset defect feature template to generate detection results including defect type and defect location annotations.

2. The method according to claim 1, characterized in that, The process involves performing anisotropic diffusion filtering and wavelet thresholding denoising on the surface morphology features and the internal defect features respectively, assigning fusion weights based on the ratio of the surface curvature gradient change rate to the reflected wave amplitude attenuation rate, and fusing the filtered surface morphology features and the denoised internal defect features into a multimodal defect distribution map based on the correlation mapping relationship, including: Anisotropic diffusion filtering and wavelet thresholding denoising are applied to the surface morphology features and the internal defect features, respectively. Initial fusion weights are assigned based on the ratio of the surface curvature gradient change rate to the attenuation rate of the reflected wave amplitude; Based on the acoustic wave propagation time data corresponding to the internal defect features, the defect depth information is determined, and a weight function with the defect depth information as the exponential factor is constructed. The fusion weight of the surface morphology features is negatively correlated with the defect depth information, and the fusion weight of the internal defect features is positively correlated with the defect depth information. The target fusion weight is determined based on the weighting function and the initial fusion weight, wherein the target fusion weight includes curvature gradient weight coefficient and reflection energy weight coefficient. Based on the aforementioned correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused using the target fusion weights to obtain the multimodal defect distribution map.

3. The method according to claim 2, characterized in that, Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused using the target fusion weights to obtain the multimodal defect distribution map, including: The feature value of each point in the filtered surface morphology feature is weighted and superimposed with the feature value of the same spatial position in the internal defect feature corresponding to the association mapping relationship to generate a fused feature value, wherein the weight of the surface morphology feature is the curvature gradient weight coefficient, and the weight of the internal defect feature is the reflection energy weight coefficient. Based on the spatial location correspondence in the association mapping relationship, the fused feature value is mapped to the three-dimensional spatial location corresponding to the point cloud coordinates of the surface deformation candidate region; In the three-dimensional spatial location, the continuous region where the fused feature value exceeds the preset defect judgment threshold is marked as a defect region, and the multimodal defect distribution map including defect region location information is generated.

4. The method according to claim 1, characterized in that, The process of filtering noise points based on density distribution from the surface point cloud data, extracting surface curvature gradients and normal vector deflection angles to obtain candidate regions for surface deformation, and performing frequency band separation processing on the ultrasonic structural image data to extract reflected wave amplitude and signal delay values ​​to obtain internal anomalous signal regions includes: Calculate the density distribution statistics of each point based on the number of points in the neighborhood of each point in the surface point cloud data, remove points in the surface point cloud data whose density distribution statistics are lower than a preset density threshold, and generate target point cloud data. In the target point cloud data, the surface curvature gradient value is obtained by calculating the three-dimensional coordinate difference between each point and its neighboring points. At the same time, the normal vector deflection angle value is calculated based on the angle between the normal vector direction of each point and the average normal vector direction of the adjacent area. In the target point cloud data, continuous point cloud regions where the surface curvature gradient value is greater than a preset curvature gradient threshold and the normal vector deflection angle value is greater than a preset deflection angle threshold are marked as surface deformation candidate regions. The ultrasonic structural image data is decomposed into reflected wave signals in multiple independent frequency bands. The maximum amplitude of the reflected wave signal in each frequency band is calculated, and the sound wave propagation time data corresponding to the maximum amplitude is recorded. In the ultrasonic structural image data, the region where the maximum amplitude exceeds a preset reflected wave threshold and the difference between the corresponding sound wave propagation time data and the preset standard propagation time data exceeds a preset difference threshold is marked as an internal abnormal signal region.

5. The method according to claim 1, characterized in that, Before establishing the association mapping relationship between surface morphology features and internal defect features through iterative nearest point matching, the method includes: Based on the surface curvature gradient of the candidate surface deformation region, curvature abrupt change points are extracted to obtain key feature points; The random sampling consensus algorithm is used to match the reflected wave amplitude distribution of the key feature points with the internal abnormal signal region to generate a spatial transformation matrix. The spatial transformation matrix is ​​used to spatially match the point cloud coordinates of the surface deformation candidate region with the sound wave propagation time data of the internal abnormal signal region. The correlation mapping relationship between surface morphology features and internal defect features is established by iterative nearest point matching.

6. The method according to claim 5, characterized in that, The step of spatially matching the point cloud coordinates of the candidate surface deformation region with the acoustic wave propagation time data of the internal abnormal signal region, and establishing the correlation mapping relationship between surface morphology features and internal defect features through iterative nearest-point matching, wherein the normal vector deflection angle and the reflected wave amplitude form a feature pair, includes: The normal vector deflection angle value of each point in the candidate surface deformation region is weighted and summed with the reflected wave amplitude value at the corresponding spatial location in the internal abnormal signal region to generate the feature correlation value of each point; Based on the spatial transformation matrix, the point cloud coordinates of the candidate surface deformation region are transformed to the coordinate system of the ultrasonic structural image data to generate transformed point cloud coordinates; Based on the time data of the spatial position corresponding to the transformed point cloud coordinates and the sound wave propagation time data, the spatial distance deviation between the transformed point cloud coordinates and the time data is calculated. By adjusting the translation and rotation parameters of the spatial transformation matrix, the sum of the product of the feature correlation value and the spatial distance deviation is maximized. Based on the adjusted spatial transformation matrix, a mapping relationship is established between the point cloud coordinates of each candidate surface deformation region and the acoustic wave propagation time data of the internal abnormal signal region, thereby generating the associated mapping relationship.

7. The method according to claim 1, characterized in that, Based on the multimodal defect distribution map, geometric shape parameters and reflection intensity parameters are extracted, and the geometric shape parameters and reflection intensity parameters are matched with a preset defect feature template to generate a detection result including defect type and defect location annotation, including: Extract the set of boundary point coordinates for each defect region from the multimodal defect distribution map, and calculate the geometric shape parameters of the defect region based on the set of boundary point coordinates. The geometric shape parameters include the area of ​​the maximum circumscribed rectangle and the concavity and convexity of the boundary contour. In the multimodal defect distribution map, the average reflection intensity value and the reflection intensity change gradient value are calculated based on the reflection intensity value of each point in the defect area to generate reflection intensity parameters; The maximum bounding rectangle area and boundary contour convexity in the geometric shape parameters, and the average reflection intensity value and reflection intensity change gradient value in the reflection intensity parameters are respectively compared with the corresponding parameter range in the preset defect feature template to generate the type matching degree of each defect region. Based on the defect regions where the type matching degree exceeds a preset matching threshold, a defect type label is determined, and the set of boundary point coordinates corresponding to the defect region is converted into three-dimensional spatial position coordinates to generate a detection result that includes defect type and three-dimensional position information.

8. A textile manufacturing defect detection system based on image processing, applied to multilayer composite textiles, characterized in that, The system includes: The acquisition module is used to acquire surface point cloud data and ultrasonic structural image data of the target textile. The surface point cloud data is generated by acquiring deformation stripe images of the surface of the target textile. The ultrasonic structural image data includes sound wave propagation time data and is generated by receiving penetrating sound wave reflection signals emitted by a multi-frequency ultrasonic probe array. The extraction module is used to filter noise points based on density distribution in the surface point cloud data, extract the surface curvature gradient and normal vector deflection angle to obtain surface deformation candidate regions, perform frequency band separation processing on the ultrasonic structure image data, and extract the reflected wave amplitude and signal delay values ​​to obtain internal abnormal signal regions. A module is established to spatially match the point cloud coordinates of the candidate surface deformation region with the acoustic wave propagation time data of the internal abnormal signal region, and to establish an association mapping relationship between surface morphology features and internal defect features through iterative nearest point matching, wherein the normal vector deflection angle and the reflected wave amplitude form a feature pair; The fusion module is used to perform anisotropic diffusion filtering and wavelet threshold denoising on the surface morphology features and the internal defect features respectively, and to allocate fusion weights according to the ratio of the surface curvature gradient change rate to the reflection wave amplitude attenuation rate. Based on the correlation mapping relationship, the filtered surface morphology features and the denoised internal defect features are fused into a multimodal defect distribution map. The generation module is used to extract geometric shape parameters and reflection intensity parameters based on the multimodal defect distribution map, and match the geometric shape parameters and reflection intensity parameters with a preset defect feature template to generate a detection result including defect type and defect location annotation.

9. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the image processing-based textile production defect detection method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the image processing-based textile production defect detection method as described in any one of claims 1-7.

Images