Textile fabric defect identification method and system based on image detection

Abstract

This application relates to the field of textile inspection technology, and discloses a method and system for defect identification of textile fabrics based on image detection. The method includes: acquiring image data of the textile fabric to be inspected at a preset frequency during the fabric transport process; extracting texture distribution features from the image data; calculating the weave topology attributes of the textile fabric to be inspected based on the texture distribution features; comparing the weave topology attributes with standard topology features corresponding to a variety of preset fabric weave methods to identify the weave category of the textile fabric to be inspected; selecting and loading a defect detection strategy that matches the identified weave category from a variety of preset defect detection strategies, wherein different weave categories correspond to different feature analysis operators; performing defect detection on the image data using the loaded defect detection strategy and outputting the identification results. By performing specialized optimization detection based on the characteristics of different weave methods, the accuracy and robustness of defect identification in mixed production scenarios are improved.

Description

Technical Field

[0001] This application relates to the field of textile inspection technology, and more specifically, to a method and system for identifying textile fabric defects based on image detection. Background Technology

[0002] In automated textile production, using image processing technology to detect fabric defects is a crucial step in ensuring product quality. The conventional inspection process typically involves continuously photographing the fabric on a conveyor belt using industrial cameras, then analyzing the images using a fixed algorithm to identify abnormal areas that deviate from the normal texture, such as stains, broken yarns, or sparse weft. This method performs adequately when processing single, continuous batches of fabric. However, modern textile production lines, in order to improve efficiency, often need to process multiple fabrics with different weaves within a single day, such as switching from simple plain weave fabrics to complex twill fabrics, and then to jacquard fabrics with large-scale repeating patterns.

[0003] The core technical problem facing existing inspection systems lies in their poor adaptability. Most of these systems employ a fixed set of inspection parameters and analysis models, optimized for a specific weave. When the fabric weave changes on the production line, the underlying texture structure of the fabric image also undergoes a fundamental change. At this point, the original fixed model becomes ineffective, unable to distinguish between normal structural textures under the new weave and genuine manufacturing defects. It often misjudges the edges of jacquard patterns or natural transitions in twill weaves as abnormalities, leading to a sharp increase in false alarm rates. To suppress false alarms, operators must raise the judgment threshold, but this reduces the system's sensitivity to subtle defects, resulting in missed detections. This one-size-fits-all inspection method fails to meet quality control requirements in terms of both accuracy and reliability under multi-variety, small-batch production models.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a method and system for identifying defects in textile fabrics based on image detection. This method can solve the technical problem of low defect identification accuracy when processing fabrics with different weaves due to the use of fixed detection strategies in existing technologies.

[0006] In a first aspect, this application provides a method for identifying defects in textile fabrics based on image detection, including: During the fabric transport process, image data of the fabric to be inspected is acquired at a preset frequency, and the texture distribution features of the image data are extracted. Based on texture distribution features, the organizational topological properties of the textile fabric to be detected are calculated. The organization's topological attributes are compared with the standard topological features corresponding to various pre-defined fabric weaves to identify the weave category of the textile fabric to be detected. Select and load a defect detection strategy that matches the identified weave type from a set of preset defect detection strategies. Different weave types correspond to different feature analysis operators. The loaded defect detection strategy is used to detect defects in image data and the recognition results are output.

[0007] By first identifying the weave type of the current fabric before defect detection, and then dynamically loading a matching dedicated defect detection strategy according to the category, the detection logic is adaptively adjusted. This fundamentally solves the problem that fixed strategies cannot adapt to varied textures and enables specialized optimization detection for the characteristics of different weaves, thereby significantly improving the accuracy and robustness of defect identification in mixed production scenarios.

[0008] Furthermore, the steps for extracting texture distribution features from image data include: Divide the image data into multiple local sub-regions; Analyze the gradient direction of pixels in each local sub-region to generate a local gradient direction distribution histogram; For each local sub-region, calculate the gray-level co-occurrence matrix and extract the energy value based on the gray-level co-occurrence matrix; By fusing the local gradient direction distribution histogram with the energy value, texture distribution features are obtained.

[0009] This technical solution integrates gradient direction and gray-level co-occurrence matrix energy to construct a more comprehensive and robust texture distribution feature. This multi-dimensional feature fusion approach can more accurately depict the microstructure of the fabric, providing a high-quality feature foundation for subsequent accurate calculation of organizational topology attributes and identification of weave types, thus improving the reliability of weave identification.

[0010] Furthermore, based on texture distribution features, the steps for calculating the organizational topological properties of the textile fabric to be detected include: The texture distribution features of all local sub-regions in the image data are weighted and averaged to obtain the global organizational topology feature vector of the image data, and the global organizational topology feature vector is used as the organizational topology attribute.

[0011] This technical solution generates a stable and representative global feature vector by weighted averaging of local features. This processing method effectively suppresses the interference of local noise or minor defects on the overall weave judgment, ensuring that the fabric topology attributes can macroscopically and accurately reflect the fabric structure essence of the entire image, making the weave category recognition results more stable and reliable.

[0012] Furthermore, before calculating the organizational topological properties of the textile fabric to be tested, the following steps are also included: During the image data acquisition process, the motion state parameters of the mechanical system used to transmit the textile fabric to be inspected are acquired. These motion state parameters include vibration frequency. The image data is evaluated based on motion state parameters to determine if there is blurring due to physical vibration. If so, the texture distribution features are compensated for or restored.

[0013] By actively acquiring motion parameters of the transmission system to predict and correct image blurring caused by mechanical vibration and other reasons, image quality is guaranteed from the data source, avoiding interference from motion artifacts on texture feature extraction, thereby improving the anti-interference capability and final detection accuracy of the entire recognition system in real industrial environments.

[0014] Furthermore, the steps for compensating or restoring texture distribution features include: Based on the intensity and direction of physical vibration, the texture distribution features in the corresponding directions are weighted and enhanced to counteract the effects of physical vibration.

[0015] This technical solution provides a precise and efficient feature compensation method. Instead of performing complex deblurring calculations on the entire image, it directly enhances the affected feature components at the feature level based on the known direction and intensity of the motion blur. This method has low computational cost, fast response speed, and can accurately counteract the impact of motion blur on key texture information, achieving targeted correction of physical disturbances.

[0016] Furthermore, after comparing the organizational topology attributes with the standard topology features corresponding to various pre-defined fabric weaves to identify the weave category of the textile fabric to be detected, and before selecting and loading a defect detection strategy that matches the identified weave category from a variety of pre-defined defect detection strategies, the method further includes: When the deviation of the current organizational topology attribute from the historical organizational topology attribute average is within a preset fluctuation range, the current weaving category remains unchanged. When the deviation exceeds the preset fluctuation range and the current organization topology attribute points to a new weave category, a weave category switching command is executed, and the new weave category is taken as the final weave category.

[0017] By establishing a fluctuation range, instantaneous jumps in weave identification results caused by minor changes in fabric tension or local texture disturbances are effectively filtered out, avoiding frequent and ineffective switching of detection strategies. Switching is only confirmed when weave characteristics undergo continuous and significant changes, ensuring the stability and continuity of weave category judgment during the production process.

[0018] Furthermore, the steps for executing the weave type switching instruction include: Continuously observe the image data acquired within a preset time period after the deviation exceeds the preset fluctuation range and the current tissue topology attribute points to a new weave category; If the proportion of the calculated tissue topology attributes pointing to the new weave category in the image data acquired within the preset time period exceeds the preset proportion, then the weave category switching command will be executed. During the period of continuous observation of image data acquired within a preset time after the deviation exceeds the preset fluctuation range and the current tissue topology attribute points to a new weave category, the method also includes: Simultaneously, referencing the defect detection strategies corresponding to the current weave type and the new weave type, a temporary defect detection strategy is generated and applied to perform defect detection during the observation period.

[0019] This technical solution not only enhances the reliability of weave switching decisions by setting observation periods and proportion thresholds, but more importantly, it creatively proposes a temporary strategy for inspection during the switching observation period. This design seamlessly fills the inspection gap during the switching process between old and new strategies, ensuring that fabrics in the weave transition area can also be effectively inspected, avoiding the risk of missed inspections due to strategy switching delays, and guaranteeing the integrity of quality control throughout the entire process.

[0020] Furthermore, after executing the weave type switching instruction, the following steps are included: The switching time point of the weave type switching instruction, the difference in the organization topology attributes before and after the switching, and the detection results of the defect detection using the temporary defect detection strategy are encapsulated into conversion segment data and output as conversion segment data.

[0021] This technical solution establishes a comprehensive switchover event traceability mechanism. By packaging and recording key information such as switchover time, characteristic differences, and transition period test results, it provides complete and accurate contextual data for subsequent quality review, process analysis, and big data statistics. This greatly enhances the transparency and traceability of the production process, facilitating precise problem identification and continuous process optimization.

[0022] Furthermore, the preset multiple defect detection strategies include a first defect detection strategy for plain weave, a second defect detection strategy for twill weave, and a third defect detection strategy for jacquard weave. The first defect detection strategy includes loading a feature analysis operator based on local binary pattern and frequency domain transform to extract local texture anomaly features of plain weave fabric from image data, and setting a first judgment tolerance to determine whether the local texture anomaly features constitute a defect. The second defect detection strategy includes loading a feature analysis operator based on directional gradient features and morphological processing to extract directional breakage features of twill fabric from image data, and setting a second judgment tolerance to determine whether the directional breakage features constitute a defect. The third defect detection strategy includes loading a feature analysis operator based on multi-scale decomposition and adaptive threshold segmentation to extract structural repetitive abnormal features of jacquard textile fabrics from image data, and setting a third judgment tolerance to determine whether the structural repetitive abnormal features constitute a defect.

[0023] This technical solution specifically illustrates differentiated detection strategies for different weaving methods and clarifies the progressive relationship of tolerance. This not only confirms the feasibility and necessity of a single weaving method and strategy but also reflects a profound understanding of textile technology: the more complex and varied the fabric structure, the greater its normal texture fluctuation range, thus requiring a higher tolerance. This refined strategy configuration is the core of achieving high-precision detection, ensuring sufficient sensitivity to subtle defects in simple fabrics while being sufficiently tolerant of the normal texture of complex fabrics.

[0024] Secondly, this application also discloses a textile fabric defect recognition system based on image detection, the system comprising: The feature extraction module is used to collect image data of the textile fabric to be detected at a preset frequency during the textile fabric transmission process and extract the texture distribution features of the image data. The attribute determination module is used to calculate the organizational topology attributes of the textile fabric to be detected based on texture distribution features. The category recognition module is used to compare the organization topology attributes with the standard topology features corresponding to various preset fabric weaves in order to identify the weave category of the textile fabric to be detected. The strategy matching module is used to select and load a defect detection strategy that matches the identified weave category from a number of preset defect detection strategies. Different weave categories correspond to different feature analysis operators. The defect recognition module is used to detect defects in image data using the loaded defect detection strategy and output the recognition results.

[0025] This application provides a method and system for textile fabric defect identification based on image detection, which has significant advantages over existing technologies. Existing defect detection systems typically employ fixed analysis algorithms and parameters. When different fabric weave types are used on the production line, the texture characteristics change, making it difficult for the fixed algorithms to adapt, leading to numerous false alarms or missed alarms and low detection accuracy. This application fundamentally solves this problem by introducing an innovative preprocessing step. Before performing defect identification, the acquired fabric image is first analyzed to calculate its unique organizational topology attributes. Based on this, the weave type of the current fabric is accurately identified, such as plain weave, twill weave, or jacquard. Subsequently, the system dynamically selects and loads a defect detection strategy specifically optimized for that weave type from a preset strategy library. This means that the system can call the most suitable analysis operator and judgment criteria for the fine structure of plain weave, the directional texture of twill weave, and the complex pattern of jacquard weave, respectively. Through this adaptive mechanism of identification followed by matching, this application achieves intelligent and dynamic adjustment of the detection strategy, ensuring that the detection system always operates in optimal condition regardless of the type of fabric produced. This not only greatly improves the accuracy of identifying various defects and significantly reduces false alarms and missed alarms, but also enhances the adaptability and robustness of the entire detection system to production changes, providing strong technical support for achieving efficient and high-precision automated quality control of textiles. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a method for identifying defects in textile fabrics based on image detection, provided in an embodiment of this application.

[0027] Figure 2 This is a schematic diagram of the structure of a textile fabric defect recognition system based on image detection, provided in an embodiment of this application.

[0028] Labeling Explanation: 210, Feature Extraction Module; 220, Attribute Determination Module; 230, Category Recognition Module; 240, Strategy Matching Module; 250, Defect Recognition Module. Detailed Implementation

[0029] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0030] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0031] In modern textile production lines, quality inspection is a crucial step. Imagine a large-scale textile factory where a fabric inspection machine is performing high-speed inspection on a roll of freshly woven fabric. This machine is equipped with an automated defect detection system based on image recognition, designed to replace manual labor and achieve uninterrupted, high-efficiency quality control. In the production plan, the factory needs to immediately switch to producing high-grade denim after completing a batch of plain cotton greige fabric. Denim uses a typical twill weave, with a clear, slanted texture. When the fabric on the conveyor belt switches from plain weave to twill, the detection system begins to generate numerous false alarms. The system identifies the inherent, normal diagonal texture of the twill fabric as continuous, large-area scratches or stripe-like defects, causing alarm lights to flash continuously and forcing frequent production line shutdowns. Operators must manually intervene to reduce the system's detection sensitivity to minimize false alarms, but this directly leads to the risk of missing truly present minor defects, such as tiny broken yarns or oil stains. This mismatch in detection strategies caused by changes in fabric weaving is a common technical bottleneck in the field of automated fabric inspection, which seriously affects production efficiency and product quality stability.

[0032] To solve the aforementioned technical problems, firstly, see [reference needed] Figure 1 This application provides a method for identifying defects in textile fabrics based on image detection, the method comprising: S1. During the fabric conveying process, image data of the fabric to be detected is collected at a preset frequency, and the texture distribution features of the image data are extracted. S2. Based on texture distribution features, calculate the organizational topological properties of the textile fabric to be detected; S3. Compare the organization's topological attributes with the standard topological features corresponding to various preset fabric weaves to identify the weave category of the textile fabric to be detected. S4. Select and load a defect detection strategy that matches the identified weave type from a set of preset defect detection strategies. Different weave types correspond to different feature analysis operators. S5. Use the loaded defect detection strategy to perform defect detection on the image data and output the recognition results.

[0033] Among them, texture distribution features refer to a set of data that can quantitatively describe the surface texture structure characteristics of textile fabric images after mathematical analysis. It is not the original pixel value of the image, but an abstract expression of the spatial arrangement of pixels, such as describing the direction of the yarn, the density of the interlacing, the roughness or regularity of the texture, etc.

[0034] Organizational topology attributes are numerical fingerprints derived from texture distribution characteristics and used to characterize the macroscopic weaving structure of fabrics. They aim to eliminate the influence of non-structural factors such as light and minute deformations, and stably reflect whether the fabric is plain weave, twill weave, or other more complex weaving structures.

[0035] Weave type refers to the classification of fabric weaving methods based on textile technology, such as plain weave, twill weave, satin weave, jacquard, etc. In this application, this is a label automatically identified and output by the system based on organizational topology attributes.

[0036] A defect detection strategy is a software package or configuration set that integrates specific algorithms, parameters, and decision logic. It is not a single algorithm, but a complete solution tailored to a specific weaving type. It includes which feature analysis operators to use to find anomalies and what tolerance level to set to determine whether an anomaly constitutes a defect.

[0037] In one specific embodiment, the method is deployed in a vision inspection system installed on a high-speed fabric inspection machine. This system mainly includes an image acquisition unit and an industrial control computer. The image acquisition unit consists of a linear industrial camera and a set of strip LED light sources, spanning a two-meter-wide fabric conveyor track. As the fabric passes beneath the camera at a speed of sixty meters per minute, the linear camera continuously scans the fabric surface at a frequency of two thousand lines per second, stitching together high-resolution continuous image data, each image covering approximately one meter of fabric. This image data is transmitted in real-time to the industrial control computer for processing via a high-speed data interface, such as CameraLink or a 10 Gigabit Ethernet port.

[0038] After receiving a complete frame of image data, the industrial control computer begins the process of extracting texture distribution features. To comprehensively describe the texture, one possible approach is to calculate the local standard deviation of the image. The computer divides the entire image into several non-overlapping local sub-regions, for example, each 32x32 pixels in size. For each sub-region, the standard deviation of all pixel grayscale values ​​within it is calculated. The magnitude of the standard deviation reflects the texture contrast or roughness of that local region. For example, for a plain weave fabric with a uniform texture, the standard deviation values ​​of most of its sub-regions will be relatively similar and small; while for a jacquard fabric with a distinct pattern, the standard deviations of different sub-regions will show significant differences. In this way, the entire image is transformed into a feature map composed of local standard deviations, which reflects the texture distribution of the image to a certain extent.

[0039] Next, based on the extracted texture distribution features, the system needs to calculate the organizational topology of the textile fabric to be detected. A simple implementation is to take the arithmetic mean of the texture distribution feature values ​​(local standard deviations in this example) of all local sub-regions obtained in the previous step. That is, add up the standard deviations of all sub-regions and then divide by the total number of sub-regions to obtain a single global average standard deviation. This global average standard deviation is then used as the organizational topology of the fabric represented by the image. This value macroscopically reflects the overall texture complexity of the fabric.

[0040] Subsequently, the system identifies the weave type. A standard topological feature library is pre-built in the system's memory. This library is obtained through offline analysis of a large number of standard sample fabrics with known weave types. For example, the library might store the following correspondences: the standard weave topological attribute (mean standard deviation) ranges from 5.0 to 8.0 for plain weave fabrics; the standard weave topological attribute ranges from 12.0 to 18.0 for twill weave fabrics; and the standard weave topological attribute range for complex jacquard fabrics might be above 25.0. When the system calculates the weave topological attribute of the current fabric to be detected to be 15.3, it compares this value with the standard range in the library. Since 15.3 falls within the range of 12.0 to 18.0, the system identifies the weave type of the current fabric as twill.

[0041] Once the weave type is identified, the system immediately selects and loads the matching strategy from a set of preset defect detection strategies. This process is similar to a computer choosing different programs to open files based on their type. The system internally maintains a strategy mapping table, for example: Weave type plain weave corresponds to loading strategy A.dll; Twill weave type corresponding to loading strategy B.dll; The weave type jacquard corresponds to the loading strategy C.dll.

[0042] In this example, since the identified weave type is twill, the system executes an instruction to load strategy B.dll into memory. This dynamic link library file encapsulates feature analysis operators and related parameters specifically designed for analyzing twill fabrics.

[0043] Finally, the system uses the loaded defect detection strategy to detect defects in the original image data. The loaded strategy B may include a directional filtering operator designed to enhance and detect textures inconsistent with the main direction of the twill weave. For example, if the twill is at a 45-degree angle, the operator will focus on finding linear anomalies in the width or length direction of the fabric being detected, which typically corresponds to warp or weft breaks. The strategy also includes a defect threshold for twill fabrics, which is higher than the threshold used for plain weaves to tolerate normal brightness fluctuations caused by variations in yarn float length. When the detection algorithm finds an abnormal response value in a region of the image exceeding this threshold, the system determines that the region has a defect and generates a defect record. This record contains the precise location (in meters) of the defect within the fabric roll, an image slice of the defect, and a preliminary determination of the defect type (e.g., warp break). These identification results are then output to the user interface and stored in the quality database.

[0044] Through the above process, when the production line switches from plain weave to twill weave, the system can automatically identify the change in weave and seamlessly switch from using strategy A for plain weave to using strategy B for twill weave. In this way, the system will no longer misreport normal twill textures as defects, while maintaining high sensitivity to actual defects on twill fabric. This solves the problem that fixed strategies in existing technologies cannot adapt to multi-variety production, significantly improving the accuracy and automation level of detection.

[0045] Furthermore, the steps for extracting texture distribution features from image data include: Divide the image data into multiple local sub-regions; Analyze the gradient direction of pixels in each local sub-region to generate a local gradient direction distribution histogram; For each local sub-region, calculate the gray-level co-occurrence matrix and extract the energy value based on the gray-level co-occurrence matrix; By fusing the local gradient direction distribution histogram with the energy value, texture distribution features are obtained.

[0046] This improvement is made because a single texture descriptor (such as the aforementioned local standard deviation) often misses the mark. For example, standard deviation reflects texture contrast but fails to capture its directionality. For twill fabrics with clear directionality, this directional information is crucial for distinguishing them from other weaves. This improvement aims to construct a more comprehensive and robust texture descriptor by integrating two complementary features.

[0047] In practice, for each local sub-region, the system first calculates the gradient of each pixel. The gradient consists of two components: magnitude and direction; here, the focus is primarily on the gradient direction. The system uses operators such as Sobel to calculate the gradient components in the x and y directions (corresponding to the length and width directions of the fabric being detected, respectively), and then calculates the gradient direction using the arctangent function. For ease of statistical analysis, the gradient direction is quantized into a limited number of intervals, for example, divided into eight 45-degree intervals. The system counts the number of pixels falling into each directional interval, thereby generating an 8-dimensional local gradient direction distribution histogram. This histogram vector effectively describes the main direction of the texture within the sub-region. For example, for twill fabric, its histogram will have a significant peak in the intervals corresponding to the diagonal direction.

[0048] Simultaneously, the system also calculates the gray-level co-occurrence matrix (GLCM) for the same local sub-region. The GLCM is a classic texture analysis tool that describes the second-order statistical properties of texture by statistically analyzing the frequency of occurrence of pixel pairs of different gray levels in a specific spatial relationship (e.g., one pixel apart, along the length direction). Based on the calculated GLCM, multiple texture metrics can be extracted; this scheme selects to extract the energy value. The energy value, also known as the angular second moment, is the sum of the squares of the elements in the GLCM, reflecting the uniformity of the image's gray-level distribution and the coarseness of the texture. A larger energy value indicates a more regular and uniform texture.

[0049] Finally, the system fuses these two feature sets. A direct fusion method is vector concatenation. This involves concatenating the 8-dimensional histogram of local gradient orientation distribution with the 1-dimensional energy value to form a 9-dimensional fused feature vector. This 9-dimensional vector contains both the directional information of the texture and the information on the uniformity and coarseness of the texture. Compared to a single feature, it can more accurately and stably depict the microstructure of the fabric, providing higher-quality input for subsequent calculation of organizational topology attributes and weave recognition.

[0050] Furthermore, based on texture distribution features, the steps for calculating the organizational topological properties of the textile fabric to be detected include: The texture distribution features of all local sub-regions in the image data are weighted and averaged to obtain the global organizational topology feature vector of the image data, and the global organizational topology feature vector is used as the organizational topology attribute.

[0051] The arithmetic mean method used in the aforementioned embodiments, while simple, suffers from the problem of treating each local sub-region equally. If a sub-region in the image happens to have abnormal texture features due to uneven lighting, lens smudges, or temporary fabric wrinkles, this outlier will indiscriminately affect the final global average, potentially leading to deviations in tissue topology attributes or even incorrect weave identification. The purpose of using weighted averaging is to introduce a mechanism for differentiated treatment, giving greater weight to those more reliable and representative sub-regions.

[0052] In practical implementation, there are several ways to determine the weight coefficients. One approach is based on the location of the sub-region. For example, the image quality of the central region can be considered the highest, and it is least affected by lens distortion and vignetting. Therefore, a higher weight can be assigned to sub-regions in the central region, decreasing towards the edge regions. Another, more adaptive approach is to determine the weights based on the characteristics of the sub-region itself. For example, the energy value of the gray-level co-occurrence matrix calculated in the previous step can be used as the weight. Sub-regions with higher energy values ​​represent clearer and more regular textures, and the extracted texture distribution features are more reliable, thus they can be given higher weights. Conversely, a sub-region with a very low energy value may correspond to a blurry or chaotic texture, and its feature reliability is low, so it should be given a lower weight.

[0053] The calculation process is as follows: For the i-th sub-region in the image, its texture distribution feature vector is Vi, and its corresponding weight is Wi. The global organization topology feature vector V_global is calculated as: V_global=(Σ(Wi*Vi)) / ΣWi. Through this weighted averaging, the obtained global organization topology feature vector can effectively suppress local noise and anomaly interference, more robustly reflecting the overall and essential weaving structure of the fabric, thereby significantly improving the stability and reliability of weave category recognition.

[0054] Furthermore, before calculating the organizational topological properties of the textile fabric to be tested, the following steps are also included: During the image data acquisition process, the motion state parameters of the mechanical system used to transmit the textile fabric to be inspected are acquired. These motion state parameters include vibration frequency. The image data is evaluated based on motion state parameters to determine if there is blurring due to physical vibration. If so, the texture distribution features are compensated for or restored.

[0055] This step introduces the idea of ​​feedforward control, which involves actively monitoring physical sources that may cause image quality degradation and intervening before the problem affects the final decision.

[0056] In practice, the system integrates a series of external sensors. For example, a triaxial accelerometer is installed on the camera bracket or the frame of the fabric inspection machine to monitor the vibration acceleration and frequency of the equipment in the x, y, and z directions in real time.

[0057] The industrial control computer continuously collects the output data from these sensors. The system has an internal vibration threshold; for example, when the accelerometer detects a vibration amplitude exceeding 0.5g in the width direction, the system determines that the currently acquired image frame may be significantly affected by motion blur. At this point, the system triggers a compensation action.

[0058] Specifically, the steps for compensating or restoring texture distribution features include: Based on the intensity and direction of physical vibration, the texture distribution features in the corresponding directions are weighted and enhanced to counteract the effects of physical vibration.

[0059] The intensity and direction of physical vibration refer to the degree and main direction of vibration generated by the mechanical system when conveying the textile fabric to be tested. Specifically, inertial measurement units (IMUs) such as accelerometers and gyroscopes can be used to monitor the motion state of the mechanical system in real time. By performing signal processing such as Fourier transform on the sensor data, parameters such as vibration frequency and amplitude can be extracted, thereby assessing the intensity of the vibration. The direction of vibration can be determined by analyzing the differences in data from multiple sensors along different axes. For example, if the acceleration value in the length direction is significantly higher than that in the width direction, it indicates that the vibration mainly occurs in the length direction.

[0060] Texture distribution features refer to information describing the surface texture structure of a fabric extracted from image data using image processing techniques. Specifically, methods such as Local Binary Pattern (LBP), Gray-Level Co-occurrence Matrix (GLCM), and Gabor filters can be used for extraction. For example, LBP can describe the contrast and structure of local textures, while GLCM can reflect the fineness, contrast, and directionality of the texture. These features are typically represented in vector or matrix form and can quantitatively describe the texture information of the fabric.

[0061] Weighted enhancement refers to applying greater weight to the most severely affected parts of the texture distribution features based on the specific circumstances of physical jitter. This can be achieved as follows: First, determine the direction and degree of impact on texture features based on the intensity and direction of the physical jitter. For example, if jitter causes blurring of texture along the length direction, then texture features along the length direction need to be enhanced more significantly. Second, design a weighting function that assigns different weights to texture features in different directions based on jitter parameters (intensity, direction). For example, assign higher weights to directions more affected by jitter and lower weights to directions less affected. Finally, apply these weights to the original texture distribution features, using methods such as weighted summation or weighted averaging, to effectively restore and highlight the damaged texture information.

[0062] In a preferred embodiment, on a textile fabric production line, an industrial camera captures image data of the textile fabric to be inspected on a conveyor belt at a frequency of 30 frames per second. Simultaneously, a triaxial accelerometer is installed on the conveyor belt's mechanical system to monitor the vibration frequency and amplitude of the system in real time. When the accelerometer detects vibration in the mechanical system, with a vibration frequency between 50Hz and 100Hz and an amplitude exceeding 0.5mm, the system determines that the image data may be blurred due to physical vibration.

[0063] At this point, the system will further analyze the intensity and direction of the physical jitter based on the data from the accelerometer. For example, if the acceleration value in the length direction is significantly higher than that in the width direction, it is determined that the jitter mainly occurs in the length direction and is of moderate intensity.

[0064] When extracting texture distribution features from image data, the system uses a Gray-Level Co-occurrence Matrix (GLCM) to calculate texture features, including energy, contrast, correlation, and entropy. Based on the previously determined physical vibration direction as the length direction, the system performs weighted enhancement on the parameters in the GLCM that reflect texture features along the length direction. Specifically, the weight coefficient for texture features along the length direction is set to 1.5, while the weight coefficients for texture features in other directions remain at 1.0. Through this weighted enhancement process, the damaged texture information along the length direction in the blurred image is effectively restored and highlighted, allowing the processed texture distribution features to more accurately reflect the true fabric structure.

[0065] Through the above technical solution, this application solves the problem that in the process of textile fabric defect identification, image blurring caused by physical vibration of the mechanical system affects the accuracy of texture distribution features. This solution counteracts the impact of physical vibration by weighting and enhancing the texture distribution features in the corresponding direction based on the intensity and direction of the physical vibration, thereby ensuring the accuracy of subsequent defect detection.

[0066] In a preferred embodiment, after the step of comparing the organization topology attributes with standard topology features corresponding to a variety of preset fabric weaves to identify the weave category of the textile fabric to be detected, and before the step of selecting and loading a defect detection strategy that matches the identified weave category from a variety of preset defect detection strategies, the method further includes: When the deviation of the current organizational topology attribute from the historical organizational topology attribute average is within a preset fluctuation range, the current weaving category remains unchanged. When the deviation exceeds the preset fluctuation range and the current organization topology attribute points to a new weave category, a weave category switching command is executed, and the new weave category is taken as the final weave category.

[0067] This step introduces a hysteresis or buffering mechanism to enhance the stability of weave category determination. Specifically, the system maintains a historical mean of fabric topology attributes, which can be a moving average of the fabric topology attributes calculated based on the past N frames (e.g., N=100) of images. This mean represents the stable weave characteristics of the fabric in the recent past.

[0068] Meanwhile, the system sets a preset fluctuation range for each weave type. This range defines the reasonable range of normal fluctuation in the organization's topology attributes under that weave. For example, suppose the system currently identifies a plain weave with a historical average of 7.5, and the preset fluctuation range is ±1.5. This means that as long as the newly calculated organization topology attribute value falls within the range of [6.0, 9.0], even if the value is theoretically closer to a certain edge value of a twill weave, the system will still consider the current weave to be a plain weave and maintain the current defect detection strategy unchanged.

[0069] The system only considers a genuine weave change to have occurred and triggers a weave category switching command when a newly calculated organizational topology attribute value, such as 13.0, exceeds the fluctuation range of [6.0, 9.0] but clearly falls within the preset twill weave feature range (e.g., its standard range is 12.0-18.0). This mechanism effectively filters out instantaneous feature drift caused by factors such as tension changes and slight wrinkles, avoids jittering of the detection strategy near the critical point, and ensures that the system only responds to the switch when the weave undergoes a continuous and significant change.

[0070] Furthermore, the steps for executing the weave type switching instruction include: Continuously observe the image data acquired within a preset time period after the deviation exceeds the preset fluctuation range and the current tissue topology attribute points to a new weave category; If the proportion of the calculated tissue topology attributes pointing to the new weave category in the image data acquired within the preset time period exceeds the preset proportion, then the weave category switching command will be executed. During the period of continuous observation of image data acquired within a preset time after the deviation exceeds the preset fluctuation range and the current tissue topology attribute points to a new weave category, the method also includes: Simultaneously, referencing the defect detection strategies corresponding to the current weave type and the new weave type, a temporary defect detection strategy is generated and applied to perform defect detection during the observation period.

[0071] Specifically, after the previous step triggers the switching command, the system does not immediately switch strategies but enters a preset observation period, such as 2 seconds. During these 2 seconds, the system continues to acquire and analyze images, calculating the organization topology attributes for each frame. The system counts how many frames among all the image frames acquired during these 2 seconds have organization topology attributes pointing to the new weave category (e.g., twill). The system internally sets a confirmation ratio, such as 85%. Only when the percentage of frames pointing to twill exceeds 85% will the system finally confirm the switch and officially load the defect detection strategy from the plain weave strategy to the twill weave strategy. If the percentage does not reach this ratio, the system will cancel the switching command, considering it to be a brief disturbance, and continue using the original plain weave strategy.

[0072] More importantly, during this 2-second observation period, the fabric continues to be conveyed, and quality inspection cannot be stopped. At this time, neither the old nor the new strategy is optimal. Therefore, this application proposes to generate and apply a temporary defect detection strategy. This temporary strategy can be generated in several ways: One approach is the union strategy, which involves running the core detection operators from both the old and new strategies simultaneously and merging the results. This provides comprehensive coverage but is computationally expensive.

[0073] Another, better approach is a hybrid strategy, which dynamically adjusts the weights of the old and new strategies based on the organization topology attribute values ​​of the current frame. For example, if the attribute values ​​of the current frame are closer to the new weave, the detection results of the new strategy will have a higher weight.

[0074] Another approach is a conservative strategy, which involves using only one general-purpose defect detection module that is insensitive to weave patterns during the observation period. For example, an algorithm specifically designed to detect large areas of oil stains, color differences, or holes may be used, while temporarily abandoning the detection of minor weave-related defects.

[0075] By applying temporary strategies, this solution ensures that the fabric can still be effectively inspected during the transition area of ​​weaving method switching, avoiding blind spots in inspection caused by decision delays and ensuring the continuity and integrity of quality control throughout the entire process.

[0076] Furthermore, after executing the weave type switching instruction, the method also includes: The switching time point of the weave type switching instruction, the difference in the organization topology attributes before and after the switching, and the detection results of the defect detection using the temporary defect detection strategy are encapsulated into conversion segment data and output as conversion segment data.

[0077] Once a successful weave changeover is completed, the system automatically generates a structured data packet, known as the transition section data. This data packet records all key information related to the changeover. A specific instance of transition section data may be in JSON format. This data packet is stored in the quality database and associated with the corresponding fabric batch number. Quality management personnel can subsequently query this data to precisely review each weave changeover process. For example, if an abnormally high defect rate is found in the weave changeover area of ​​a particular batch, this data can be used to trace back to the specific time point and characteristic changes, providing precise data support for analyzing process issues (such as whether the loom is stable when changing fabric types) or optimizing temporary inspection strategies.

[0078] Furthermore, the preset multiple defect detection strategies include a first defect detection strategy for plain weave, a second defect detection strategy for twill weave, and a third defect detection strategy for jacquard weave. The first defect detection strategy includes loading a feature analysis operator based on local binary pattern and frequency domain transform to extract local texture anomaly features of plain weave fabric from image data, and setting a first judgment tolerance to determine whether the local texture anomaly features constitute a defect. The second defect detection strategy includes loading a feature analysis operator based on directional gradient features and morphological processing to extract directional breakage features of twill fabric from image data, and setting a second judgment tolerance to determine whether the directional breakage features constitute a defect. The third defect detection strategy includes loading a feature analysis operator based on multi-scale decomposition and adaptive threshold segmentation to extract structural repetitive abnormal features of jacquard textile fabrics from image data, and setting a third judgment tolerance to determine whether the structural repetitive abnormal features constitute a defect.

[0079] The first defect detection strategy is a detection scheme specifically designed for plain weave fabrics. Plain weave fabrics are characterized by their simple structure and uniform texture, and their defects often manifest as local texture anomalies. Local Binary Pattern (LBP) is an operator used to describe local texture features of an image. It generates a binary pattern by comparing the gray values ​​of a pixel with its neighboring pixels, effectively capturing microscopic changes in local texture and showing high sensitivity to abrupt changes in local texture caused by yarn breaks, stains, etc., in plain weave fabrics. Frequency domain transformation, such as Fourier transform, can convert the image from the spatial domain to the frequency domain, thereby analyzing the periodicity and directionality of the texture and further enhancing the ability to identify abnormal textures. By extracting local texture anomaly features of plain weave fabrics from image data using these operators, the unique defect types of plain weave fabrics can be accurately located. Simultaneously, a first judgment tolerance is set to determine whether local texture anomalies constitute a defect. The first judgment tolerance can be a numerical threshold; for example, when a certain index of the LBP feature or frequency domain feature exceeds this threshold, it is judged as a defect. This threshold is set based on the characteristics of plain weave fabrics through experiments or expert experience, aiming to avoid misjudging normal texture fluctuations as defects and to prevent overlooking subtle defects.

[0080] The second defect detection strategy is a detection scheme specifically designed for twill fabrics. Twill fabrics are characterized by their distinct diagonal weave, and defects may manifest as breaks or deformations in this weave. Oriented gradient features capture edge direction information by calculating the rate of grayscale change of image pixels in different directions, demonstrating strong recognition capabilities for directional breaks or distortions in twill fabrics caused by yarn breaks, uneven weaving, etc. Morphological processing involves a series of image shape-based processing techniques, such as erosion, dilation, opening, and closing operations. These can be used to connect broken weaves, fill holes, or smooth edges, thereby more clearly extracting the directional breakage features of twill fabrics and avoiding misjudgments caused by subtle changes in the weave. A second judgment tolerance is set to determine whether the directional breakage features constitute a defect. This second judgment tolerance can be a threshold based on the intensity of the directional gradient or the results of morphological processing, allowing the system to adjust its sensitivity to directional breaks according to the characteristics of the twill fabric, ensuring detection accuracy.

[0081] The third defect detection strategy is a detection scheme specifically designed for jacquard fabrics. Jacquard fabrics are characterized by complex, large-scale repeating patterns, and defects may manifest as structural repetition anomalies in these patterns. Multi-scale decomposition, such as wavelet transform or pyramid decomposition, can analyze images at different scales, capturing structural information of the jacquard pattern at different levels, which is highly effective for identifying local or global anomalies. Adaptive threshold segmentation is a method that automatically adjusts the threshold based on the grayscale distribution of local image regions, thereby more accurately segmenting the structure of the jacquard pattern from complex backgrounds and extracting structural repetition anomalies in jacquard textile fabrics, such as pattern misalignment, missing parts, or deformation. A third judgment tolerance is set to determine whether structural repetition anomalies constitute a defect. The third judgment tolerance can be a threshold based on multi-scale decomposition coefficients or segmentation region features, allowing the system to set a flexible judgment criterion according to the complexity of the jacquard fabric, effectively distinguishing normal pattern variations from genuine structural defects.

[0082] As a preferred embodiment, a defect detection strategy for three common weave types—plain weave, twill weave, and jacquard—is pre-configured in a textile fabric defect detection system.

[0083] When the system determines through the image recognition module that the fabric to be detected is plain weave, the system automatically loads the first defect detection strategy. This strategy activates an LBP feature extractor and a Fourier transform-based frequency domain analysis module. The LBP feature extractor scans the image in a 3x3 window, calculates the grayscale difference between each pixel and its eight neighboring pixels, generates an 8-bit binary pattern, and calculates a local histogram. The frequency domain analysis module performs a two-dimensional Fourier transform on the image, analyzing its spectral energy distribution and directionality. These extracted features are fed into a classifier and compared with a preset first judgment tolerance (e.g., a peak in the LBP histogram deviates from the normal range by no more than 15%, or an abnormal increase in frequency domain energy in a specific high-frequency region does not exceed 20%). If the feature value exceeds the tolerance range, it is judged as a local texture abnormality defect in the plain weave fabric.

[0084] When the system identifies the fabric as a twill weave, it loads a second defect detection strategy. This strategy activates a gradient directional calculation module and a morphological processing module. The gradient directional calculation module uses the Sobel operator to calculate the gradient intensity and direction of the image in multiple directions, such as 0°, 45°, 90°, and 135°. The morphological processing module uses opening operations (erosion followed by dilation) to remove minor noise and broken lines in the image. The extracted gradient directional features and morphological processing results are fed into a classifier and compared with a preset second tolerance (e.g., the gradient intensity in a certain direction decreases by no more than 30% in a continuous region, or the length of the broken line after morphological processing does not exceed 5 pixels). If the feature value exceeds the tolerance range, it is determined to be a directional break defect in the twill fabric.

[0085] When the system identifies the fabric as jacquard, it loads a third defect detection strategy. This strategy activates a multi-scale wavelet decomposition module and an adaptive threshold segmentation module. The multi-scale wavelet decomposition module performs a three-level wavelet decomposition on the image, obtaining low-frequency and high-frequency sub-band images at different scales to capture the macroscopic and microscopic structural information of the jacquard pattern. The adaptive threshold segmentation module automatically determines the segmentation threshold based on the grayscale histogram of each local region, accurately separating the jacquard pattern from the background. The extracted multi-scale features and the segmented pattern structural features are fed into a classifier and compared with a preset third judgment tolerance (e.g., the variance of wavelet coefficients in a specific sub-band deviates from the normal value by no more than 25%, or the repetitive periodicity error of the segmented pattern does not exceed 10%). If the feature value exceeds the tolerance range, it is judged as a structural repetitive abnormal defect in the jacquard fabric.

[0086] Through the above technical solution, this application can provide customized defect detection solutions for textile fabrics with different weave types, solving the problem of poor adaptability of fixed detection models in the prior art when dealing with fabrics with different weaves. This solution designs specific defect detection strategies for different weave types such as plain weave, twill weave, and jacquard, and equips each strategy with specific feature analysis operators and decision tolerances, thereby significantly improving the accuracy and specificity of defect detection for multiple types of textile fabrics. This solution avoids misjudging the edges of jacquard patterns or natural transitions in twill weaves as anomalies, reducing the false alarm rate, while improving sensitivity to subtle defects and reducing missed detections, thus meeting the refined requirements for defect detection of multiple types of fabrics.

[0087] Secondly, see Figure 2 This application also provides an image detection-based textile fabric defect recognition system, the system comprising: The feature extraction module 210 is used to collect image data of the textile fabric to be detected at a preset frequency during the textile fabric conveying process and extract the texture distribution features of the image data. The attribute determination module 220 is used to calculate the organizational topology attributes of the textile fabric to be detected based on the texture distribution features. The category recognition module 230 is used to compare the organization topology attributes with the standard topology features corresponding to a variety of preset fabric weaves in order to identify the weave category of the textile fabric to be detected. The strategy matching module 240 is used to select and load a defect detection strategy that matches the identified weave type from a plurality of preset defect detection strategies, wherein different weave types correspond to different feature analysis operators; The defect recognition module 250 is used to perform defect detection on image data using the loaded defect detection strategy and output the recognition results; The physical vibration compensation module is used before calculating the tissue topology properties of the textile fabric to be tested. During the image data acquisition process, the motion state parameters of the mechanical system used to transmit the textile fabric to be inspected are acquired. These motion state parameters include vibration frequency. The motion state parameters are used to assess whether the image data is blurred due to physical vibration. If so, the texture distribution features are compensated or restored. The new category identification module is used to maintain the current weave category unchanged when the deviation of the current organizational topology attribute from the average historical organizational topology attribute is within a preset fluctuation range, after the step of comparing the organizational topology attribute with the standard topology features corresponding to a variety of preset fabric weaves to identify the weave category of the textile fabric to be detected, and before the step of selecting and loading a defect detection strategy that matches the identified weave category from a variety of preset defect detection strategies. When the deviation exceeds the preset fluctuation range and the current organization topology attribute points to a new weave category, a weave category switching command is executed, and the new weave category is taken as the final weave category.

[0088] This technical solution provides a physical device capable of executing the aforementioned method. Through modular functional division, the hardware or software structure required for adaptive defect detection is clearly defined, providing a clear implementation blueprint for transforming this method into practically usable industrial inspection equipment, and possessing strong engineering practical value.

[0089] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for identifying defects in textile fabrics based on image detection, characterized in that, The method includes: During the fabric transport process, image data of the fabric to be tested is acquired at a preset frequency, and the texture distribution features of the image data are extracted. Based on the texture distribution features, the organizational topological properties of the textile fabric to be detected are calculated; The organization topology attributes are compared with the standard topology features corresponding to a variety of preset fabric weaves to identify the weave category of the textile fabric to be detected. Select and load a defect detection strategy that matches the identified weave type from a set of preset defect detection strategies, wherein different weave types correspond to different feature analysis operators; The image data is used to detect defects using the loaded defect detection strategy, and the recognition results are output.

2. The method for identifying textile fabric defects based on image detection according to claim 1, characterized in that, The step of extracting the texture distribution features of the image data includes: The image data is divided into multiple local sub-regions; The gradient direction of each pixel in each local sub-region is statistically analyzed to generate a local gradient direction distribution histogram. For each of the local sub-regions, a gray-level co-occurrence matrix is ​​calculated, and energy values ​​are extracted based on the gray-level co-occurrence matrix; The texture distribution feature is obtained by fusing the local gradient direction distribution histogram with the energy value.

3. The method for identifying textile fabric defects based on image detection according to claim 2, characterized in that, The step of calculating the organizational topological properties of the textile fabric to be detected based on the texture distribution features includes: The texture distribution features of all the local sub-regions in the image data are weighted and averaged to obtain the global organizational topology feature vector of the image data, and the global organizational topology feature vector is used as the organizational topology attribute.

4. The method for identifying textile fabric defects based on image detection according to claim 1, characterized in that, Before calculating the organizational topological properties of the textile fabric to be detected, the following steps are also included: During the acquisition of the image data, the motion state parameters of the mechanical system used to transmit the textile fabric to be detected are obtained, and the motion state parameters include vibration frequency. The motion state parameters are used to assess whether the image data is blurred due to physical vibration. If so, the texture distribution features are compensated or restored.

5. The method for identifying textile fabric defects based on image detection according to claim 4, characterized in that, The step of compensating or restoring the texture distribution features includes: Based on the intensity and direction of the physical vibration, the texture distribution features in the corresponding direction are weighted and enhanced to counteract the effects of the physical vibration.

6. The method for identifying textile fabric defects based on image detection according to claim 1, characterized in that, After the step of comparing the organizational topology attributes with standard topology features corresponding to a variety of preset fabric weaves to identify the weave category of the textile fabric to be detected, and before the step of selecting and loading a defect detection strategy that matches the identified weave category from a variety of preset defect detection strategies, the method further includes: When the deviation of the current organizational topology attribute from the historical average organizational topology attribute is within a preset fluctuation range, the current weaving category is maintained unchanged. When the deviation exceeds the preset fluctuation range and the current organization topology attribute points to the new weave category, a weave category switching instruction is executed, and the new weave category is taken as the final determined weave category.

7. The method for identifying textile fabric defects based on image detection according to claim 6, characterized in that, The steps for executing the weaving type switching instruction include: The image data acquired within a preset time period after the deviation exceeds the preset fluctuation range and the current tissue topology attribute points to the new weave category is continuously observed; If, within the preset time period, the proportion of the calculated tissue topology attribute pointing to the new weave type in the image data acquired exceeds a preset proportion, then the weave type switching instruction is executed. During the period of continuously observing the image data acquired within a preset time after the deviation exceeds the preset fluctuation range and the current tissue topology attribute points to a new weave category, the method further includes: Simultaneously, referencing both the current defect detection strategy corresponding to the weave type and the new defect detection strategy corresponding to the weave type, a temporary defect detection strategy is generated and applied to perform defect detection during the observation period.

8. The method for identifying textile fabric defects based on image detection according to claim 7, characterized in that, After executing the weave type switching instruction, the following steps are included: The switching time point of the weave type switching instruction, the difference in the tissue topology attributes before and after the switching, and the detection results of the defect detection using the temporary defect detection strategy are encapsulated into conversion segment data and the conversion segment data is output.

9. The method for identifying textile fabric defects based on image detection according to claim 1, characterized in that, The preset multiple defect detection strategies include a first defect detection strategy for plain weave, a second defect detection strategy for twill weave, and a third defect detection strategy for jacquard weave. The first defect detection strategy includes loading a feature analysis operator based on local binary pattern and frequency domain transformation to extract local texture anomaly features of plain weave fabric from the image data, and setting a first judgment tolerance to determine whether the local texture anomaly features constitute a defect. The second defect detection strategy includes loading a feature analysis operator based on directional gradient features and morphological processing to extract directional breakage features of twill fabric from the image data, and setting a second judgment tolerance to determine whether the directional breakage features constitute a defect. The third defect detection strategy includes loading a feature analysis operator based on multi-scale decomposition and adaptive threshold segmentation to extract structural repeatability anomalies of jacquard textile fabrics from the image data, and setting a third judgment tolerance to determine whether the structural repeatability anomalies constitute a defect.

10. A textile fabric defect identification system based on image detection, used to execute the textile fabric defect identification method based on image detection as described in any one of claims 1 to 9, characterized in that, The system includes: The feature extraction module is used to collect image data of the textile fabric to be detected at a preset frequency during the textile fabric conveying process, and extract the texture distribution features of the image data. The attribute determination module is used to calculate the organizational topology attributes of the textile fabric to be detected based on the texture distribution features. The category identification module is used to compare the organization topology attributes with the standard topology features corresponding to a variety of preset fabric weaves in order to identify the weave category of the textile fabric to be detected. The strategy matching module is used to select and load a defect detection strategy that matches the identified weave type from a plurality of preset defect detection strategies, wherein different weave types correspond to different feature analysis operators; The defect identification module is used to perform defect detection on the image data using the loaded defect detection strategy and output the identification results.

Images