A Method and System for Nonwoven Fabric Quality Inspection Based on Image Recognition and AI Analysis

Abstract

This invention belongs to the field of quality inspection technology and provides a method and system for nonwoven fabric quality inspection based on image recognition and AI analysis. It obtains a steady-state image by preprocessing the target grayscale image of the nonwoven fabric surface and performs edge detection to obtain multiple grayscale sub-regions. Then, multiple clustering domains are constructed based on grayscale primitives. The similarity fitting degree of all clustering domains is calculated, and abnormal clustering domains are screened out. Finally, the connectivity feature value between the abnormal clustering domain and adjacent clustering domains is calculated, and the presence of fiber nodules in the abnormal clustering domain is determined based on this feature value. This method can quickly and accurately locate areas that may have defects, reduce manual intervention, and improve the automation and accuracy of inspection. Therefore, while ensuring product quality, it also improves production efficiency and avoids potential quality problems.

Description

Technical Field

[0001] This invention belongs to the field of quality inspection technology, specifically relating to a nonwoven fabric quality inspection method and system based on image recognition and AI analysis. Background Technology

[0002] Non-woven fabric is a new type of environmentally friendly material with advantages such as good breathability, moisture resistance, softness, lightness, non-flammability, easy decomposition, non-toxicity, non-irritation, recyclability, and low price. It is widely used in daily life, medical care, agriculture, industry, and many other fields. The quality of non-woven fabric directly affects its final application effect; therefore, effective detection of surface defects in non-woven fabric is particularly important during the production process. Traditional methods for detecting surface defects in non-woven fabric mainly rely on manual inspection. Workers visually inspect and manually screen out defective products on the production line. However, manual inspection not only has disadvantages such as low accuracy and slow detection speed, but also easily leads to missed detections, causing unqualified products to enter the market, thereby affecting product quality and brand reputation. To improve detection efficiency and accuracy, some factories have adopted automated inspection systems based on industrial vision and AI vision technologies. These systems typically acquire images of nonwoven fabric surfaces using CCD array cameras and process and analyze these images using machine learning algorithms. In practical applications, image processing steps include image acquisition and preprocessing, feature extraction, and training and optimization of machine learning models. The trained machine learning model can detect and classify defects in newly acquired images, thereby achieving intelligent identification of surface defects in nonwoven fabrics. During feature extraction, existing technologies primarily identify surface defects by extracting edge features from images. Currently, commonly used edge feature extraction methods are based on threshold algorithms, which identify edges in images by setting fixed thresholds. However, this method has poor flexibility and limitations when dealing with complex scenarios. To improve detection accuracy, some algorithms use dynamic threshold adjustment for edge feature extraction. However, in areas where fibers aggregate to form knots or blocky nodules on the nonwoven fabric surface, these algorithms often suffer from redundant calculations, leading to decreased detection efficiency. Summary of the Invention

[0003] The purpose of this invention is to propose a nonwoven fabric quality inspection method and system based on image recognition and AI analysis, so as to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0004] To achieve the above objectives, according to one aspect of the present invention, a nonwoven fabric quality inspection method based on image recognition and AI analysis is provided, the nonwoven fabric quality inspection method based on image recognition and AI analysis comprising the following steps:

[0005] S100: Obtain the target grayscale image of the nonwoven fabric surface, preprocess it to obtain a steady-state image, and perform edge detection on the steady-state image to obtain multiple grayscale sub-regions.

[0006] S200: Obtain grayscale primitives within all grayscale sub-regions, and construct multiple clustering domains based on the grayscale primitives;

[0007] S300, calculate the similarity fit of all clustering regions, and filter out abnormal clustering regions based on the similarity fit.

[0008] S400: Calculate the connectivity feature value between the abnormal cluster and the adjacent cluster based on the abnormal cluster, and determine whether there are fibrous nodules in the abnormal cluster based on the connectivity feature value.

[0009] Fiber nodules on the surface of nonwoven fabrics are usually caused by a combination of factors. For example, the uniformity of raw fiber or the presence of impurities can cause fibers to clump together during production, forming nodules. Improper control of fiber tension or excessive spinning speed during production can lead to uneven fiber distribution, resulting in localized fiber accumulation. Fluctuations in humidity and temperature can also affect the physical properties of fibers, causing fiber adhesion or accumulation and nodule formation. Existing methods include using CCD array cameras to acquire nonwoven fabric images for machine learning-based intelligent defect detection. This involves first acquiring and preprocessing images to extract relevant features, then training a machine learning model using labeled data, evaluating and optimizing its performance, and finally applying the trained model to detect and classify defects in new images. Feature extraction typically involves extracting edge features from the image. Current edge feature extraction algorithms are primarily based on thresholding, which lacks flexibility. While other algorithms dynamically adjust thresholds for feature extraction, when fiber aggregation creates nodules or blocky nodules on the nonwoven fabric surface, the features at the nodules are extremely similar to the texture features at the edges, leading to repetitive calculations and affecting detection efficiency.

[0010] Further, in S100, the target grayscale image of the nonwoven fabric surface is acquired, and a steady-state image is obtained after preprocessing. The specific method for edge detection of the steady-state image to obtain multiple sub-regions is as follows: the nonwoven fabric surface image is acquired using a linear CCD camera under uniform illumination conditions, the nonwoven fabric surface image is grayscale processed to obtain the target grayscale image of the nonwoven fabric surface, the target grayscale image is processed by filtering, denoising and contrast enhancement to obtain a steady-state image, and the steady-state image is edge detected by the Canny operator to obtain multiple grayscale sub-regions.

[0011] Furthermore, in S200, the gray-level gradient primitives within all gray-level sub-regions are obtained. The specific method for constructing multiple clustering domains based on these gray-level gradient primitives is as follows: Using i as the index of the gray-level sub-region, Area... iLet represent the i-th grayscale sub-region. Iterate through the grayscale values ​​of each point in all grayscale sub-regions. Within each grayscale sub-region, subtract the grayscale value of each grayscale point from the grayscale value of the next grayscale point to obtain multiple grayscale difference values. Replace each grayscale value in the original grayscale sub-region with the grayscale difference value to obtain multiple grayscale difference image. Iterate through the 8-neighborhood matrix of each point in each grayscale difference image (if the point has no 8-neighborhood, fill the blank position with 0). Iterate through the grayscale difference values ​​in all 8-neighborhoods and compare the grayscale difference value of the center grayscale difference value with the grayscale difference values ​​of other neighboring grayscale differences. If the grayscale difference value of the center grayscale difference value is greater than the grayscale difference values ​​of other grayscale differences, replace the other differences with 1; otherwise, replace them with 0. Replace the difference value of the center point of the 8-neighborhood matrix with the average value of the other differences. Use the updated 8-neighborhood matrix as the grayscale primitive of each point in the grayscale sub-region.

[0012] The grayscale primitives described above are calculated based on the differences between adjacent grayscale values ​​in grayscale sub-regions. Due to the uniformity of the nonwoven fabric arrangement during the spinning process, the spinning will result in a regularly arranged grating structure. At this time, when using a CCD camera to collect optical information, light diffraction is likely to occur, causing color interference in some images and resulting in slight deviations between some grayscale values ​​and the actual values. The grayscale primitives of the above method directly reflect the degree of difference between adjacent grayscale values ​​and are not affected by grayscale errors after diffraction. They can more accurately depict the local differences and their spatial distribution patterns in grayscale images. The above methods improve the sensitivity to grayscale changes and enhance the ability to extract local features.

[0013] Iterate through the grayscale primitives of all points in all grayscale sub-regions. Construct a block matrix from the grayscale primitives contained in the same grayscale sub-region. Perform an XOR operation on corresponding points in the grayscale primitives of adjacent grayscale sub-regions. Mark points with a result of 1 as common points (SE) and points with a result of 0 as different points (DE). Calculate the regression equation Y1 for all common points (SE) and the regression equation Y2 for all different points (DE) using the least squares method. If the slopes of Y1 and Y2 are both less than zero, or both are greater than zero, then calculate the Area using the aggregation criterion. i With Area i+1 The clustering value among the i-th points in the grayscale sub-region, where the clustering criterion is: Where COS(·) represents the cosine of the angle, θ(·) represents the acute angle between two lines, and P (i，j) -P (i，j+1) Area i With Area i+1 The difference between the center points of the grayscale primitive at point j in the grayscale sub-region; j represents the index of the grayscale primitive, P (i，j)This represents the gray value of the center point of the gray-level primitive at point j in the i-th gray-level sub-region; otherwise, the aggregation criterion is... Where SIN(·) represents the sine value of the angle, for Area i With Area i+1 The clustering values ​​corresponding to all points in the grayscale sub-region are summed. If the sum is greater than zero, then the area is... i With Area i+1 Gray-scale sub-regions are stitched together into a new gray-scale sub-region using an image stitching algorithm. The gray-scale primitives of each point in the sub-region are directly ANDed to form new gray-scale primitives (the dimensions of the gray-scale primitives remain unchanged). All the new gray-scale primitives are used as the block matrix of the new gray-scale sub-region generated by stitching. Then, the new block matrix is ​​used to calculate the clustering value with the block matrix corresponding to the next gray-scale sub-region until the clustering criterion is no longer met. By following the above method, all gray-scale sub-regions are clustered to obtain multiple clustering regions.

[0014] Existing methods for extracting and identifying defects in nonwoven fabrics typically involve first segmenting the fabric into multiple regions using a threshold, and then dynamically extracting and identifying feature details by adjusting the acquisition window size. However, this method is prone to local overcomputation at region edges, affecting computational efficiency. Furthermore, the fiber arrangement of nonwoven fabrics exhibits strong regularity, with significant differences in grayscale patterns at nodules compared to the surrounding areas, leading to substantial errors when adjusting the window at these locations. Agglomeration domains, by utilizing grayscale distribution patterns to merge multiple sub-regions into one (these sub-regions may have some differences in grayscale values), effectively avoid interference from optical diffraction. Simultaneously, pre-identifying regions with similar structures greatly reduces the frequency of overcomputation. Furthermore, by using aggregation criteria to merge grayscale sub-regions with essentially consistent grayscale distribution patterns into the same agglomeration domain, it effectively avoids grayscale deviations in some sub-regions caused by optical diffraction. Most existing nonwoven fabric grayscale feature extraction techniques are based on grayscale distribution features or grayscale thresholds for feature classification. When grayscale interference occurs... The optical diffraction phenomenon on the surface of nonwoven fibers, which causes grayscale distortion, is prone to significant errors and can easily interfere with traditional methods. Furthermore, when fiber nodules appear in the nonwoven fabric, optical diffraction around the nodules can make it difficult to capture the nodule region, leading to missed detections and misidentifications. Moreover, if the distribution of fiber nodules is regular, the segmentation algorithm may repeatedly run in this area, causing an infinite loop within the computer. The aggregation criterion described above extracts areas with consistent grayscale differences from the fiber nodule region, preventing the algorithm from running in an infinite loop at this point. The merged cluster region consists of parts of the image exhibiting the same optical diffraction (due to the consistency of grayscale differences, the light propagation path is highly consistent, resulting in highly consistent diffraction phenomena), which is more conducive to subsequent nodule identification. This improves the accuracy and consistency of region identification, accurately identifies and merges regions with similar structures, ensures the consistency and integrity of image segmentation, reduces the need for manual intervention, and significantly improves the efficiency and accuracy of image analysis.

[0015] Furthermore, in S300, the similarity fit of all clusters is calculated. The specific method for filtering out anomalous clusters based on the similarity fit is as follows: The focal points *n* in all clusters are sequentially marked, where each focal point is the geometric center of the grayscale sub-region of the clustered region. The number of focal points can be one or more, according to the formula... The similarity fit SIM of the clustering region is calculated, where exp(·) represents the exponential function, DIS(·) represents the distance between two points, Smax(·) represents the maximum area, r represents the index of the foci, N represents the number of foci, s represents the index of the edge points of the clustering region, and e sLet represent the s-th edge point, and M represent the number of clusters. All clusters with a similarity fit less than the average are marked as abnormal clusters.

[0016] An abnormal clustering region is an image region on the surface of a nonwoven fabric that exhibits a significant difference in gray-level distribution compared to its surrounding areas due to the presence of fiber nodules. The similarity fitting degree of the above method is calculated based on the extreme feature points and edge gray-level points in each pair of clustering regions. If a clustering region contains fiber nodules, the similarity fitting degree between the corresponding clustering region and other clustering regions will be significantly lower than other values ​​(because the gray-level distribution pattern at the fiber nodules is significantly different from other parts, leading to a significant difference in the formula's Smax(Area). r (The area is significantly larger than other area values ​​due to the difference in the distribution of edge points). The abnormal clustering region is the region where the gray-scale distribution pattern is significantly different from other images. These areas may contain fibrous nodules, thereby accelerating the subsequent nodule identification and recognition work.

[0017] Furthermore, in S400, the connectivity characteristic value between the anomalous cluster and its adjacent clusters is calculated based on the anomalous cluster. The specific method for determining whether fibrous nodules exist in the anomalous cluster based on the connectivity characteristic value is as follows: using k as the index of the anomalous cluster, POL... k Let represent the k-th anomalous cluster. The nearest clusters surrounding the anomalous cluster are sequentially obtained. Specifically, all clusters in the 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315° directions are obtained. The clusters whose geometric centers in each of these eight directions are closest to the geometric center of the anomalous cluster are selected as its neighboring clusters, and these neighboring clusters are denoted as NPOL. t Let t be the index of the adjacent clusters, and let the Euclidean distance between the geometric center of the adjacent cluster and the geometric center of the anomalous cluster be the reference distance D1. k Let the limiting distance D2 be the distance with the largest difference from the reference distance among all Euclidean distances between the geometric centers of adjacent clique regions and their edge points. k Through formula Calculate the connectivity eigenvalue Fe between the anomalous cluster and its neighboring clusters. k Where Max(·) represents calculating the maximum value, T represents the number of adjacent clusters, K represents the number of anomalous clusters, and D1 (k，t) D2 represents the baseline distance between the k-th anomalous cluster and its t-th neighboring cluster. (k，t) This represents the limiting distance between the k-th anomalous cluster and its t-th neighboring cluster;

[0018] Iterate through the connectivity features Fe between adjacent anomalous clusters. If the connectivity features satisfy the connectivity criterion, then fibrous nodules are considered to exist in the current anomalous cluster; otherwise, fibrous nodules are considered not to exist in the current anomalous cluster. The connectivity criterion is... Where lg(·) represents the logarithmic function with base 10, Ave(·) represents the average gray value, MAx(ave_k-ave_t) represents the maximum gray value difference between the abnormal cluster and all adjacent clusters, Min(ave_k-ave_t) represents the minimum gray value difference between the abnormal cluster and all adjacent clusters, ave_k represents the average gray value in the k-th abnormal cluster, and ave_t represents the average gray value in the t-th adjacent cluster.

[0019] The beneficial effects of the above steps are as follows: The above method evaluates the connectivity between regions by calculating the geometric center and edge distance between abnormal agglomerates and their adjacent agglomerates, and is used to detect whether there are fiber nodules on the surface of nonwoven fabrics. By analyzing the geometric center and grayscale differences of abnormal agglomerates and their surrounding adjacent agglomerates, the structural features of local areas on the surface can be effectively captured to obtain the reference distance and limit distance between adjacent agglomerates and abnormal agglomerates. The connectivity feature value is calculated through these distances. When the connectivity feature value meets the set criteria, it indicates that there is a certain connectivity between the abnormal agglomerate and its adjacent agglomerates, thereby determining the presence of fiber nodules. Through this method, the distribution and interconnection of fibers on the surface of nonwoven fabrics can be accurately analyzed, possible fiber nodule areas can be found, and irregular fiber aggregation areas can be effectively identified, ensuring that problems are detected in a timely manner during the production process, thereby improving production efficiency and product quality.

[0020] Beneficial effects: This invention can effectively identify and detect fiber nodules that may exist on the surface of nonwoven fabrics. Through comprehensive analysis of the geometric relationship and grayscale difference between abnormal agglomeration domains and their adjacent agglomeration domains, it can accurately assess the distribution characteristics and abnormal aggregation of fibers on the surface. This not only improves the detection sensitivity of fiber nodules on the surface of nonwoven fabrics, but also optimizes the detection process through connectivity analysis, ensuring efficient operation under different scales and complex structures. For quality control in the production process, it can quickly and accurately locate areas that may have defects, reduce manual intervention, and improve the automation and accuracy of detection. Thus, while ensuring product quality, it also improves production efficiency and avoids potential quality problems.

[0021] This invention also provides a nonwoven fabric quality inspection system based on image recognition and AI analysis. The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program within the following system units:

[0022] The image acquisition unit acquires the target grayscale image of the nonwoven fabric surface, preprocesses it to obtain a steady-state image, and performs edge detection on the steady-state image to obtain multiple grayscale sub-regions.

[0023] The region construction unit obtains grayscale primitives within all grayscale sub-regions and constructs multiple clustering domains based on the grayscale primitives;

[0024] The image filtering unit calculates the similarity fit of all clustering regions and filters out abnormal clustering regions based on the similarity fit.

[0025] Nodule determination unit: Calculates the connectivity feature value between the abnormal cluster and the adjacent cluster based on the abnormal cluster, and determines whether there are fibrous nodules in the abnormal cluster based on the connectivity feature value.

[0026] The beneficial effects of this invention are as follows: This invention can effectively identify and detect fiber nodules that may exist on the surface of nonwoven fabrics. Through comprehensive analysis of the geometric relationship and grayscale difference between abnormal agglomeration domains and their adjacent agglomeration domains, it can accurately assess the distribution characteristics and abnormal aggregation of fibers on the surface. This not only improves the detection sensitivity of fiber nodules on the surface of nonwoven fabrics, but also optimizes the detection process through connectivity analysis, ensuring efficient operation under different scales and complex structures. For quality control in the production process, it can quickly and accurately locate areas that may have defects, reduce manual intervention, and improve the automation and accuracy of detection. Thus, while ensuring product quality, it improves production efficiency and avoids potential quality problems. Attached Figure Description

[0027] Figure 1 The flowchart shown is a nonwoven fabric quality inspection method based on image recognition and AI analysis;

[0028] Figure 2 The diagram shows the structure of a nonwoven fabric quality inspection system based on image recognition and AI analysis. Detailed Implementation

[0029] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0030] Example 1:

[0031] Figure 1 The flowchart shown is a nonwoven fabric quality inspection method based on image recognition and AI analysis.

[0032] Reference Figure 1This invention proposes a nonwoven fabric quality inspection method based on image recognition and AI analysis, the method comprising the following steps:

[0033] S100: Obtain the target grayscale image of the nonwoven fabric surface, preprocess it to obtain a steady-state image, and perform edge detection on the steady-state image to obtain multiple grayscale sub-regions.

[0034] S200: Obtain grayscale primitives within all grayscale sub-regions, and construct multiple clustering domains based on the grayscale primitives;

[0035] S300, calculate the similarity fit of all clustering regions, and filter out abnormal clustering regions based on the similarity fit.

[0036] S400: Calculate the connectivity feature value between the abnormal cluster and the adjacent cluster based on the abnormal cluster, and determine whether there are fibrous nodules in the abnormal cluster based on the connectivity feature value.

[0037] Further, in S100, the target grayscale image of the nonwoven fabric surface is acquired, and a steady-state image is obtained after preprocessing. The specific method for edge detection of the steady-state image to obtain multiple sub-regions is as follows: the nonwoven fabric surface image is acquired using a linear CCD camera under uniform illumination conditions, the nonwoven fabric surface image is grayscale processed to obtain the target grayscale image of the nonwoven fabric surface, the target grayscale image is processed by filtering, denoising and contrast enhancement to obtain a steady-state image, and the steady-state image is edge detected by the Canny operator to obtain multiple grayscale sub-regions.

[0038] Furthermore, in S200, the gray-level gradient primitives within all gray-level sub-regions are obtained. The specific method for constructing multiple clustering domains based on these gray-level gradient primitives is as follows: Using i as the index of the gray-level sub-region, Area... i Let represent the i-th grayscale sub-region. Iterate through the grayscale values ​​of each point in all grayscale sub-regions. Within each grayscale sub-region, subtract the grayscale value of each grayscale point from the grayscale value of the next grayscale point to obtain multiple grayscale difference values. Replace each grayscale value in the original grayscale sub-region with the grayscale difference value to obtain multiple grayscale difference image. Iterate through the 8-neighborhood matrix of each point in each grayscale difference image (if the point has no 8-neighborhood, fill the blank position with 0). Iterate through the grayscale difference values ​​in all 8-neighborhoods and compare the grayscale difference value of the center grayscale difference value with the grayscale difference values ​​of other neighboring grayscale differences. If the grayscale difference value of the center grayscale difference value is greater than the grayscale difference values ​​of other grayscale differences, replace the other differences with 1; otherwise, replace them with 0. Replace the difference value of the center point of the 8-neighborhood matrix with the average value of the other differences. Use the updated 8-neighborhood matrix as the grayscale primitive of each point in the grayscale sub-region.

[0039] Iterate through the grayscale primitives of all points in all grayscale sub-regions. Construct a block matrix from the grayscale primitives contained in the same grayscale sub-region. Perform an XOR operation on corresponding points in the grayscale primitives of adjacent grayscale sub-regions. Mark points with a result of 1 as common points (SE) and points with a result of 0 as different points (DE). Calculate the regression equation Y1 for all common points (SE) and the regression equation Y2 for all different points (DE) using the least squares method. If the slopes of Y1 and Y2 are both less than zero, or both are greater than zero, then calculate the Area using the aggregation criterion. i With Area i+1 The clustering value among the i-th points in the grayscale sub-region, where the clustering criterion is: Where COS(·) represents the cosine of the angle, θ(·) represents the acute angle between two lines, and P (i，j) -P (i，j+1) Area i With Area i+1 The difference between the center points of the grayscale primitive at point j in the grayscale sub-region; j represents the index of the grayscale primitive, P (i，j) This represents the gray value of the center point of the gray-level primitive at point j in the i-th gray-level sub-region; otherwise, the aggregation criterion is... Where SIN(·) represents the sine value of the angle, for Area i With Area i+1 The clustering values ​​corresponding to all points in the grayscale sub-region are summed. If the sum is greater than zero, then the area is... i With Area i+1 Gray-scale sub-regions are stitched together into a new gray-scale sub-region using an image stitching algorithm. The gray-scale primitives of each point in the sub-region are directly ANDed to form new gray-scale primitives (the dimensions of the gray-scale primitives remain unchanged). All the new gray-scale primitives are used as the block matrix of the new gray-scale sub-region generated by stitching. Then, the new block matrix is ​​used to calculate the clustering value with the block matrix corresponding to the next gray-scale sub-region until the clustering criterion is no longer met. By following the above method, all gray-scale sub-regions are clustered to obtain multiple clustering regions.

[0040] Furthermore, in S300, the similarity fit of all clusters is calculated. The specific method for filtering out anomalous clusters based on the similarity fit is as follows: The focal points *n* in all clusters are sequentially marked, where each focal point is the geometric center of the grayscale sub-region of the clustered region. The number of focal points can be one or more, according to the formula... The similarity fit SIM of the clustering region is calculated, where exp(·) represents the exponential function, DIS(·) represents the distance between two points, Smax(·) represents the maximum area, r represents the index of the foci, N represents the number of foci, s represents the index of the edge points of the clustering region, and e s Let represent the s-th edge point, and M represent the number of clusters. All clusters with a similarity fit less than the average are marked as abnormal clusters.

[0041] Furthermore, in S400, the connectivity characteristic value between the anomalous cluster and its adjacent clusters is calculated based on the anomalous cluster. The specific method for determining whether fibrous nodules exist in the anomalous cluster based on the connectivity characteristic value is as follows: using k as the index of the anomalous cluster, POL... k Let represent the k-th anomalous cluster. The nearest clusters surrounding the anomalous cluster are sequentially obtained. Specifically, all clusters in the 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315° directions are obtained. The clusters whose geometric centers in each of these eight directions are closest to the geometric center of the anomalous cluster are selected as its neighboring clusters, and these neighboring clusters are denoted as NPOL. t Let t be the index of the adjacent clusters, and let the Euclidean distance between the geometric center of the adjacent cluster and the geometric center of the anomalous cluster be the reference distance D1. k Let the limiting distance D2 be the distance with the largest difference from the reference distance among all Euclidean distances between the geometric centers of adjacent clique regions and their edge points. k Through formula Calculate the connectivity eigenvalue Fe between anomalous clusters and their neighboring clusters, where Max(·) represents the maximum value, T represents the number of neighboring clusters, K represents the number of anomalous clusters, and D1 represents the maximum value. (k，t) D2 represents the baseline distance between the k-th anomalous cluster and its t-th neighboring cluster. (k，t) This represents the limiting distance between the k-th anomalous cluster and its t-th neighboring cluster;

[0042] Traversing the connectivity features Fe between adjacent clusters of anomalous clusters k If the connectivity eigenvalues ​​satisfy the connectivity criterion, then fibrous nodules are considered to exist in the current anomalous cluster; otherwise, fibrous nodules are considered not to exist in the current anomalous cluster. The connectivity criterion is... Where lg(·) represents the logarithmic function with base 10, Ave(·) represents the average gray value, MAx(ave_k-ave_t) represents the maximum gray value difference between the abnormal cluster and all adjacent clusters, Min(ave_k-ave_t) represents the minimum gray value difference between the abnormal cluster and all adjacent clusters, ave_k represents the average gray value in the k-th abnormal cluster, and ave_t represents the average gray value in the t-th adjacent cluster.

[0043] Furthermore, this invention also provides embodiments of a nonwoven fabric quality inspection system based on image recognition and AI analysis, such as... Figure 2 The diagram shows the structure of the nonwoven fabric quality inspection system based on image recognition and AI analysis according to the present invention. The nonwoven fabric quality inspection system based on image recognition and AI analysis in this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above embodiment of the nonwoven fabric quality inspection system based on image recognition and AI analysis.

[0044] The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program in units of the following system:

[0045] The image acquisition unit acquires the target grayscale image of the nonwoven fabric surface, preprocesses it to obtain a steady-state image, and performs edge detection on the steady-state image to obtain multiple grayscale sub-regions.

[0046] The region construction unit obtains grayscale primitives within all grayscale sub-regions and constructs multiple clustering domains based on the grayscale primitives;

[0047] The image filtering unit calculates the similarity fit of all clustering regions and filters out abnormal clustering regions based on the similarity fit.

[0048] Nodule determination unit: Calculates the connectivity feature value between the abnormal cluster and the adjacent cluster based on the abnormal cluster, and determines whether there are fibrous nodules in the abnormal cluster based on the connectivity feature value.

[0049] The image recognition and AI analysis-based nonwoven fabric quality inspection system can run on computing devices such as desktop minicomputers, laptops, handheld computers, and cloud servers. The system that can run on the image recognition and AI analysis-based nonwoven fabric quality inspection system may include, but is not limited to, processors and memory. Those skilled in the art will understand that the examples described are merely illustrations of an image recognition and AI analysis-based nonwoven fabric quality inspection system and do not constitute a limitation on such systems. The system may include more or fewer components, combinations of certain components, or different components. For example, the image recognition and AI analysis-based nonwoven fabric quality inspection system may also include input / output devices, network access devices, buses, etc.

[0050] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the nonwoven fabric quality inspection system based on image recognition and AI analysis, connecting various parts of the system through various interfaces and lines.

[0051] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the nonwoven fabric quality inspection system based on image recognition and AI analysis by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0052] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A nonwoven fabric quality inspection method based on image recognition and AI analysis, characterized in that, The method includes the following steps: S100: Obtain the target grayscale image of the nonwoven fabric surface, preprocess it to obtain a steady-state image, and perform edge detection on the steady-state image to obtain multiple grayscale sub-regions. S200: Obtain grayscale primitives within all grayscale sub-regions, and construct multiple clustering domains based on the grayscale primitives; S300, calculate the similarity fit of all clustering regions, and filter out abnormal clustering regions based on the similarity fit. S400, calculate the connectivity feature value between the abnormal cluster and the adjacent cluster based on the abnormal cluster, and determine whether there are fibrous nodules in the abnormal cluster based on the connectivity feature value; In step S200, grayscale primitives within all grayscale sub-regions are obtained. The method for constructing multiple clustering regions based on these grayscale primitives is as follows: Using i as the index of the grayscale sub-region, Area... i Let represent the i-th gray-level sub-region. Iterate through the gray values ​​of each point in all gray-level sub-regions. Within each gray-level sub-region, subtract the gray value of each gray point from the gray value of the next gray point to obtain multiple gray-level difference values. Replace each gray value in the original gray-level sub-region with the gray-level difference values ​​to obtain multiple gray-level difference images. Iterate through the 8-neighborhood matrix of each point in each gray-level difference image. Iterate through the gray-level difference values ​​in all 8-neighborhoods and compare the gray-level difference value of the center with the gray-level difference values ​​of other neighboring gray-levels. If the gray-level difference value of the center is greater than the gray-level difference values ​​of other gray-levels, replace the other differences with 1; otherwise, replace them with 0. Replace the difference value of the center point of the 8-neighborhood matrix with the average value of the other differences. Use the updated 8-neighborhood matrix as the gray-level primitive of each point in the gray-level sub-region. Iterate through the grayscale primitives of all points in all grayscale sub-regions. Construct a block matrix from the grayscale primitives contained in the same grayscale sub-region. Perform an XOR operation on corresponding points in the grayscale primitives of adjacent grayscale sub-regions. Mark points with a result of 1 as common points (SE) and points with a result of 0 as different points (DE). Calculate the regression equation Y1 for all common points (SE) and the regression equation Y2 for all different points (DE) using the least squares method. If the slopes of Y1 and Y2 are both less than zero, or both are greater than zero, then calculate the Area using the aggregation criterion. i With Area i+1 The clustering value between the i-th points in the grayscale sub-region, for Area i With Area i+1 The clustering values ​​corresponding to all points in the grayscale sub-region are summed. If the sum is greater than zero, then the area is... i With Area i+1 Gray-scale sub-regions are stitched together into a new gray-scale sub-region using an image stitching algorithm. The gray-scale primitives of each point in the sub-region are directly ANDed to form new gray-scale primitives. All the new gray-scale primitives are used as the block matrix of the new gray-scale sub-region generated by stitching. Then, the new block matrix is ​​used to calculate the clustering value with the block matrix corresponding to the next gray-scale sub-region until the clustering criterion is no longer met. By following the above method, all gray-scale sub-regions are clustered to obtain multiple clustering regions.

2. The nonwoven fabric quality inspection method based on image recognition and AI analysis according to claim 1, characterized in that, In S100, the target grayscale image of the nonwoven fabric surface is acquired, and a steady-state image is obtained after preprocessing. The method for edge detection of the steady-state image to obtain multiple grayscale sub-regions is as follows: the nonwoven fabric surface image is acquired using a linear CCD camera under uniform illumination conditions, the nonwoven fabric surface image is grayscale processed to obtain the target grayscale image of the nonwoven fabric surface, the target grayscale image is processed by filtering, denoising and contrast enhancement to obtain a steady-state image, and the steady-state image is edge detected by the Canny operator to obtain multiple grayscale sub-regions.

3. The nonwoven fabric quality inspection method based on image recognition and AI analysis according to claim 2, characterized in that, In S300, the similarity fit of all clustered regions is calculated. The method for filtering out anomalous clustered regions based on the similarity fit is as follows: The focal points *n* in all clustered regions are sequentially marked, where each focal point is the geometric center of the grayscale sub-region of the clustered region generated by stitching together the data. This is done according to the formula... The similarity fit SIM of the clustering region is calculated, where exp(·) represents the exponential function, DIS(·) represents the distance between two points, Smax(·) represents the maximum area, r represents the index of the foci, N represents the number of foci, s represents the index of the edge points of the clustering region, and e s Let represent the s-th edge point, and M represent the number of clusters. All clusters with a similarity fit less than the average are marked as abnormal clusters.

4. The nonwoven fabric quality inspection method based on image recognition and AI analysis according to claim 3, characterized in that, In S400, the connectivity characteristic value between the anomalous cluster and its adjacent clusters is calculated based on the anomalous cluster. The method for determining whether there are fibrous nodules in the anomalous cluster based on the connectivity characteristic value is as follows: using k as the index of the anomalous cluster, POL... k Let represent the k-th anomalous cluster. We sequentially acquire all clusters in the directions of 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. We then select the clusters in each of these eight directions whose geometric centers are closest to the geometric centers of the anomalous clusters, and denote these neighboring clusters as NPOL. t Let t be the index of the adjacent clusters, and let the Euclidean distance between the geometric center of the adjacent cluster and the geometric center of the anomalous cluster be the reference distance D1. k The limiting distance D2 is defined as the distance with the largest difference from the reference distance among all Euclidean distances between the edge points of adjacent clusters and the geometric center of anomalous clusters. k Through formula Calculate the connectivity eigenvalue Fe between the anomalous cluster and its neighboring clusters. k , where Max(·) represents the maximum value to be calculated, T represents the number of adjacent clusters, K represents the number of anomalous clusters, D1(k,t) represents the baseline distance between the k-th anomalous cluster and its t-th adjacent cluster, and D2(k,t) represents the limit distance between the k-th anomalous cluster and its t-th adjacent cluster. Traverse the connectivity feature values ​​Fe between adjacent anomalous clusters. If the connectivity feature values ​​satisfy the connectivity criterion, it is considered that there are fibrous nodules in the current anomalous cluster; otherwise, it is considered that there are no fibrous nodules in the current anomalous cluster.

5. A nonwoven fabric quality inspection system based on image recognition and AI analysis, characterized in that, The nonwoven fabric quality inspection system based on image recognition and AI analysis includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the nonwoven fabric quality inspection method based on image recognition and AI analysis as described in any one of claims 1-4.

Images