Online detection method for cloth cover uniformity of non-woven cloth

Abstract

The invention relates to an online detection method for the cloth cover uniformity of non-woven cloth. The online detection method comprises the following steps: (1) carrying out online image acquisition and pretreatment; 2) dividing the cloth cover into rectangular detection units, and calculating a gray feature value in each detection unit; (3) the divided units are numbered in the X-axis direction and the Y-axis direction respectively; sequences are established along the Y-axis direction and the X-axis direction for the gray feature values of the X-axis detection unit and the Y-axis detection unit respectively; 4) analyzing an abnormal cluster of the gray feature value of the sequence by adopting a K-means clustering algorithm, marking a unit where an abnormal value is located, and calculating a uniformity index according to the total number of abnormal units; all results after sequence processing are marked on a cloth cover, and analysis results are visualized; 5, multi-stage early warning is triggered according to the detection result, and online detection of the uniformity of the non-woven fabric is achieved.According to the non-woven fabric uniformity detection method, instant local detection of the non-woven fabric is changed into long-time overall detection, and the production quality is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a cloth surface defect detection method, in particular to a non-woven cloth surface uniformity online detection method, and belongs to the technical field of machine vision. BACKGROUND

[0002] In modern industrial production, non-woven cloth as a new type of material is widely used in textile, medical, environmental protection and other fields. However, in the production process, due to uneven spraying of raw materials, environmental contamination, electrostatic adsorption and other reasons, non-woven cloth is prone to uneven appearance defects, which affects its use effect and product quality. The traditional non-woven cloth surface defect detection mainly relies on manual visual detection, which is low in efficiency and easy to be affected by the fatigue, emotion and misjudgment of the detector, resulting in unstable detection results and difficult to guarantee the quality, at the same time, long-time cloth inspection work will also cause harm to the workers.

[0003] In order to solve this problem, some non-woven cloth surface defect detection systems based on machine vision have been put into use. For example, the SIMV non-woven cloth online detection system integrates advanced machine image vision acquisition, photoelectric identification technology, and computer vision image software and hardware specially designed for domestic non-woven cloth products, and realizes accurate and rapid online non-woven cloth surface defect high-speed detection. However, these technologies still have limitations, and can only capture instantaneous local defects based on single frame image, and cannot effectively identify long-term and overall uneven problems which need to be judged by multiple frames of images.

[0004] Therefore, in order to solve the above problems, it is necessary to provide an innovative non-woven cloth surface uniformity online detection method to overcome the defects in the prior art. SUMMARY

[0005] The purpose of the present application is to provide a non-woven cloth surface uniformity online detection method which can accurately detect the overall uniformity of non-woven cloth in a period of time and improve the production quality of non-woven cloth.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: a non-woven cloth surface uniformity online detection method, which comprises the following process steps: 1), online image acquisition and pretreatment; 2), defining X axis in the width direction of cloth surface and Y axis in the running direction, dividing rectangular detection units Cell(i,j) along X axis and Y axis direction according to fixed width L, wherein i=1,2,...,M, j=1,2,...,N, calculating the gray value of each detection unit; ​3), the cells divided in step 2) are numbered along the X-axis direction and the Y-axis direction respectively; for the gray scale feature value of each X-axis detection cell, a sequence is established along the Y-axis direction; for the gray scale feature value of each Y-axis detection cell, a sequence is established along the X-axis direction; 4), a feature vector is constructed for the detection cell Cell , ) in step 3), the abnormal clusters of the gray scale feature values are analyzed by using a K-means clustering algorithm, and the cells where the abnormal values are located are marked, and the uniformity index is calculated according to the total number of abnormal cells; and the results after processing all sequences are marked on the fabric surface, and the analysis results are visualized. 5), according to the detection results, a warning is triggered to prompt the staff to adjust in time, so that the non-woven fabric uniformity online detection is realized.

[0007] The non-woven fabric surface uniformity online detection method further comprises that step 1) specifically comprises: 1-1), real-time capturing images of the non-woven fabric moving on the conveying belt by means of an industrial camera; 1-2), converting the collected color images into gray scale images; 1-3), using median filtering to eliminate noise in the images, and then highlighting the filament bundle profile through edge detection technology to form a pre-processed gray scale profile image.

[0008] The non-woven fabric surface uniformity online detection method further comprises that the gray scale feature value in step 2) comprises a mean value , a standard deviation , and a maximum value Max i,j , and the specific calculation method is as follows: , ,

[0009] In the formula, i and j are the numbers of the detection cells on the X-axis and the Y-axis respectively, and the gray scale mean value of the jth cell; is the total number of pixels in a single detection cell, which is determined by the camera resolution; is the gray scale value of the kth pixel in the ith and jth detection cell.

[0010] The non-woven fabric surface uniformity online detection method further comprises that the sequence construction method of step 3) specifically comprises: for each X-axis position X i , the feature values of the corresponding detection cells are collected in time sequence to form a three-dimensional time sequence:

[0011] wherein, represents the gray scale feature variation sequence of the i-th detection unit along the Y axis from the starting end to the current real-time detection position range in the X axis, i represents the number of the detection unit in the X axis, ; is the N-th feature vector in the detection unit numbered i along the Y axis direction from the starting position to the detection position ; , respectively represent the gray scale mean value, standard deviation, maximum value; For each Y axis position , the feature values of the corresponding detection unit are collected in position order to form a three-dimensional position sequence: wherein represents the gray scale feature variation sequence of the j-th detection unit along the X axis from one end of the cloth surface to the other end in the Y axis, j represents the number of the detection unit in the Y axis, ; is the M-th feature vector in the detection unit numbered j along the X axis direction from one end of the cloth surface to the other end .

[0012] The non-woven fabric surface uniformity online detection method of the application further comprises: in step 4), the feature vector constructed by the detection unit Cell( , ) is subjected to standardization processing, and the formula is as follows:

[0013] : the standardized feature value; : the original value of the k-th feature of the i, j-th unit; : the mean value of the k-th feature of all units; : the standard deviation of the k-th feature of all units.

[0014] The non-woven fabric surface uniformity online detection method of the application further comprises: in step 4), the method for K-means clustering algorithm to analyze the abnormal jump of the gray scale feature value is that: the preset cluster number K=2, the feature space is divided into a normal cluster and an abnormal cluster, which correspond to a standard cloth surface and a non-standard cloth surface respectively, wherein the cluster label of the normal cluster is 0, and the normal cluster center is C0; the cluster label of the abnormal cluster is 1, and the abnormal cluster center is C1; In the iteration process, the standardized gray scale feature vector sequence , the maximum iteration number , and the convergence threshold value are inputted: , and the cluster label 0 or 1 of each detection unit is outputted; the iteration process is specifically as follows: Assignment stage: for each feature vector, calculate the Euclidean distance with the current cluster center , , assign to the cluster with closer distance: its expression is as follows:

[0015] In the formula , is the normalized gray feature vector, is the normal cluster center, is the abnormal cluster center; Update stage: recalculate the mean of all feature vectors in each cluster as the new cluster center : its expression is as follows:

[0016] Wherein, is the normalized feature vector, is the total number of feature vectors in the kth cluster; is the set of all feature vectors contained in the kth cluster.

[0017] If the cluster center has no change or reaches the maximum number of iterations , terminate the iteration; otherwise, replace the old cluster center with the new cluster center , and return to the assignment stage to continue iteration.

[0018] The non-woven fabric surface uniformity online detection method of the application further comprises that: in step 4), the specific method for marking the unit where the abnormal value is located is: For the sequence T i with i as the X-axis number, if a certain feature vector is determined to be an abnormal cluster, the region where the abnormal cluster is located is marked with yellow, and the number of abnormal clusters is recorded as i ; if 2 or more consecutive detection units are used, the region is marked with dark yellow; For the sequence with j as the Y-axis number, if a certain feature vector is determined to be an abnormal cluster, the region where the abnormal cluster is located is marked with yellow, and the number of abnormal clusters is recorded as j ; if 3 or more consecutive detection units are used, the region is marked with dark yellow; For the same detection unit, if both the Y-axis time sequence and the X-axis position sequence are determined to be abnormal, the number of such units is recorded as , and the region is marked with red.

[0019] The non-woven fabric surface uniformity online detection method of the present application further has: in the step 4), the uniformity index is U, that is, the proportion of the number of normal units, and the calculation method is as follows:

[0020] Wherein is the total number of detection units of the fabric surface;

[0021] Wherein: i is the number of abnormal clusters of sequence T i ; j is the number of abnormal clusters of sequence P j ; is the number of the same detection unit being determined as an abnormal cluster in the bidirectional sequence; If U is greater than or equal to 0.95, it is marked as high-quality fabric surface, if 0.85 is less than or equal to U and U is less than 0.95, it is marked as good fabric surface, and if U is less than 0.85, it is marked as extremely uneven fabric surface.

[0022] The non-woven fabric surface uniformity online detection method of the present application can also be: when the total number of abnormal detection units determined by K-means clustering in step 4) exceeds 5% of the total number of detection units of the fabric surface or in the same sequence, 5 or more consecutive detection units are determined as abnormal clusters, the system automatically triggers a voice warning to prompt the worker; the warning feedback information needs to include the specific coordinate unit Cell , of the abnormal detection unit, so as to facilitate the worker to quickly locate the abnormal area of the fabric surface; at the same time, the system records the feature data of this abnormality in the background, that is, the three-dimensional feature value of the abnormal unit.

[0023] Compared with the prior art, the present application has the following beneficial effects: 1. The non-woven fabric surface uniformity online detection method of the present application precisely detects the overall uniformity of the fabric by fusing the spatiotemporal features of multiple images and dynamic model training, overcomes the limitations of traditional single-frame image detection, and can effectively identify the subtle unevenness accumulated over a long period of time.

[0024] 2. The non-woven fabric surface uniformity online detection method of the present application disassembles the fabric surface into standardized detection units and calculates multiple-dimensional features such as gray mean, standard deviation and maximum value, avoids the blurring of subtle unevenness defects, and comprehensively quantifies the uniformity state of the unit.

[0025] 3. The non-woven fabric surface uniformity online detection method of the present application triggers multi-level early warning based on detection results, realizes in-process control of production parameters, replaces manual detection to avoid fatigue misjudgment. Finally, it realizes accurate and efficient online detection of non-woven fabric uniformity, successfully upgrades the detection mode from "instantaneous local" to "long-term overall", and significantly improves production quality and control ability. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of the non-woven fabric surface uniformity online detection method of the present application.

[0027] Figure 2 is a flowchart of K-means clustering analysis of abnormal units in step 4) of the present application.

[0028] Figure 3 is the final generated uniformity visualization pattern of step 4) of the present application.

[0029] Figure 4 is a screenshot of actual detection of a certain non-woven fabric according to the present application. DETAILED DESCRIPTION

[0030] Please refer to the drawings in the specification Figure 1 The present application is a non-woven fabric surface uniformity online detection method, which comprises the following process steps: 1), online image acquisition and preprocessing, that is, real-time acquisition of dynamic fabric surface images on the non-woven fabric production line scene, and preprocessing through gray scale conversion and median filtering to generate preprocessed images meeting the requirements of gray scale feature extraction of detection units. This step specifically includes: 1-1), real-time capture of images of non-woven fabric moving on the conveyor belt with the help of an industrial camera.

[0031] 1-2), convert the captured color image into a gray scale image.

[0032] 1-3), use median filtering to eliminate noise in the image, and then highlight the filament bundle profile through edge detection technology to form a preprocessed gray scale profile image, providing a data basis for subsequent operations.

[0033] 2), define the width direction of the fabric surface as the X axis and the running direction as the Y axis, and divide the rectangular detection unit Cell(i,j) along the X and Y axes according to a fixed width L. , ), where i=1,2,...,M, j=1,2,...,N, M is the total number of units in the width direction, and N is the total number of units in the motion direction. Then calculate the gray scale feature values in each detection unit.

[0034] The gray scale feature values include mean , standard deviation , the maximum value Max i,j . The mean value may reflect the average density of the fibers of the detection unit; the standard deviation embodies the degree of dispersion of the pixel gray scale, and can reflect the density fluctuation of the fibers of the detection unit, and the specific calculation method is as follows: , ,

[0035] In the formula, i and j are the numbering of the detection unit on the X axis and the gray mean value of the jth unit on the Y axis; is the total number of pixels in a single detection unit, which is determined by the camera resolution; is the gray value of the kth pixel in the ith, jth detection unit.

[0036] Step 2) avoids blurring of subtle uneven defects by disassembling the cloth surface into standardized detection units and calculating gray mean value, standard deviation, maximum value and other multi-dimensional features, and comprehensively quantifies the uniformity of the unit.

[0037] 3), the units divided in step 2) are numbered along the X axis direction and the Y axis direction respectively; for each gray feature value of the X axis detection unit, a sequence is established along the Y axis direction; for each gray feature value of the Y axis detection unit, a sequence is established along the X axis direction, thereby forming a bidirectional feature sequence, breaking through the limitation of the prior art that can only detect locally and instantaneously, and realizing long-time overall monitoring in time and space dimensions.

[0038] Specifically, the sequence construction method is specifically: For each X axis position X i , the feature values of its corresponding detection unit are collected in time sequence to form a three-dimensional time sequence:

[0039] Among them, represents the gray feature change sequence of the ith detection unit on the X axis along the Y axis from the starting end to the current real-time detection position range; i represents the numbering of the detection unit on the X axis, ; is the Nth feature vector in the detection unit numbered i along the Y axis direction from the starting position to the detection place ; , are the gray mean value, standard deviation and maximum value respectively; For each Y axis position , the feature values of its corresponding detection unit are collected in position sequence to form a three-dimensional position sequence: wherein, represents the sequence of gray scale feature variation along X axis from one end of the cloth to the other end in the jth detection unit on Y axis; j represents the number of the detection unit on Y axis, is the Mth feature vector along X axis direction from one end of the cloth to the detection position in the detection unit numbered j

[0040] 4), the feature vector is constructed for the detection unit Cell( ), and the sequence in step 3) is analyzed by using the K-means clustering algorithm to analyze the abnormal cluster of the gray scale feature value, and the unit where the abnormal value is located is marked, and the uniformity index is calculated according to the total number of abnormal units. Then, the results of processing all sequences are marked on the cloth surface, and the analysis results are visualized.

[0041] Specifically, in order to eliminate the influence of different feature quantities on the clustering results, first, the feature vector constructed by the detection unit Cell( ) is standardized, and the formula is as follows:

[0042] wherein, : the standardized feature value; : the original value of the kth feature of the ith,jth unit; : the mean value of the kth feature of all units; : the standard deviation of the kth feature of all units.

[0043] The method for analyzing the abnormal jump of the gray scale feature value by using the K-means clustering algorithm is that: the cluster number K is preset as 2, the feature space is divided into a normal cluster and an abnormal cluster, which respectively correspond to a standard cloth surface and a non-standard cloth surface, wherein the cluster label of the normal cluster is 0, and the normal cluster center is C0; the cluster label of the abnormal cluster is 1, and the abnormal cluster center is C1.

[0044] The normal cluster center is the mean value of the gray scale feature vector of 100 qualified non-woven fabric samples ; the abnormal cluster center : the mean value of the gray scale feature vector of the historical abnormal sample, if there is no historical data, it is initialized as .

[0045] In the iteration process, the sequence of the standardized gray scale feature vector is input, the maximum iteration number is , and the convergence threshold is: , and the cluster label 0 or 1 of each detection unit is output; the iteration process is specifically as follows: ​​​​Assignment phase: for each feature vector, calculate the Euclidean distance to the current cluster center , , and assign to the cluster with closer distance: its expression is as follows:

[0046] where is the normalized gray feature vector, is the normal cluster center, is the abnormal cluster center.

[0047] Update phase: recalculate the mean of all feature vectors in each cluster as the new cluster center : its expression is as follows:

[0048] where, is the normalized feature vector, is the total number of feature vectors in the kth cluster; is the set of all feature vectors contained in the kth cluster.

[0049] If the cluster center has no change or reaches the maximum number of iterations , terminate the iteration; otherwise, replace the old cluster center with the new cluster center , and return to the assignment phase to continue iteration.

[0050] The result of iteration is: the feature vectors of normal clusters are concentrated around , and satisfy: where, is the normal fluctuation threshold, which can be set to an initial value according to historical data. The feature vectors of abnormal clusters are far away from , which can be manifested as a significant deviation of the mean, a too large standard deviation, and an abnormal maximum value.

[0051] Further, to adapt to the subtle changes of raw materials, environment and other factors in the production process, the system updates the normal cluster center every hour based on the latest normal samples (feature vectors of nearly 100 qualified detection units), and adjusts it using the exponential smoothing method, with the formula as follows:

[0052] is the normal cluster center at the tth hour; is the smoothing weight, fixed at 0.3; is the mean of the feature vectors of the latest 100 normal units at the tth hour; is the normal cluster center at the (t−1)th hour.

[0053] Continue to return to the distribution phase to continue iteration after the cluster center update is completed.

[0054] Further, the specific method for marking the unit where the abnormal value is located is: For the sequence T numbered i on the X axis i , if a certain feature vector is determined to be an abnormal cluster, the area where the abnormal cluster is located is marked with yellow, and the number of abnormal clusters is recorded as i ; if 2 or more consecutive detection units, the area is marked with dark yellow.

[0055] For the sequence P numbered j on the Y axis , if a certain feature vector is determined to be an abnormal cluster, the area where the abnormal cluster is located is marked with yellow, and the number of abnormal clusters is recorded as j ; if 3 or more consecutive detection units, the area is marked with dark yellow.

[0056] For the same detection unit in the Y axis time sequence and the X axis position sequence, if both are determined to be abnormal, the number of such units is recorded as , and the area is marked with red.

[0057] Further, the uniformity index U, i.e. the proportion of normal unit number, is calculated as follows:

[0058] wherein, is the total number of detection units of the cloth surface;

[0059] wherein: i is the number of abnormal clusters of the sequence T i ; j is the number of abnormal clusters of the sequence P j ; is the number of the same detection unit in the bidirectional sequence that is determined to be an abnormal cluster.

[0060] If U≥0.95, it is marked as high-quality cloth surface, if 0.85≤U<0.95, it is marked as good cloth surface, and if U<0.85, it is marked as extremely uneven cloth surface.

[0061] In summary, the step 4) adopts the K-means clustering algorithm with dynamic update of normal cluster center, the single sequence + bidirectional cross classification abnormal marking, and the quantitative uniformity rate calculation, and the cloth surface condition is visualized.

[0062] 5), according to the detection result, triggering early warning, prompting workers to adjust in time, realizing the online detection of non-woven fabric unevenness.

[0063] The specific steps are as follows: when the proportion of the total number of abnormal detection units determined by K-means clustering in step 4) to the total number of detection units on the cloth surface exceeds 5% of the preset threshold or in the same sequence, 5 or more consecutive detection units are determined to be abnormal clusters, the system automatically triggers voice warning to prompt workers; the early warning feedback information needs to include the specific coordinate unit Cell( , ) of the abnormal detection unit, so as to facilitate workers to quickly locate the abnormal area of the cloth surface; at the same time, the system records the feature data of this abnormality in the background, that is, the three-dimensional feature value of the abnormal unit. Workers can check the running parameters of the system combined with the feedback early warning information, and adjust the process parameters in time to correct the unevenness of the cloth surface, so as to realize real-time online quality control.

[0064] The present application triggers multi-level early warning based on the detection result, realizes the in-process control of production parameters, replaces manual detection to avoid fatigue misjudgment, and finally realizes the accurate and efficient online detection of non-woven fabric uniformity, successfully upgrades the detection mode from "instantaneous local" to "long-time overall", and significantly improves the production quality and control ability.

[0065] The above specific embodiments are only the preferred embodiments of the present application, and are not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An online detection method for the surface uniformity of nonwoven fabrics, characterized in that: The process includes the following steps: 1) Online image acquisition and preprocessing; 2) Define the fabric width direction as the X-axis and the running direction as the Y-axis. Divide the fabric into rectangular detection cells (Cells) along the X and Y axes according to a fixed width L. , ), where i=1,2,...,M, j=1,2,...,N, calculate the grayscale feature value within each detection unit; 3) Number the units divided in step 2) along the X-axis and Y-axis respectively; establish a sequence of gray-scale feature values ​​of each detection unit along the Y-axis; establish a sequence of gray-scale feature values ​​of each detection unit along the X-axis. 4) For the detection unit Cell ( , ) Construct feature vectors, and use the K-means clustering algorithm to analyze the outlier clusters of gray-level feature values ​​of all sequences described in step 3), mark the units where the outliers are located, and calculate the uniformity index based on the total number of outlier units; then mark the results of all processed sequences on the fabric surface to visualize the analysis results; 5) Trigger an early warning based on the test results to prompt staff to make timely adjustments, thereby achieving online detection of nonwoven fabric unevenness.

2. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 1, characterized in that: Step 1) specifically refers to: 1-1), using an industrial camera to capture real-time images of the nonwoven fabric moving on the conveyor belt; 1-2), convert the acquired color images into grayscale images; (1-3) Median filtering is used to eliminate noise in the image, and edge detection technology is used to highlight the outline of the filament bundle, forming a preprocessed grayscale outline image.

3. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 1, characterized in that: The grayscale feature values ​​in step 2) include the mean. Standard deviation Maximum value Max i,j The specific calculation method is as follows: ， ， In the formula, i and j are the gray average values ​​of the j-th unit, which is the number of the detection unit on the X-axis and Y-axis, respectively. It is the total number of pixels within a single detection unit, determined by the camera resolution; It is the grayscale value of the k-th pixel within the i-th and j-th detection units.

4. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 3, characterized in that: The sequence construction method in step 3) is as follows: For each X-axis position X i The feature values ​​of the corresponding detection units are collected in chronological order to form a three-dimensional time series: in, This represents the sequence of grayscale feature changes along the Y-axis from the starting point to the current real-time detection position for the i-th detection unit on the X-axis; i represents the number of the detection unit on the X-axis. ; Let i be the Nth feature vector along the Y-axis from the starting position to the detection point in the detection unit numbered i. ; , These are the mean, standard deviation, and maximum grayscale values, respectively. For each Y-axis position The feature values ​​of the corresponding detection units are collected in order of position to form a three-dimensional position sequence: Where represents the sequence of grayscale feature changes of the j-th detection unit on the Y-axis from one end of the fabric to the other along the X-axis; j represents the number of the detection unit on the Y-axis. ; Let j be the Mth feature vector along the X-axis from one end of the fabric to the detection point in the detection unit. .

5. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 1, characterized in that: In step 4), the detection unit Cell ( , The constructed feature vectors are standardized using the following formula: Standardized eigenvalues; The original value of the k-th feature of the i-th,j-th unit; The mean of the k-th feature of all units; : The standard deviation of the k-th feature of all units.

6. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 5, characterized in that: In step 4), the method of K-means clustering algorithm to analyze abnormal jumps in gray-level feature values ​​is as follows: the number of clusters is preset to K=2, and the feature space is divided into normal clusters and abnormal clusters, corresponding to standard and non-standard fabric surfaces, respectively. The cluster label of the normal cluster is 0, and the center of the normal cluster is C0; the cluster label of the abnormal cluster is 1, and the center of the abnormal cluster is C1. The standardized gray-scale feature vector sequence is input during the iteration process. Maximum number of iterations The convergence threshold is: Output the cluster label 0 or 1 for each detection unit; the iterative process is as follows: Allocation phase: For each feature vector, calculate the value relative to the current cluster center. , The Euclidean distance is used to assign individuals to clusters that are closer in distance: its expression is as follows: In the formula This is the standardized grayscale feature vector. As a normal cluster center, It is the center of an abnormal cluster; Update phase: Recalculate the mean of all eigenvectors within each cluster, and use this mean as the new cluster center. Its expression is as follows: in, The standardized feature vector, The total number of eigenvectors within the k-th cluster; The set of all feature vectors contained in the Kth cluster; If the cluster center remains unchanged or the maximum number of iterations has been reached... Terminate the iteration if necessary; otherwise, use the new cluster center. Replace the old cluster center Then return to the allocation phase to continue iterating.

7. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 6, characterized in that: In step 4), the specific method for marking the cell containing the outlier is as follows: For the sequence T with X-axis number i i If a certain feature vector is identified as an anomalous cluster, the region containing the anomalous cluster is marked in yellow, and the number of anomalous clusters is denoted as . i If two or more consecutive detection units are detected, the area will be marked in dark yellow. For the sequence numbered j on the Y-axis If a certain feature vector is identified as an anomalous cluster, the region containing the anomalous cluster is marked in yellow, and the number of anomalous clusters is denoted as . j If three or more consecutive detection units are detected, the area will be marked in dark yellow. If the same detection unit is judged as abnormal in both the Y-axis time series and the X-axis position series, the number of such units is denoted as . The area is marked in red.

8. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 7, characterized in that: In step 4), the uniformity index is U, which represents the percentage of normal units. Its calculation method is as follows: in This represents the total number of detection units on the fabric surface. in: i For sequence T i The number of anomalous clusters; j For sequence P j The number of anomalous clusters; It is the number of clusters in which the same detection unit is identified as an anomalous cluster in both bidirectional sequences; If U ≥ 0.95, it is marked as high-quality fabric; if 0.85 ≤ U < 0.95, it is marked as good fabric; if U < 0.85, it is marked as extremely uneven fabric.

9. The online detection method for the surface uniformity of nonwoven fabrics as described in claim 1, characterized in that: Step 5) specifically involves: when the total number of abnormal detection units determined by K-means clustering in step 4) exceeds a preset threshold of 5% or when 5 or more consecutive detection units in the same sequence are identified as abnormal clusters, the system automatically triggers a voice warning to alert the worker; the warning feedback information must include the specific coordinate unit (Cell) of the abnormal detection unit. , This allows staff to quickly locate abnormal areas on the fabric; at the same time, the system records the characteristic data of this abnormality in the background, namely the three-dimensional characteristic value of the abnormal unit.

Images