Control method and system based on flaw detection

Abstract

The invention discloses a control method and system based on flaw detection, and relates to the technical field of printing and dyeing detection. The method comprises the following steps: obtaining K weaving printing plates and K weaving printing patterns; performing pattern complexity analysis, and determining K first control coefficients; neighborhood association is carried out, and K second control coefficients are determined; when the K weaving printing plates complete printing on the weaving cloth, image acquisition is carried out, and K weaving cloth images are obtained; performing flaw detection to generate K first flaw detection results and K second flaw detection results; integrating to generate a target flaw detection result; and performing control scheme identification based on the target flaw detection result, and generating a target control scheme. The technical problem that the printing and dyeing quality is reduced due to the fact that the defect detection efficiency and accuracy are low in the textile printing and dyeing process in the prior art and the control scheme cannot be adjusted in time is solved, the control scheme is optimized through the defect detection result, and the technical effect of improving the printing and dyeing quality is achieved.

Description

Technical Field

[0001] This invention relates to the field of dyeing and printing inspection technology, specifically to a control method and system based on defect detection. Background Technology

[0002] In the textile manufacturing industry, dyeing and printing have always held a crucial position. They not only endow textiles with vibrant and colorful appearances but also significantly enhance product added value and market competitiveness. However, with consumers' ever-increasing demands for textile quality and the growing complexity and diversity of printing designs, defects in the dyeing and printing process have become increasingly prominent. Traditional defect detection methods, such as manual inspection and simple machine vision systems, are no longer sufficient to meet the modern textile manufacturing industry's demand for efficient and accurate defect detection. First, manual inspection is inefficient and easily affected by human factors such as fatigue and experience differences, leading to inconsistent and inaccurate results. In today's massive textile production, this method can no longer meet the needs of rapid and continuous defect detection on production lines. Second, while traditional machine vision systems can partially automate detection, their simple algorithms and limited functionality often only detect obvious defects, rendering them ineffective against subtle flaws hidden beneath complex printing designs. This not only limits the improvement of textile quality but also poses a significant challenge to product quality control for enterprises. Summary of the Invention

[0003] This application provides a control method and system based on defect detection, which solves the technical problem that the efficiency and accuracy of defect detection in the textile printing and dyeing process are low in the prior art, which leads to the inability to adjust the control scheme in time and thus reduces the printing and dyeing quality.

[0004] A first aspect of this application provides a control method based on defect detection, the method comprising: Based on the fabric information and printing requirements of the target fabric, the fabric printing and dyeing equipment is configured to obtain K fabric printing plates and K fabric printing patterns, wherein the K fabric printing plates have K printing numbers; pattern complexity analysis is performed on the K fabric printing patterns to determine K first control coefficients; neighborhood association is performed on the K fabric printing patterns based on the K printing numbers to determine K second control coefficients; based on the K first control coefficients, the K second control coefficients, and the K fabric printing patterns, K first detection branches and K second detection branches of a parallel fabric defect detection channel are constructed. The process involves several steps: First, when K fabric printing plates have finished printing on the fabric, K color CCD cameras positioned above the K printing plates are used to acquire images of the fabric, resulting in K fabric images. The K first detection branches and K second detection branches are then used to detect defects in the K fabric images, generating K first defect detection results and K second defect detection results. These results are then integrated to generate a target defect detection result. Finally, a control scheme is identified based on the target defect detection result to generate a target control scheme.

[0005] A second aspect of this application provides a defect detection-based control system, the system comprising: A configuration module is used to configure the weaving and printing device according to the fabric information and printing requirements of the target fabric, obtaining K weaving printing plates and K weaving printing patterns, wherein the K weaving printing plates have K printing numbers; a parsing module is used to perform pattern complexity analysis on the K weaving printing patterns respectively, and determine K first control coefficients; an association module is used to perform neighborhood association on the K weaving printing patterns based on the K printing numbers, and determine K second control coefficients; a detection channel construction module is used to construct K first detection branches and K second detection branches of a parallel fabric defect detection channel based on the K first control coefficients, the K second control coefficients and the K weaving printing patterns. The system comprises: a detection branch; an image acquisition module, which, when K fabric printing plates have been printed on the fabric, uses K color CCD cameras positioned above the K fabric printing plates to acquire images of the fabric, obtaining K fabric images; a defect detection module, which uses the K first detection branches and the K second detection branches to detect defects in the K fabric images, generating K first defect detection results and K second defect detection results; an integration module, which integrates the K first defect detection results and the K second defect detection results to generate a target defect detection result; and an identification module, which identifies a control scheme based on the target defect detection result and generates a target control scheme.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, the fabric printing and dyeing equipment is configured based on the fabric information and printing requirements of the target fabric, resulting in K fabric printing plates and K fabric printing patterns. Each of the K fabric printing plates has a printing number. Next, pattern complexity analysis is performed on each of the K fabric printing patterns to determine K first control coefficients. Based on the K printing numbers, neighborhood association is performed on the K fabric printing patterns to determine K second control coefficients. Then, based on the K first control coefficients, K second control coefficients, and K fabric printing patterns, K first detection branches and K second detection branches of a parallel fabric defect detection channel are constructed. When the K fabric printing plates have finished printing on the fabric, K color CCD cameras positioned above the K fabric printing plates are used to acquire images of the fabric, obtaining K fabric images. Then, the K first detection branches and K second detection branches are used to perform defect detection on the K fabric images, generating K first defect detection results and K second defect detection results. Furthermore, the K first defect detection results and K second defect detection results are integrated to generate target defect detection results. Finally, based on the target defect detection results, control scheme identification is performed to generate a target control scheme. This solves the technical problem in existing technologies where the efficiency and accuracy of defect detection in textile printing and dyeing processes are low, leading to a failure to adjust control schemes in a timely manner and resulting in reduced printing and dyeing quality. By optimizing the control scheme through defect detection results, the technical effect of improving printing and dyeing quality is achieved. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the control method based on defect detection provided in an embodiment of this application; Figure 2 This is a schematic diagram of the control system structure based on defect detection provided in an embodiment of this application.

[0009] Figure labeling: Configuration module 11, parsing module 12, association module 13, detection channel construction module 14, image acquisition module 15, defect detection module 16, integration module 17, recognition module 18. Detailed Implementation

[0010] This application provides a control method and system based on defect detection, which solves the technical problem in the prior art that the efficiency and accuracy of defect detection in the textile printing and dyeing process are low, which leads to the inability to adjust the control scheme in a timely manner and thus reduces the printing and dyeing quality.

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0012] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0013] Example 1, as Figure 1 As shown, this application provides a control method based on defect detection, wherein the method includes: Based on the fabric information and printing requirements of the target fabric, the weaving and dyeing equipment is configured to obtain K weaving printing plates and K weaving printing patterns, wherein the K weaving printing plates have K printing numbers.

[0014] In textile printing and dyeing processes, in order to achieve diverse printing designs and meet the needs of different fabrics, it is necessary to precisely configure the weaving and printing equipment based on the fabric information of the target fabric and the specific printing requirements. The fabric information includes the characteristics of the fabric such as material, density, thickness, and color, while the printing requirements cover the specific requirements such as the design, color, size, and position of the printing pattern. Based on the configuration of the fabric printing and dyeing equipment, K fabric printing plates and K fabric printing patterns can be obtained. Each fabric printing plate is precisely designed and manufactured according to specific printing requirements, with different shapes, sizes, and textures to adapt to the needs of different fabrics and printing patterns. For easy identification, each fabric printing plate is assigned a unique printing number. Corresponding to the K fabric printing plates are the K fabric printing patterns. These patterns are engraved or drawn on the printing plates according to the printing requirements. They will serve as templates in the printing process, guiding the printing press to form the final printing effect on the fabric. Each fabric printing pattern corresponds to a specific printing plate and has a matching printing number to ensure accurate identification and matching during the printing process.

[0015] The pattern complexity of each of the K fabric printing patterns is analyzed to determine the K first control coefficients.

[0016] By performing pattern complexity analysis on K fabric printing patterns, including identifying features such as color, lines, and texture in the patterns, and calculating the number and density of these features, K first control coefficients are obtained, which are used to represent the complexity of the patterns.

[0017] Furthermore, this includes: Extract K pattern size features and K pattern distribution uniformity features from the K woven fabric printing patterns; use a pattern complexity recognizer to identify the K pattern size features and K pattern distribution uniformity features to obtain the K first control coefficients.

[0018] Preferably, K pattern size features and K pattern distribution uniformity features are extracted from K printed fabric patterns. The pattern size features include the overall size of the pattern, the size distribution of each element (such as lines, color blocks, etc.), and the relative size ratio. The pattern distribution uniformity features refer to the distribution and arrangement of the pattern on the fabric, such as the pattern being repeatedly distributed on the fabric. A deep neural network or convolutional neural network is selected, and complex graphic data with labeled complexity levels is used as training data to train the model. After training, a pattern complexity recognizer is obtained. The pattern complexity recognizer can output the corresponding pattern complexity evaluation value based on the input feature vector. The K pattern size features and K pattern distribution uniformity features are input into the pattern complexity recognizer, and the pattern complexity recognizer will output K first control coefficients.

[0019] Based on the K printing numbers, neighborhood association is performed on the K fabric printing patterns to determine K second control coefficients.

[0020] Based on the K printing numbers, neighborhood association is performed on the K woven printing patterns, that is, it is determined which patterns are close to each other in spatial position. Then, the second control coefficient can be determined based on the distance between the two patterns, and K second control coefficients are obtained.

[0021] Furthermore, this includes: Based on the K printing numbers, any two adjacent fabric printing patterns are grouped into a pattern group to obtain K-1 pattern groups; the distance between the centers of two adjacent fabric printing patterns in each of the K-1 pattern groups is statistically analyzed to obtain the center distance of the K-1 pattern groups; the ratio of the center distance of each of the K-1 pattern groups to the sum of the center distances of the K-1 pattern groups is calculated, and the reciprocal of the ratio is used as the control coefficient of the K-1 pattern groups; the control coefficient of the K-1 pattern groups is retrieved using the K fabric printing patterns as an index to obtain the set of control coefficients of the K pattern groups of the K fabric printing patterns; the average value of the set of control coefficients of the K pattern groups is calculated by traversing the set of control coefficients of the K pattern groups to obtain the K second control coefficients.

[0022] Preferably, based on K printing numbers, all fabric printing patterns are traversed, and any two adjacent patterns are selected to form a pattern group, resulting in K-1 pattern groups. For each pattern group, the center points of two patterns are found, and the distance between these two center points is calculated. This distance is the center distance of the pattern group. This process is repeated to calculate the center distances of all K-1 pattern groups. For each pattern group, the ratio of its center distance to the sum of the center distances of all pattern groups is calculated, and the reciprocal of this ratio is used as the control coefficient of the pattern group. This process is repeated to calculate the control coefficients of all K-1 pattern groups. Since each pattern appears in multiple pattern groups (except for patterns on the boundary), each pattern will be associated with multiple pattern group control coefficients. Using the K fabric printing patterns as an index, all pattern group control coefficients associated with each pattern are retrieved and collected. For each pattern, all its associated pattern group control coefficients are placed into a set, forming a set of K pattern group control coefficients. The set of pattern group control coefficients for each pattern is traversed, and the average of the control coefficients in the set is calculated to obtain the second control coefficient of the pattern. This process is repeated to calculate the second control coefficients of all K patterns.

[0023] Based on the K first control coefficients, the K second control coefficients, and the K woven printing patterns, K first detection branches and K second detection branches are constructed for the parallel fabric defect detection channel.

[0024] By combining K first control coefficients, K second control coefficients, and K fabric printing patterns, K first detection branches and K second detection branches can be created respectively to improve detection efficiency.

[0025] Furthermore, this includes: Positioning points are configured based on the K first control coefficients and the K second control coefficients to generate K first positioning points and K second positioning points. Edge points are randomly extracted from the K woven fabric printing patterns based on the K first positioning points and the K second positioning points, resulting in K sets of first random positioning points and K sets of second random positioning points. The K sets of first random positioning points and the K sets of second random positioning points are sequentially connected to generate K first randomly extracted patterns and K second randomly extracted patterns. The K first randomly extracted patterns and the K second randomly extracted patterns are verified using the K woven fabric printing patterns. If the verification is successful, the K first detection branches and the K second detection branches of a parallel fabric defect detection channel are constructed based on the K sets of first random positioning points and the K sets of second random positioning points.

[0026] Preferably, positioning points are configured based on K first control coefficients and K second control coefficients, generating K first positioning points and K second positioning points. For each woven printing pattern, edge points are randomly selected as first random positioning points based on their corresponding number of first positioning points, forming K sets of first random positioning points. For each pattern group, edge points are randomly selected as second random positioning points within the pattern group based on their corresponding number of second positioning points, forming K sets of second random positioning points. The points in each set of first random positioning points are connected to generate K first randomly extracted patterns, and the points in each set of second random positioning points are connected to generate K second randomly extracted patterns. The first randomly extracted pattern and K second randomly extracted patterns are compared and verified with the original K fabric printing patterns to ensure that the randomly extracted patterns can represent the main features and structure of the original patterns without losing important information. If the verification is successful, the randomly extracted pattern can accurately reflect the features and structure of the original pattern. If the verification fails, the positioning point configuration or random extraction strategy needs to be readjusted, and the extraction and verification should be performed again. Based on the verified set of K first random positioning points and K second random positioning points, K first detection branches and K second detection branches of the parallel fabric defect detection channel are constructed. In each detection branch, the corresponding random positioning points are used for defect detection and analysis.

[0027] Furthermore, this includes: A Siamese neural network is obtained, wherein the Siamese neural network includes a first Siamese branch, a second Siamese branch, and a loss function; the K fabric printing patterns are input into the first Siamese branch, and the K first randomly extracted patterns are input into the second Siamese branch to obtain K first printed pattern feature sets and K first random pattern feature sets; the loss function is used to calculate K first random loss values ​​for the K first printed pattern feature sets and K first random pattern feature sets; the K fabric printing patterns and the K first randomly extracted patterns are input into the Siamese neural network to obtain K second random loss values; it is determined whether the K first random loss values ​​and the K second random loss values ​​are greater than or equal to a preset loss value threshold; if not, the verification is successful.

[0028] Preferably, a Siamese neural network is constructed, comprising a first Siamese branch, a second Siamese branch, and a loss function. K fabric printing patterns are input into the first Siamese branch, and K first randomly extracted patterns are input into the second Siamese branch. Features are extracted from the K input fabric printing patterns through the first Siamese branch to obtain a feature set of K first printing patterns. Features are extracted from the K input first randomly extracted patterns through the second Siamese branch to obtain a feature set of K first random patterns. The loss function in the Siamese neural network is used to calculate the feature sets of K first printing patterns and the K first random patterns. K first random loss values ​​are obtained between the feature sets of the case, which reflect the similarity between the original pattern and the randomly extracted pattern. K fabric printing patterns and K first randomly extracted patterns are input into the Siamese neural network to obtain K second random loss values. A preset loss value threshold is set, which represents the acceptable similarity level. It is determined whether the K first random loss values ​​and K second random loss values ​​are all less than or equal to the preset loss value threshold. If all loss values ​​meet the condition (i.e., less than or equal to the threshold), the verification is passed, indicating that the randomly extracted pattern is sufficiently similar to the original fabric printing pattern in terms of features.

[0029] Furthermore, the loss function is: Where MSE is the first random loss, n is the total number of features in the first printed pattern feature set, and n is a positive integer greater than or equal to 1. Let i be the i-th feature vector of the first printed pattern feature set. Let be the i-th feature vector of the first random pattern feature set.

[0030] The loss function in a Siamese neural network is Where MSE is the first random loss, representing the square of the average difference between the two feature sets, n is the total number of features in the first printed pattern feature set, and n is a positive integer greater than or equal to 1. Let i be the i-th feature vector of the first printed pattern feature set. Let be the i-th feature vector of the first random pattern feature set. Specifically, K fabric printing patterns are input into the first twin branch of the Siamese neural network to obtain K first printing pattern feature sets. K first randomly extracted patterns are input into the second twin branch of the Siamese neural network to obtain K first random pattern feature sets. For each feature set, the feature vectors are traversed, and the square of the difference between the corresponding feature vectors is calculated. The squares of the differences between all feature vectors are summed to obtain the MSE value (i.e., the first random loss) of the feature set. The above process is repeated to calculate the MSE values ​​of all K pairs of feature sets. It is then determined whether these K MSE values ​​(i.e., K first random losses) are all less than or equal to a preset loss threshold. If all conditions are met, the verification is successful; otherwise, the verification fails.

[0031] When the K fabric printing plates have finished printing on the fabric, the K color CCD cameras positioned above the K fabric printing plates are used to capture images of the fabric, thus obtaining K fabric images.

[0032] When the K printing plates have finished printing on the fabric, the K color CCD cameras positioned above the K printing plates are used to capture images of the fabric that has just been printed. Each color CCD camera will capture the fabric image of its corresponding printing plate area, thus obtaining K fabric images.

[0033] The K first detection branches and the K second detection branches are used to perform defect detection on the K fabric images respectively, generating K first defect detection results and K second defect detection results.

[0034] K fabric images are input into K first detection branches and K second detection branches respectively to generate K first defect detection results and K second defect detection results.

[0035] Furthermore, this includes: K sample fabric images are collected, and defect detection is performed on K first random positioning points in each of the K sample fabric images to obtain K sample first defect detection results. K sample fabric images are also collected, and defect detection is performed on K second random positioning points in each of the K sample fabric images to obtain K sample second defect detection results. Using the K sample fabric images, the K first random positioning points, and the K sample first defect detection results as construction data, K first detection branches are constructed. Using the K sample fabric images, the K second random positioning points, and the K sample second defect detection results as construction data, K second detection branches are constructed. The K first detection branches and the K second detection branches are then connected in parallel to generate the parallel fabric defect detection channel.

[0036] Preferably, a set of K sample fabric images is acquired. For each sample fabric image, a location point is selected from the corresponding set of K first random location points. Defect detection is performed on the selected first random location point and its surrounding area. For each sample fabric image, the defect detection results of the first random location point and its surrounding area are recorded and saved, forming K sample first defect detection results. For each sample fabric image, a location point is selected from the corresponding set of K second random location points. Defect detection is performed on the selected second random location point and its surrounding area. For each sample fabric image, the defect detection results of the second random location point and its surrounding area are recorded and saved, forming K sample second defect detection results. The set of K sample fabric images, the set of K first random location points, and the K sample first defect detection results are used as construction data. Using this data, a corresponding first detection branch is constructed for each sample fabric image. This involves training multiple machine learning models to identify and locate defects at specific random locations, resulting in K first detection branches. Then, using a set of K sample fabric images, K sets of K second random locations, and K sample second defect detection results as construction data, similar to the construction of the first detection branches, these data are used to construct the second detection branches. Each second detection branch will perform defect detection on its corresponding sample fabric image and random location. The K first and K second detection branches are then connected in parallel, meaning that during runtime, all detection branches will work simultaneously to perform defect detection on the input fabric images in parallel. Through parallel connection, a complete parallel fabric defect detection channel is formed. This channel can process multiple sample fabric images simultaneously and perform defect detection at different random locations, thereby improving detection efficiency and accuracy.

[0037] The K first defect detection results and the K second defect detection results are integrated to generate the target defect detection result.

[0038] By comparing K first defect detection results and K second defect detection results, including information such as the location, size, and type of the defect, duplicate defect detection results are removed. For example, if two detection results are very consistent in terms of the location and type of the defect, one of the results is selected as the final result. After integration, the target defect detection result is obtained.

[0039] Based on the target defect detection results, a control scheme is identified and a target control scheme is generated.

[0040] Based on the target defect detection results, identify which parameters in the control scheme cause the defects, and compensate the control scheme based on historical experience and in conjunction with the expert team to generate a target control scheme. This scheme is used to guide the implementation of quality control and corrective measures in the production process, thereby improving production efficiency and product quality, and reducing scrap rate and costs.

[0041] In summary, the embodiments of this application have at least the following technical effects: First, the fabric printing and dyeing equipment is configured based on the fabric information and printing requirements of the target fabric, resulting in K fabric printing plates and K fabric printing patterns. Each of the K fabric printing plates has a printing number. Next, pattern complexity analysis is performed on each of the K fabric printing patterns to determine K first control coefficients. Based on the K printing numbers, neighborhood association is performed on the K fabric printing patterns to determine K second control coefficients. Then, based on the K first control coefficients, K second control coefficients, and K fabric printing patterns, K first detection branches and K second detection branches of a parallel fabric defect detection channel are constructed. When the K fabric printing plates have finished printing on the fabric, K color CCD cameras positioned above the K fabric printing plates are used to acquire images of the fabric, obtaining K fabric images. Then, the K first detection branches and K second detection branches are used to perform defect detection on the K fabric images, generating K first defect detection results and K second defect detection results. Furthermore, the K first defect detection results and K second defect detection results are integrated to generate target defect detection results. Finally, based on the target defect detection results, control scheme identification is performed to generate a target control scheme. This solves the technical problem in existing technologies where the efficiency and accuracy of defect detection in textile printing and dyeing processes are low, leading to a failure to adjust control schemes in a timely manner and resulting in reduced printing and dyeing quality. By optimizing the control scheme through defect detection results, the technical effect of improving printing and dyeing quality is achieved.

[0042] Example 2, based on the same inventive concept as the defect detection-based control method in the foregoing examples, such as... Figure 2 As shown, this application provides a defect detection-based control system, wherein the system includes: Configuration module 11 is used to configure the weaving and printing device according to the fabric information and printing requirements of the target fabric, and obtain K weaving printing plates and K weaving printing patterns, wherein the K weaving printing plates have K printing numbers; Analysis module 12 is used to analyze the pattern complexity of the K weaving printing patterns respectively, and determine K first control coefficients; Association module 13 is used to perform neighborhood association on the K weaving printing patterns based on the K printing numbers, and determine K second control coefficients; Detection channel construction module 14 is used to construct K first detection branches and K second detection branches of a parallel fabric defect detection channel based on the K first control coefficients, the K second control coefficients and the K weaving printing patterns. The system includes: a detection branch; an image acquisition module 15, which, when K fabric printing plates are printed on the fabric, calls K color CCD cameras positioned above the K fabric printing plates to acquire images of the fabric, obtaining K fabric images; a defect detection module 16, which uses the K first detection branches and the K second detection branches to perform defect detection on the K fabric images, generating K first defect detection results and K second defect detection results; an integration module 17, which integrates the K first defect detection results and the K second defect detection results to generate a target defect detection result; and an identification module 18, which identifies a control scheme based on the target defect detection result and generates a target control scheme.

[0043] Furthermore, the parsing module 12 is used to perform the following methods: Extract K pattern size features and K pattern distribution uniformity features from the K woven fabric printing patterns; use a pattern complexity recognizer to identify the K pattern size features and K pattern distribution uniformity features to obtain the K first control coefficients.

[0044] Furthermore, the association module 13 is used to perform the following method: Based on the K printing numbers, any two adjacent fabric printing patterns are grouped into a pattern group to obtain K-1 pattern groups; the distance between the centers of two adjacent fabric printing patterns in each of the K-1 pattern groups is statistically analyzed to obtain the center distance of the K-1 pattern groups; the ratio of the center distance of each of the K-1 pattern groups to the sum of the center distances of the K-1 pattern groups is calculated, and the reciprocal of the ratio is used as the control coefficient of the K-1 pattern groups; the control coefficient of the K-1 pattern groups is retrieved using the K fabric printing patterns as an index to obtain the set of control coefficients of the K pattern groups of the K fabric printing patterns; the average value of the set of control coefficients of the K pattern groups is calculated by traversing the set of control coefficients of the K pattern groups to obtain the K second control coefficients.

[0045] Furthermore, the detection channel construction module 14 is used to perform the following method: Positioning points are configured based on the K first control coefficients and the K second control coefficients to generate K first positioning points and K second positioning points. Edge points are randomly extracted from the K woven fabric printing patterns based on the K first positioning points and the K second positioning points, resulting in K sets of first random positioning points and K sets of second random positioning points. The K sets of first random positioning points and the K sets of second random positioning points are sequentially connected to generate K first randomly extracted patterns and K second randomly extracted patterns. The K first randomly extracted patterns and the K second randomly extracted patterns are verified using the K woven fabric printing patterns. If the verification is successful, the K first detection branches and the K second detection branches of a parallel fabric defect detection channel are constructed based on the K sets of first random positioning points and the K sets of second random positioning points.

[0046] Furthermore, the detection channel construction module 14 is used to perform the following method: A Siamese neural network is obtained, wherein the Siamese neural network includes a first Siamese branch, a second Siamese branch, and a loss function; the K fabric printing patterns are input into the first Siamese branch, and the K first randomly extracted patterns are input into the second Siamese branch to obtain K first printed pattern feature sets and K first random pattern feature sets; the loss function is used to calculate K first random loss values ​​for the K first printed pattern feature sets and K first random pattern feature sets; the K fabric printing patterns and the K first randomly extracted patterns are input into the Siamese neural network to obtain K second random loss values; it is determined whether the K first random loss values ​​and the K second random loss values ​​are greater than or equal to a preset loss value threshold; if not, the verification is successful.

[0047] Furthermore, the detection channel construction module 14 is used to perform the following method: The loss function is: Where MSE is the first random loss, n is the total number of features in the first printed pattern feature set, and n is a positive integer greater than or equal to 1. Let i be the i-th feature vector of the first printed pattern feature set. Let be the i-th feature vector of the first random pattern feature set.

[0048] Furthermore, the detection channel construction module 14 is used to perform the following method: K sample fabric images are collected, and defect detection is performed on K first random positioning points in each of the K sample fabric images to obtain K sample first defect detection results. K sample fabric images are also collected, and defect detection is performed on K second random positioning points in each of the K sample fabric images to obtain K sample second defect detection results. Using the K sample fabric images, the K first random positioning points, and the K sample first defect detection results as construction data, K first detection branches are constructed. Using the K sample fabric images, the K second random positioning points, and the K sample second defect detection results as construction data, K second detection branches are constructed. The K first detection branches and the K second detection branches are then connected in parallel to generate the parallel fabric defect detection channel.

[0049] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0050] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0051] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A control method based on defect detection, characterized in that, The method includes: Based on the fabric information and printing requirements of the target fabric, the weaving and dyeing equipment is configured to obtain K weaving printing plates and K weaving printing patterns, wherein the K weaving printing plates have K printing numbers. The pattern complexity of the K fabric printing patterns is analyzed to determine the K first control coefficients; Based on the K printing numbers, neighborhood association is performed on the K fabric printing patterns to determine K second control coefficients; Based on the K first control coefficients, the K second control coefficients, and the K fabric printing patterns, construct K first detection branches and K second detection branches for a parallel fabric defect detection channel; When the K fabric printing plates have finished printing on the fabric, the K color CCD cameras positioned above the K fabric printing plates are used to capture images of the fabric and obtain K fabric images. The K first detection branches and the K second detection branches are used to perform defect detection on the K fabric images respectively, generating K first defect detection results and K second defect detection results; The K first defect detection results and the K second defect detection results are integrated to generate the target defect detection result; Based on the target defect detection results, a control scheme is identified and a target control scheme is generated.

2. The control method based on defect detection as described in claim 1, characterized in that, include: Extract the K pattern size features and K pattern distribution uniformity features of the K fabric printing patterns; The K pattern size features and the K pattern distribution uniformity features are identified using a pattern complexity recognizer to obtain the K first control coefficients.

3. The control method based on defect detection as described in claim 1, characterized in that, include: Based on the K printing numbers, any two adjacent fabric printing patterns are grouped into a pattern group to obtain K-1 pattern groups; The distance between the centers of two adjacent fabric printing patterns in the K-1 pattern groups is statistically analyzed to obtain the center distance of the K-1 pattern groups. Calculate the ratio of the center distance of each of the K-1 pattern groups to the sum of the center distances of the K-1 pattern groups, and use the reciprocal of the ratio as the control coefficient of the K-1 pattern groups; Using K fabric printing patterns as indexes, the control coefficients of the K-1 pattern groups are retrieved to obtain the set of K pattern group control coefficients for the K fabric printing patterns; The average value of the K pattern group control coefficient sets is calculated to obtain the K second control coefficients.

4. The control method based on defect detection as described in claim 1, characterized in that, include: Based on the K first control coefficients and the K second control coefficients, positioning points are configured to generate K first positioning point quantities and K second positioning point quantities; Based on the number of the K first positioning points and the number of the K second positioning points, edge points are randomly extracted from the K woven and printed patterns to obtain a set of K first random positioning points and a set of K second random positioning points. The K sets of first random positioning points and the K sets of second random positioning points are connected sequentially to generate K first random extraction patterns and K second random extraction patterns; The K first randomly extracted patterns and the K second randomly extracted patterns are verified using the K fabric printing patterns. If the verification is successful, the K first detection branches and the K second detection branches of the parallel fabric defect detection channel are constructed based on the K first random positioning point sets and the K second random positioning point sets.

5. The control method based on defect detection as described in claim 4, characterized in that, include: Obtain a Siamese neural network, wherein the Siamese neural network includes a first Siamese branch, a second Siamese branch, and a loss function; The K fabric printing patterns are input into the first twin branch, and the K first randomly extracted patterns are input into the second twin branch to obtain the K first printed pattern feature sets and the K first random pattern feature sets; The loss function is used to calculate the K first random loss values ​​of the K first printed pattern feature sets and the K first random pattern feature sets; The K fabric printing patterns and the K first randomly extracted patterns are input into the twin neural network to obtain K second random loss values; Determine whether the K first random loss values ​​and the K second random loss values ​​are greater than or equal to a preset loss value threshold. If not, the verification is successful.

6. The control method based on defect detection as described in claim 5, characterized in that, The loss function is: ； Where MSE is the first random loss, n is the total number of features in the first printed pattern feature set, and n is a positive integer greater than or equal to 1. Let i be the i-th feature vector of the first printed pattern feature set. Let be the i-th feature vector of the first random pattern feature set.

7. The control method based on defect detection as described in claim 1, characterized in that, include: Collect a set of K sample fabric images, and perform defect detection on the K first random positioning points in the set of K sample fabric images to obtain the first defect detection results of K samples; Collect a set of K sample fabric images, and perform defect detection on the K sets of K second random positioning points in the set of K sample fabric images to obtain the second defect detection results of K samples; Using the set of K sample fabric images, the set of K first random positioning points, and the detection results of the first defects of the K samples as construction data, the K first detection branches are constructed. Using the set of K sample fabric images, the set of K second random positioning points, and the detection results of the K sample second defects as construction data, the K second detection branches are constructed. The K first detection branches and the K second detection branches are connected in parallel to generate the parallel fabric defect detection channel.

8. A control system based on defect detection, characterized in that, The system is used to implement the defect detection-based control method according to any one of claims 1-7, the system comprising: The configuration module is used to configure the weaving and printing device according to the fabric information and printing requirements of the target fabric, and to obtain K weaving printing plates and K weaving printing patterns, wherein the K weaving printing plates have K printing numbers. The analysis module is used to analyze the pattern complexity of the K fabric printing patterns respectively and determine the K first control coefficients; The association module is used to perform neighborhood association on the K fabric printing patterns based on the K printing numbers, and determine K second control coefficients; A detection channel construction module is used to construct K first detection branches and K second detection branches of a parallel fabric defect detection channel based on the K first control coefficients, the K second control coefficients and the K fabric printing patterns; An image acquisition module is used to capture images of the fabric by calling K color CCD cameras positioned above the K fabric printing plates when the K fabric printing plates have finished printing on the fabric, thereby obtaining K fabric images. A defect detection module is used to perform defect detection on the K fabric images using the K first detection branches and the K second detection branches respectively, and generate K first defect detection results and K second defect detection results; An integration module is used to integrate the K first defect detection results and the K second defect detection results to generate a target defect detection result; The identification module is used to identify control schemes based on the target defect detection results and generate target control schemes.

Images