A clothing product quality information management system and management method

Abstract

The invention discloses a clothing product quality information management system and a management method, comprising the following steps: obtaining quality demand indicators of clothing products and clothing product images within a predetermined time period; constructing a preliminary recognition model based on the quality demand indicators, building a rule engine, importing the clothing product images into the preliminary recognition model, outputting a number of abnormal prompt signals, generating a number of abnormal type events and a corresponding defect supervision mechanism based on the number of abnormal prompt signals; extracting feature information based on the clothing product images, integrating the feature information and the output information into a target data set, importing the target data set into a pre-constructed deep recognition model for deep analysis, outputting deep analysis results, including chi-square statistics, and obtaining a defect index based on the chi-square statistics; generating a selective report; the invention can improve the accuracy of model calculation, improve the efficiency of information management and improve the overall production efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of clothing product management, and in particular to a clothing product quality information management system and a management method. Background Art

[0002] The production process of clothing products is complex, involving design, material selection, production, inspection, logistics and other links. The quality of clothing products refers to the degree to which clothing meets consumer needs and usage functions during the design, production, processing and sales process. It not only includes the appearance and materials of clothing, but also covers the performance, comfort, durability and environmental protection of clothing. Especially in the production process of clothing products, quality control and tracking are particularly important.

[0003] However, traditional clothing product quality information management often uses manual quality inspection to identify abnormalities of clothing products, which is not only inefficient but also prone to deviations. With the development of technology, machine learning algorithms are often introduced for judgment. In the process of defect judgment, images are usually used for rapid training. Omissions may occur in the processing process, making it difficult to achieve systematic data analysis, resulting in inaccurate defect judgment and identification, causing defective clothing products to directly enter the sales market, resulting in information blockage and untimely feedback. This not only affects the quality of a single product, but also affects the overall production efficiency and quality management level in the context of large-scale production, reducing the efficiency of clothing product quality management. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides a clothing product quality information management system and a management method, which build a preliminary recognition model, import the clothing product image into the preliminary recognition model, identify high or low abnormal prompt signals, generate a first abnormal event or a second abnormal event, and assign a corresponding defect supervision mechanism; extract feature information based on the clothing product image, integrate the feature information and output information into a target data set, and import the target data set into a pre-built deep recognition model for deep analysis, output the deep analysis result, the deep analysis result includes a chi-square statistic, and a defect index is obtained based on the chi-square statistic; in this process, the accuracy of classified defects is improved, the efficiency of management is further improved, and the problems raised in the background technology are solved.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] In a first aspect, the present application provides a method for managing clothing product quality information, the method comprising the following steps:

[0009] Acquire quality requirement indicators of clothing products and pictures of clothing products within a predetermined time period; wherein the quality requirement indicators include quality text descriptions and image data;

[0010] Build a preliminary recognition model based on quality demand indicators, build a rule engine, import clothing product images into the preliminary recognition model, output several abnormal prompt signals, and generate output information based on several abnormal prompt signals. The output information includes several abnormal type events and corresponding defect supervision mechanisms;

[0011] Extract feature information based on clothing product images, integrate the feature information and output information into a target data set, import the target data set into a pre-built deep recognition model for deep analysis, and output the deep analysis results, which include chi-square statistics, and obtain a defect index based on the chi-square statistics; wherein the feature information includes appearance quality, specification deviation, surface defects, and recognition time, the target data set includes a training set, a test set, and a validation set, and the deep analysis includes a first judgment analysis and a second judgment analysis;

[0012] Generates a selective report that includes the category of defect supervision mechanism, category of abnormal type event, defect index, and identification time.

[0013] Furthermore, the construction of a preliminary identification model based on quality requirement indicators includes:

[0014] Data preparation and preprocessing: Collect quality demand indicators of clothing products and preprocess them, including filtering and denoising, contrast enhancement, and data enhancement;

[0015] Feature extraction: Perform continuous convolution operations on the processed data and introduce an attention mechanism to determine feature points according to channel attention and spatial attention, and extract the corresponding feature vectors, which at least include style outline features and seam defect features;

[0016] Feature classification: Perform spherical clustering operations on each feature point, and preset the radius index of each feature point, and the radius index includes monitoring radius, radius interval and radius window; filter out abnormal feature points that exceed the monitoring radius, cut the radius value into several equidistant radius intervals based on the radius window, classify the radius values ​​of the filtered abnormal feature points into the corresponding radius intervals, and label and assign weights to several equidistant radius intervals, and keep the labels and weights synchronized;

[0017] Feature combination: weighted processing and feature concatenation of the feature vectors corresponding to the classified feature points to form a feature matrix, and each row and column of the feature matrix corresponds to a combined eigenvalue;

[0018] Event classification: Based on the combined eigenvalues ​​corresponding to the feature matrix, the product of the combined eigenvalues ​​and the frequency of abnormal feature points is marked as an outlier; the outlier is compared and analyzed with the set standard threshold, and a high or low abnormal prompt signal and a first abnormal event or a second abnormal event are generated accordingly, and the high abnormal prompt signal corresponds to the first abnormal event, and the low abnormal prompt signal corresponds to the second abnormal event.

[0019] Furthermore, the defect supervision mechanism includes:

[0020] Matching a type of defect supervision mechanism for the first abnormal event, wherein the type of defect supervision mechanism includes performing a first determination analysis on the first abnormal event;

[0021] Matching a second-type defect supervision mechanism for the second abnormal event, wherein the second-type defect mechanism includes temporarily storing the second abnormal event and performing a second determination analysis;

[0022] The first determination analysis and the second determination analysis are performed in the deep recognition model.

[0023] Furthermore, the step of constructing the depth recognition model includes:

[0024] Data preparation and processing: Collect known clothing product images to be tested, extract corresponding feature information, and combine them with the corresponding output information to integrate them into the target data set;

[0025] Model training and evaluation: Support vector machine is used as a machine learning model. The target data set is divided into training set, test set and validation set. The training set is used to train the model, select the parameters in the model and optimize the algorithm to obtain the optimal classification model. The test set is input into the trained model for classification test. The validation set is used to evaluate the classification performance of the model through accuracy, recall rate, precision and F1 score.

[0026] Model prediction: Use the trained model to predict new abnormal events and determine whether they meet quality requirements.

[0027] Furthermore, the process of training the model with the training set includes: calculating the distribution coefficient through the mean of the training set, and calculating the expected abnormal frequency of each group of training, calculating the chi-square statistic by comparing the actual abnormal probability and the expected abnormal frequency in the training set, sorting the distribution coefficient in descending order, and drawing the corresponding dynamic change curve in combination with the corresponding chi-square statistic, and selecting the distribution coefficient with the most stable curve change as the optimal parameter, and iterating the cycle.

[0028] Furthermore, the first determination analysis step includes: identifying a first evaluation parameter set for each first abnormal event, performing dimensionless processing on the first evaluation parameter set, and generating a first defect index corresponding to each first abnormal event; wherein the first evaluation parameter set includes an abnormal value, a relative deviation coefficient, and an identification time of each first abnormal event; the second determination analysis step includes: identifying a second evaluation parameter set for each second abnormal event, performing dimensionless processing on the second evaluation parameter set, and generating a second defect index corresponding to each second abnormal event; wherein the second evaluation parameter set includes a chi-square statistic, a feedback evaluation coefficient, and an identification time of each second abnormal event.

[0029] Furthermore, the step of obtaining the relative deviation coefficient comprises:

[0030] Construct an interval analysis model: mark the input information as interval Q, and interval Q has several sub-intervals, integrate the points belonging to the same interval, obtain the regional area proportions of several sub-intervals, and integrate and mark the regional area proportions of each region as a proportion set; assign corresponding weight factors to several sub-intervals respectively, and integrate and mark the weight factors of each sub-interval as a weight set; obtain the corresponding deviation proportion by multiplying and accumulating the regional area proportions and the corresponding weight factors, and each sub-interval corresponds to a deviation proportion, and the deviation proportions include low deviation proportion, medium deviation proportion and high deviation proportion; wherein, mark any sub-interval as i, mark the regional area proportion of interval i as Si, and mark the weight factor of interval i as Mi;

[0031] Apply the interval analysis model: extract the appearance quality from the extracted feature information. The appearance quality includes the color value and the fabric density value. For any clothing product, divide it into N1 regions, mark any region as N2, extract N3 points under region N2 and their color values ​​and fabric density values, mark any point as d, and mark the color value of point d as Sd and the fabric density value as ld; and set M1 chromaticity intervals and M2 density intervals respectively, substitute them into the interval analysis model respectively, and obtain the first deviation proportion zb and the second deviation proportion zb2;

[0032] Interval analysis: By comparing and analyzing the corresponding deviation ratio with the preset first comparison interval, second comparison interval and third comparison interval, the corresponding correction coefficient is assigned, and the formula is set to obtain the relative deviation coefficient; then the cumulative sum of the product of the relative deviation coefficient and the abnormal value is marked as the first defect index; wherein, the first comparison interval is smaller than the second comparison interval and smaller than the third comparison interval;

[0033] The formula on which the relative coefficient of deviation is based is: ;

[0034] Where pc represents the relative deviation coefficient, and λ1 and λ2 are matching correction coefficients.

[0035] Furthermore, the step of obtaining the feedback evaluation coefficient includes:

[0036] Specification deviations and surface defects are extracted from the extracted feature information and defined as key parameters. Grade scores and weights corresponding to the identification parameters are assigned, and the grade scores and weights are kept synchronized. The grade scores and weights are multiplied to obtain the corresponding feedback evaluation coefficients. The cumulative sum of the product of the feedback evaluation coefficients and the outliers is then marked as the second defect index.

[0037] In a second aspect, the present application provides a clothing product quality information management system, the system comprising:

[0038] A data acquisition module, used to acquire quality requirement indicators of clothing products and pictures of clothing products within a predetermined time period; wherein the quality requirement indicators include quality text descriptions and image data;

[0039] The preliminary identification module is used to build a preliminary identification model based on quality requirement indicators, build a rule engine, import clothing product images into the preliminary identification model, output a number of abnormal prompt signals, and generate output information based on the abnormal prompt signals. The output information includes a number of abnormal type events and corresponding defect supervision mechanisms;

[0040] A deep recognition module is used to extract feature information based on a clothing product image, integrate the feature information and output information into a target data set, import the target data set into a pre-built deep recognition model for deep analysis, and output a deep analysis result, the deep analysis result includes a chi-square statistic, and a defect index is obtained based on the chi-square statistic; wherein the feature information includes appearance quality, specification deviation, surface defects, and recognition time, the target data set includes a training set, a test set, and a validation set, and the deep analysis includes a first judgment analysis and a second judgment analysis;

[0041] The feedback module is used to generate a selective report, which includes the category of defect supervision mechanism, the category of abnormal type event, defect index and identification time.

[0042] (III) Beneficial effects

[0043] The present invention provides a clothing product quality information management system and management method, which have the following beneficial effects:

[0044] 1. The present invention provides data support for subsequent data analysis by setting quality demand indicators, which improves the accuracy of analysis to a certain extent; by building a rule engine, analyzing abnormal values ​​through several feature matrices, and generating several abnormal prompt signals through threshold analysis; by identifying the clothing product images to be tested within a predetermined time period, matching the corresponding defect supervision mechanism, a first type of defect supervision mechanism performs a first judgment analysis on the first abnormal event, and a second type of defect supervision mechanism performs a second judgment analysis on the second abnormal event, further performing the corresponding judgment analysis through the deep recognition model, and identifying the corresponding defect index, and generating a selective report; to a certain extent, the accuracy of clothing quality identification is improved, and the efficiency of information management is improved;

[0045] 2. The present invention sorts the distribution coefficients in descending order, draws the corresponding dynamic change curve in combination with the corresponding chi-square statistic, and selects the distribution coefficient with the most stable curve change as the optimal parameter, thereby ensuring that the prediction results output by the deep recognition model have higher accuracy and precision. Through iterative cycles, the deep recognition model is refined and adjusted to a certain extent, which can better adapt to the natural fluctuations of clothing product process data and further improve the stability of quality management. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of a method for managing clothing product quality information according to an exemplary embodiment;

[0047] Figure 2 It is a module diagram of a clothing product quality information management system according to an exemplary embodiment. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Example 1

[0050] An embodiment of the present invention provides a method for managing clothing product quality information. Figure 1 FIG. 1 is a flow chart of a method for managing clothing product quality information according to an exemplary embodiment. Figure 1 , the method comprises the following steps:

[0051] Acquire quality requirement indicators of clothing products and pictures of clothing products within a predetermined time period; wherein the quality requirement indicators include quality text descriptions and image data;

[0052] It should be noted that the quality text description includes type description, appearance description, size description, and fabric description, and the image data is the picture data that meets the quality requirement indicators; at the same time, when the corresponding clothing products are produced, the corresponding images of the clothing products will be taken and stored in the database as the clothing product pictures to be tested;

[0053] Build a preliminary recognition model based on quality demand indicators, build a rule engine, import clothing product images into the preliminary recognition model, output several abnormal prompt signals, and generate output information based on several abnormal prompt signals. The output information includes several abnormal type events and corresponding defect supervision mechanisms;

[0054] Among them, a preliminary identification model is built based on quality requirement indicators, including:

[0055] Data preparation and preprocessing: Collect quality demand indicators of clothing products and preprocess them, including filtering and denoising, contrast enhancement, and data enhancement;

[0056] Filter denoising: used to remove various types of noise introduced during the original image data collection process, and retain the detailed features of the style outline and seams; contrast enhancement: used to increase the contrast; data enhancement: used to improve the generalization ability of the model;

[0057] Feature extraction: Perform continuous convolution operations on the processed data and introduce an attention mechanism to determine feature points according to channel attention and spatial attention, and extract the corresponding feature vectors, which include color features, texture features, style outline features, and seam defect features;

[0058] Feature classification: Perform spherical clustering operations on each feature point, and preset the radius index of each feature point, and the radius index includes monitoring radius, radius interval and radius window; filter out abnormal feature points that exceed the monitoring radius, cut the radius value into several equidistant radius intervals based on the radius window, classify the radius values ​​of the filtered abnormal feature points into the corresponding radius intervals, and label and assign weights to several equidistant radius intervals, and keep the labels and weights synchronized;

[0059] Feature combination: weighted processing and feature concatenation of the feature vectors corresponding to the classified feature points to form a feature matrix, and each row and column of the feature matrix corresponds to a combined eigenvalue;

[0060] Setting of radius index: Through spherical clustering operation, each cluster is quantified as a "spherical" feature, which not only reflects the size of each cluster, but also helps determine the clustering density and spatial distribution of feature points;

[0061] Setting of radius interval: the range of distance from the feature points in the cluster to the cluster center, that is, the minimum radius value and the maximum radius value of each cluster, and the cluster center is the center of the sphere, and its center is the mean value of each feature point in the cluster;

[0062] The monitoring radius is set as follows: the mean value of the radius interval and a certain multiple of the standard deviation, and the multiple is between 2 and 3;

[0063] Radius window setting: Specifically, it is a sliding window used to locally smooth or aggregate the distance from the feature points in the cluster to the center of the cluster. First, the distances of all points are sorted according to the distances from the feature points in each cluster to the center, and a window size (for example, 10 points) is selected. The radius range of each abnormal feature point in the window is calculated and the corresponding labels and weights are assigned.

[0064] For example, for the radius interval number 1, a weight of 1 is assigned, for the radius interval number 2, a weight of 2 is assigned, and so on. The weight 1 and weight 2 at this time are just an example, and the specific values ​​are explained according to the actual situation; it can be expressed that the larger the number, the greater the weight;

[0065] Event classification: Based on the combined eigenvalues ​​corresponding to the feature matrix, the product of the combined eigenvalues ​​and the frequency of abnormal feature points is marked as an outlier; the outlier is compared and analyzed with the set standard threshold:

[0066] When the abnormal value is greater than or equal to the standard threshold, a high abnormality prompt signal is generated, and the abnormal value is marked as qx1, and edited as a first-level character at the same time, and qx1 and the first-level character are combined to generate the first abnormal event;

[0067] When the abnormal value is less than the standard threshold, a low abnormality prompt signal is generated, and the abnormal value is marked as qx2, and edited as a secondary character at the same time, and qx2 and the secondary character are combined to generate a second abnormal event;

[0068] Extract feature information based on clothing product images, integrate feature information and output information into a target data set, import the target data set into a pre-built deep recognition model for deep analysis, and output deep analysis results, which include chi-square statistics, and obtain a defect index based on the chi-square statistics;

[0069] The feature information includes appearance quality, specification deviation, surface defects and recognition time, the target data set includes training set, test set and validation set, and the in-depth analysis includes first judgment analysis and second judgment analysis; it should be noted that NLP technology and named entity recognition technology can be used to extract feature information, and the extraction process is not described in detail here;

[0070] Among them, the steps of building a deep recognition model include:

[0071] Data preparation and processing: Collect known clothing product images to be tested, extract corresponding feature information, and combine them with the corresponding output information to integrate them into the target data set;

[0072] The defect supervision mechanism specifically includes: matching a first abnormal event with a first defect supervision mechanism, and the first defect supervision mechanism includes performing a first determination analysis on the first abnormal event; matching a second abnormal event with a second defect supervision mechanism, and the second defect mechanism includes temporarily storing the second abnormal event and performing a second determination analysis; and the first determination analysis and the second determination analysis are performed in the deep recognition model;

[0073] Model training and evaluation: Support vector machine is used as a machine learning model. The target data set is divided into training set, test set and validation set. The model is trained with the training set, the parameters in the model and the optimization algorithm are selected to obtain the optimal classification model. The test set is input into the trained model for classification test. The classification performance of the model is evaluated with the validation set through accuracy, recall rate, precision and F1 score.

[0074] The training process of the training set includes: calculating the distribution coefficient by the mean of the training set, calculating the expected abnormal frequency of each training group, and calculating the chi-square statistic by comparing the actual abnormal probability and the expected abnormal frequency in the training set;

[0075] The distribution coefficient is based on the formula: ;

[0076] In the formula, Distribution represents the distribution coefficient, Num represents the total number of outliers in the corresponding training set, and prob0 represents the actual anomaly probability of the corresponding training set, that is, the probability of an actual outlier occurring.

[0077] The formula on which the chi-square statistic is based is: ;

[0078] In the formula, represents the chi-square statistic of the rth training set, represents the expected abnormal frequency of the bth item in the rth group;

[0079] Sort the distribution coefficients in descending order, and draw the corresponding dynamic change curve in combination with the corresponding chi-square statistic, and select the distribution coefficient with the most stable curve change as the optimal parameter, and iterate the cycle;

[0080] The steps of performing the first decision analysis include:

[0081] By identifying a first evaluation parameter set of each first abnormal event, performing dimensionless processing on the first evaluation parameter set, and generating a first defect index corresponding to each first abnormal event; wherein the first evaluation parameter set includes an abnormal value, a relative deviation coefficient, and an identification time of each first abnormal event;

[0082] The steps to obtain the relative deviation coefficient are:

[0083] Construct an interval analysis model: mark the input information as interval Q, and interval Q has several sub-intervals, integrate the points belonging to the same interval, obtain the regional area proportions of several sub-intervals, and integrate and mark the regional area proportions of each region as a proportion set; assign corresponding weight factors to several sub-intervals respectively, and integrate and mark the weight factors of each sub-interval as a weight set; obtain the corresponding deviation proportion by multiplying and accumulating the regional area proportions and the corresponding weight factors, and each sub-interval corresponds to a deviation proportion, and the deviation proportions include low deviation proportion, medium deviation proportion and high deviation proportion; wherein, mark any sub-interval as i, mark the regional area proportion of interval i as Si, and mark the weight factor of interval i as Mi;

[0084] Apply the interval analysis model: extract the appearance quality from the extracted feature information. The appearance quality includes the color value and the fabric density value. For any clothing product, divide it into N1 regions, mark any region as N2, extract N3 points under region N2 and their color values ​​and fabric density values, mark any point as d, and mark the color value of point d as Sd and the fabric density value as ld; and set M1 chromaticity intervals and M2 density intervals respectively, substitute them into the interval analysis model respectively, and obtain the first deviation proportion zb and the second deviation proportion zb2;

[0085] The first deviation proportion zb1: ;

[0086] In the formula, sj represents the chromaticity area proportion Sj, βj represents the weight factor corresponding to the chromaticity interval, and βj is greater than 0;

[0087] The second deviation accounts for zb2: ;

[0088] In the formula, dj represents the density area ratio, φj represents the weight factor corresponding to the density interval, and φj is greater than 0;

[0089] Interval analysis: By comparing the deviation ratio with the preset first comparison interval, second comparison interval and third comparison interval respectively:

[0090] When the corresponding deviation ratio is in the first comparison interval, a first correction coefficient is allocated;

[0091] When the corresponding deviation ratio is in the second comparison interval, a second correction coefficient is allocated;

[0092] When the corresponding deviation ratio is in the third comparison interval, a third correction coefficient is allocated;

[0093] It should be noted that the maximum interval of the first comparison interval is smaller than the minimum interval of the second comparison interval, and the maximum interval of the second comparison interval is smaller than the minimum interval of the third comparison interval; in this embodiment, the comparison interval is defined as the historical deviation proportion data is sorted in order and separated by the tertile interval, the first tertile takes the position of 33.33% of the data set, and the second tertile takes the position of 66.67% of the data set, thereby dividing it into the first comparison interval, the second comparison interval and the third comparison interval;

[0094] Set the formula to obtain the relative deviation coefficient: ;

[0095] In the formula, pc represents the relative deviation coefficient, λ1 and λ2 are the matching correction coefficients;

[0096] Then the cumulative sum of the product of the relative deviation coefficient and the outlier value is marked as the first defect index;

[0097] The steps of performing the second decision analysis include:

[0098] By identifying a second evaluation parameter set of each second abnormal event, performing dimensionless processing on the second evaluation parameter set, and generating a second defect index corresponding to each second abnormal event; wherein the second evaluation parameter set includes a chi-square statistic, a feedback evaluation coefficient, and an identification time of each second abnormal event;

[0099] The process of feedback evaluation coefficient is as follows: extract specification deviation and surface defects from the extracted feature information and define them as key parameters, assign grade scores and weights corresponding to the identification parameters, and keep the grade scores and weights synchronized, and multiply the grade scores and weights to obtain the corresponding feedback evaluation coefficient;

[0100] For example, specification deviations include length deviation, waist circumference deviation, and sleeve / trouser / skirt length deviation; surface defects include printing deviation, fabric defects, and thread ends;

[0101] Specification deviation: clothing length error, assigned weight 3> waist circumference error, assigned weight 2> sleeve length / trouser length / skirt length error, assigned weight 1;

[0102] Surface defects: printing deviation, assigned weight 3> fabric defects, assigned weight 2> thread ends, assigned weight 1;

[0103] Then, the cumulative sum of the product of the feedback evaluation coefficient and the abnormal value is marked as the second defect index;

[0104] Model prediction: Use the trained model to predict new abnormal events and determine whether they meet quality requirements.

[0105] Generate selective reports: including defect supervision mechanism categories, abnormal event categories, defect index, and identification time; by uploading the generated reports to appropriate platforms and displaying them (for example: cloud storage, file servers, web applications), relevant personnel can make corresponding decisions based on the generated reports.

[0106] In summary, the present invention constructs a preliminary recognition model based on quality demand indicators, identifies clothing product images to be tested within a predetermined time period through clothing product images, identifies high or low abnormal prompt signals, generates a first abnormal event or a second abnormal event, and allocates a corresponding defect supervision mechanism; in this process, the number of training samples of the preliminary recognition model is greatly reduced, and the accuracy of preliminary recognition is improved; by extracting feature information based on clothing product images, and integrating the feature information and output information into a target data set, and then importing the target data set into a pre-built deep recognition model for deep analysis, the deep analysis results are output, and the deep analysis results include a chi-square statistic, and a defect index is obtained based on the chi-square statistic; in this process, the accuracy of classified defects is improved, the accuracy of model calculation is further improved, and the efficiency of management is improved.

[0107] Example 2

[0108] An embodiment of the present invention provides a clothing product quality information management system. Figure 2 is a schematic diagram of a module of a clothing product quality information management system according to an exemplary embodiment. Figure 2 The system includes: a data acquisition module, a preliminary recognition module, a depth recognition module and a feedback module, and the data acquisition module, the preliminary recognition module, the depth recognition module and the feedback module are communicatively connected;

[0109] A data acquisition module, used to acquire quality requirement indicators of clothing products and pictures of clothing products within a predetermined time period; wherein the quality requirement indicators include quality text descriptions and image data;

[0110] The preliminary identification module is used to build a preliminary identification model based on quality requirement indicators, build a rule engine, import clothing product images into the preliminary identification model, output a number of abnormal prompt signals, and generate output information based on the abnormal prompt signals. The output information includes a number of abnormal type events and corresponding defect supervision mechanisms;

[0111] A deep recognition module is used to extract feature information based on a clothing product image, integrate the feature information and output information into a target data set, import the target data set into a pre-built deep recognition model for deep analysis, and output a deep analysis result, the deep analysis result includes a chi-square statistic, and a defect index is obtained based on the chi-square statistic; wherein the feature information includes appearance quality, specification deviation, surface defects, and recognition time, the target data set includes a training set, a test set, and a validation set, and the deep analysis includes a first judgment analysis and a second judgment analysis;

[0112] The feedback module is used to generate a selective report, which includes the category of defect supervision mechanism, the category of abnormal type event, defect index and identification time.

[0113] In the application, the several formulas involved are all calculated by removing dimensions and taking their numerical values, and the formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The formula is set by technical personnel in this field according to actual conditions.

[0114] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0115] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0116] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for managing clothing product quality information, characterized in that: The steps include: Acquire quality requirement indicators of clothing products and pictures of clothing products within a predetermined time period; wherein the quality requirement indicators include quality text descriptions and image data; Build a preliminary recognition model based on quality demand indicators, build a rule engine, import clothing product images into the preliminary recognition model, output several abnormal prompt signals, and generate output information based on several abnormal prompt signals. The output information includes several abnormal type events and corresponding defect supervision mechanisms; Among them, a preliminary identification model is built based on quality requirement indicators, including: Data preparation and preprocessing: Collect quality demand indicators of clothing products and preprocess them, including filtering and denoising, contrast enhancement, and data enhancement; Feature extraction: Perform continuous convolution operations on the processed data and introduce an attention mechanism to determine feature points according to channel attention and spatial attention, and extract the corresponding feature vectors, which at least include style outline features and seam defect features; Feature classification: Perform spherical clustering operations on each feature point, and preset the radius index of each feature point, and the radius index includes monitoring radius, radius interval and radius window; filter out abnormal feature points that exceed the monitoring radius, cut the radius value into several equidistant radius intervals based on the radius window, classify the radius values ​​of the filtered abnormal feature points into the corresponding radius intervals, and label and assign weights to several equidistant radius intervals, and keep the labels and weights synchronized; Feature combination: weighted processing and feature concatenation of the feature vectors corresponding to the classified feature points to form a feature matrix, and each row and column of the feature matrix corresponds to a combined eigenvalue; Event classification: Based on the combined eigenvalues ​​corresponding to the feature matrix, the product of the combined eigenvalues ​​and the frequency of the abnormal feature points is marked as an abnormal value; the abnormal value is compared and analyzed with the set standard threshold, and a high or low abnormal prompt signal and a first abnormal event or a second abnormal event are generated accordingly, and the high abnormal prompt signal corresponds to the first abnormal event, and the low abnormal prompt signal corresponds to the second abnormal event; Extract feature information based on clothing product images, integrate the feature information and output information into a target data set, import the target data set into a pre-built deep recognition model for deep analysis, and output the deep analysis results, which include chi-square statistics, and obtain a defect index based on the chi-square statistics; wherein the feature information includes appearance quality, specification deviation, surface defects, and recognition time, the target data set includes a training set, a test set, and a validation set, and the deep analysis includes a first judgment analysis and a second judgment analysis; Generates a selective report that includes the category of defect supervision mechanism, category of abnormal type event, defect index, and identification time.

2. A clothing product quality information management method according to claim 1, characterized in that: The defect supervision mechanism includes: Matching a type of defect supervision mechanism for the first abnormal event, wherein the type of defect supervision mechanism includes performing a first determination analysis on the first abnormal event; Matching a second-type defect supervision mechanism for the second abnormal event, wherein the second-type defect mechanism includes temporarily storing the second abnormal event and performing a second determination analysis; The first determination analysis and the second determination analysis are performed in the deep recognition model.

3. A clothing product quality information management method according to claim 1, characterized in that: The steps of constructing the depth recognition model include: Data preparation and processing: Collect known clothing product images to be tested, extract corresponding feature information, and combine them with the corresponding output information to integrate them into the target data set; Model training and evaluation: Support vector machine is used as a machine learning model. The target data set is divided into training set, test set and validation set. The training set is used to train the model, select the parameters in the model and optimize the algorithm to obtain the optimal classification model. The test set is input into the trained model for classification test. The validation set is used to evaluate the classification performance of the model through accuracy, recall rate, precision and F1 score. Model prediction: Use the trained model to predict new abnormal events and determine whether they meet quality requirements.

4. A clothing product quality information management method according to claim 3, characterized in that: The process of training the model with the training set includes: calculating the distribution coefficient through the mean of the training set, and calculating the expected abnormal frequency of each group of training, calculating the chi-square statistic by comparing the actual abnormal probability and the expected abnormal frequency in the training set, sorting the distribution coefficient in descending order, and drawing the corresponding dynamic change curve in combination with the corresponding chi-square statistic, and selecting the distribution coefficient with the most stable curve change as the optimal parameter, and iterating the cycle.

5. A clothing product quality information management method according to claim 2, characterized in that: The first determination and analysis step includes: identifying a first evaluation parameter set of each first abnormal event, performing dimensionless processing on the first evaluation parameter set, and generating a first defect index corresponding to each first abnormal event; wherein the first evaluation parameter set includes an abnormal value, a relative deviation coefficient, and an identification time of each first abnormal event; The step of the second determination analysis includes: identifying a second evaluation parameter set for each second abnormal event, performing dimensionless processing on the second evaluation parameter set, and generating a second defect index corresponding to each second abnormal event; wherein the second evaluation parameter set includes a chi-square statistic, a feedback evaluation coefficient, and an identification time of each second abnormal event.

6. A clothing product quality information management method according to claim 5, characterized in that: The step of obtaining the relative deviation coefficient comprises: Construct an interval analysis model: mark the input information as interval Q, and interval Q has several sub-intervals, integrate the points belonging to the same interval, obtain the regional area proportions of several sub-intervals, and integrate and mark the regional area proportions of each region as a proportion set; assign corresponding weight factors to several sub-intervals respectively, and integrate and mark the weight factors of each sub-interval as a weight set; obtain the corresponding deviation proportion by multiplying and accumulating the regional area proportions and the corresponding weight factors, and each sub-interval corresponds to a deviation proportion, and the deviation proportions include low deviation proportion, medium deviation proportion and high deviation proportion; wherein, mark any sub-interval as i, mark the regional area proportion of interval i as Si, and mark the weight factor of interval i as Mi; Apply the interval analysis model: extract the appearance quality from the extracted feature information. The appearance quality includes the color value and the fabric density value. For any clothing product, divide it into N1 regions, mark any region as N2, extract N3 points under region N2 and their color values ​​and fabric density values, mark any point as d, and mark the color value of point d as Sd and the fabric density value as ld; and set M1 chromaticity intervals and M2 density intervals respectively, substitute them into the interval analysis model respectively, and obtain the first deviation proportion zb and the second deviation proportion zb2; Interval analysis: By comparing and analyzing the corresponding deviation ratio with the preset first comparison interval, second comparison interval and third comparison interval, the corresponding correction coefficient is assigned, and the formula is set to obtain the relative deviation coefficient; then the cumulative sum of the product of the relative deviation coefficient and the abnormal value is marked as the first defect index; among which, the first comparison interval is smaller than the second comparison interval and smaller than the third comparison interval; the formula based on the relative deviation coefficient is: ; Where pc represents the relative deviation coefficient, and λ1 and λ2 are matching correction coefficients.

7. A method for managing clothing product quality information according to claim 5, characterized in that: The step of obtaining the feedback evaluation coefficient comprises: Specification deviations and surface defects are extracted from the extracted feature information and defined as key parameters. Grade scores and weights corresponding to the identification parameters are assigned, and the grade scores and weights are kept synchronized. The grade scores and weights are multiplied to obtain the corresponding feedback evaluation coefficients. The cumulative sum of the product of the feedback evaluation coefficients and the outliers is then marked as the second defect index.

8. A clothing product quality information management system, characterized in that: include: A data acquisition module, used to acquire quality requirement indicators of clothing products and pictures of clothing products within a predetermined time period; wherein the quality requirement indicators include quality text descriptions and image data; The preliminary identification module is used to build a preliminary identification model based on quality requirement indicators, build a rule engine, import clothing product images into the preliminary identification model, output a number of abnormal prompt signals, and generate output information based on the abnormal prompt signals. The output information includes a number of abnormal type events and corresponding defect supervision mechanisms; Among them, a preliminary identification model is built based on quality requirement indicators, including: Data preparation and preprocessing: Collect quality demand indicators of clothing products and preprocess them, including filtering and denoising, contrast enhancement, and data enhancement; Feature extraction: Perform continuous convolution operations on the processed data and introduce an attention mechanism to determine feature points according to channel attention and spatial attention, and extract the corresponding feature vectors, which at least include style outline features and seam defect features; Feature classification: Perform spherical clustering operations on each feature point, and preset the radius index of each feature point, and the radius index includes monitoring radius, radius interval and radius window; filter out abnormal feature points that exceed the monitoring radius, cut the radius value into several equidistant radius intervals based on the radius window, classify the radius values ​​of the filtered abnormal feature points into the corresponding radius intervals, and label and assign weights to several equidistant radius intervals, and keep the labels and weights synchronized; Feature combination: weighted processing and feature concatenation of the feature vectors corresponding to the classified feature points to form a feature matrix, and each row and column of the feature matrix corresponds to a combined eigenvalue; Event classification: Based on the combined eigenvalues ​​corresponding to the feature matrix, the product of the combined eigenvalues ​​and the frequency of the abnormal feature points is marked as an abnormal value; the abnormal value is compared and analyzed with the set standard threshold, and a high or low abnormal prompt signal and a first abnormal event or a second abnormal event are generated accordingly, and the high abnormal prompt signal corresponds to the first abnormal event, and the low abnormal prompt signal corresponds to the second abnormal event; A deep recognition module is used to extract feature information based on a clothing product image, integrate the feature information and output information into a target data set, import the target data set into a pre-built deep recognition model for deep analysis, and output a deep analysis result, the deep analysis result includes a chi-square statistic, and a defect index is obtained based on the chi-square statistic; wherein the feature information includes appearance quality, specification deviation, surface defects, and recognition time, the target data set includes a training set, a test set, and a validation set, and the deep analysis includes a first judgment analysis and a second judgment analysis; The feedback module is used to generate a selective report, which includes the category of defect supervision mechanism, the category of abnormal type event, defect index and identification time.

Images