An intelligent detection method for stitching defects of overcoats

Abstract

This application relates to the field of sewing technology and provides an intelligent detection method for the sewing defects of overcoats. The method includes: obtaining the design drawing of the target overcoat for functional area division, matching the material information according to the division result, and generating a material-functional area mapping diagram; collecting sewing images zone by zone to obtain an image data stream; activating a defect feature extraction model, inputting the data stream after channel matching for defect characterization to obtain a stitch deviation stream and a feature set; collecting process parameters, matching them with the feature set and the stitch deviation stream to generate a process-deviation correlation data set; performing multi-scale fitting to identify the key data set and evaluating to generate a defect detection report. This application solves the technical problems of inaccurate defect detection and insufficient real-time performance caused by the fluctuation of process parameters in different functional areas during the sewing process of overcoats, and realizes the effect of improving the detection accuracy and adaptability by combining the synchronous analysis of process parameters and stitch deviation data and multi-scale fitting.

Description

Technical Field

[0001] This application relates to the field of sewing technology, and particularly to an intelligent detection method for stitching defects of overcoats. Background Art

[0002] In the traditional clothing production process, especially in the stitching process of overcoats, problems such as stitching deviation, uneven stitch density, and stitching defects in the zipper and button areas often affect the quality of the finished product. The existing stitching quality inspection mainly relies on manual inspection, which not only takes a long time and has low efficiency, but also is prone to omissions or misjudgments due to human factors. Especially in the case of complex stitching processes and diverse materials, it is difficult for manual inspection to maintain consistency. In addition, as high-end clothing, overcoats usually use a variety of different materials, such as wool, silk, polyester, etc., and these materials have different requirements for stitching processes, further increasing the difficulty of defect detection. Summary of the Invention

[0003] This application provides an intelligent detection method for stitching defects of overcoats, aiming to solve the technical problems of inaccurate defect detection and insufficient real-time performance caused by fluctuations in process parameters in different functional areas during the stitching process of overcoats.

[0004] In view of the above problems, this application provides an intelligent detection method for stitching defects of overcoats.

[0005] This application provides an intelligent detection method for stitching defects of overcoats. The method includes: obtaining the design drawing of the target overcoat, dividing the functional areas based on the design drawing of the target overcoat, matching the material information according to the division result, and generating a material-functional area mapping diagram; collecting the stitching images of each area one by one based on a high-precision visible light camera for the division result to obtain the stitching image data stream of each functional area; activating the defect feature extraction model, performing channel matching according to the material-functional area mapping diagram, and inputting the stitching image data stream of each functional area according to the matching result to extract defect characteristics, obtaining the stitching deviation stream and the stitching deviation feature set of different materials; collecting the real-time process parameters of the stitching equipment, combining the stitching deviation feature set, performing homologous functional area time series matching on the real-time process parameters of the stitching equipment and the stitching deviation stream, and generating a process-deviation correlation data set for different functional areas; performing multi-scale fitting on the process-deviation correlation data set for different functional areas to identify the key process-deviation correlation data set; inputting the key process-deviation correlation data set into the stitching defect correlation analyzer for quality assessment, and combining the stitching quality tolerance to generate a stitching defect detection report.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The above intelligent detection method for stitching defects of a coat divides the functional areas of the coat based on the design drawing of the target coat, and matches the corresponding material information according to the division results of each area to generate a material-functional area mapping diagram. The role of this mapping diagram is to ensure a one-to-one correspondence between each area and the material characteristics, providing a basis for subsequent detection. Subsequently, a high-precision visible light camera is used to collect images of the stitching process of each area one by one according to the division of the functional areas, obtaining the stitching image data stream of each area. Then, the defect feature extraction model is activated, and channel matching is performed through the material-functional area mapping diagram, and the corresponding stitching images are input into the internal channels of the model to extract the stitch deviation stream and defect feature set corresponding to each material. The purpose of this step is to ensure that the detection process can accurately identify potential stitching defects in each area based on different material characteristics. After that, the real-time process parameters of the stitching equipment are collected, and these process parameters are matched with the stitch deviation data of each functional area in the same-source time series. Through this matching, a data set reflecting the correlation between the stitching process and defects in each functional area is generated. Then, multi-scale fitting is performed on the process-deviation correlation data sets of each area to identify the process parameters and deviation features that are most critical to the stitch quality, and the identified key process-deviation data set is input into the stitching defect correlation analyzer. Combining with the tolerance standard of the actual stitching quality of different materials, a stitching defect detection report is generated. This report can help production personnel understand the stitching quality of different functional areas of the coat in real time, thereby effectively improving the detection accuracy and reducing the defect rate.

[0008] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specific embodiments of this application are specifically given. Brief Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0010] Figure 1 It is a schematic flow chart of an intelligent detection method for stitching defects of a coat in an embodiment;

[0011] Figure 2 It is a schematic flow chart of the construction of a defect feature extraction model for an intelligent detection method for stitching defects of a coat in an embodiment. Detailed Description of the Embodiments

[0012] Embodiments of this application provide an intelligent detection method for overcoat stitching defects, which solves the technical problems of inaccurate defect detection and insufficient real-time performance caused by fluctuations in process parameters in different functional areas during overcoat stitching.

[0013] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts shall fall within the protection scope of this application.

[0014] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0015] Embodiment, as Figure 1 shown, this application provides an intelligent detection method for overcoat stitching defects, and the method includes:

[0016] Obtain the design drawing of the target overcoat, divide the functional areas based on the design drawing of the target overcoat, and match the material information according to the division result to generate a material-functional area mapping diagram.

[0017] In the embodiments of this application, the system terminal first obtains the design drawing of the target overcoat, which details the overall structure of the overcoat and the specific designs of each part. Based on this drawing, the system terminal divides the various functional areas of the overcoat, such as the collar, cuffs, buttons, zippers, pockets, etc., to determine different functional areas. Subsequently, according to the divided functional areas and combined with the specific design information of each part, the system terminal obtains the material information used in these functional areas, such as different materials like cotton, wool, silk, etc. Through this matching process, the system terminal generates a material-functional area mapping diagram. This mapping diagram details each functional area of the overcoat and the corresponding material information. For example, collar area - wool, button area - polyester fiber, etc. This mapping diagram will serve as the basis for subsequent image acquisition and defect detection, ensuring that in different functional areas, the system terminal can accurately match the corresponding material information.

[0018] Based on a high-precision visible light camera, perform stitching image acquisition for each area of the division result to obtain the stitching image data stream of each functional area.

[0019] In one embodiment, after the division of each functional area of the coat is completed, the system terminal uses a high-precision visible light camera to take pictures of each divided functional area one by one, and collect image data during the stitching process area by area. The high-precision visible light camera will focus on different functional areas of the coat in sequence to ensure accurate capture of the stitching details of each area. Through this process, the system terminal obtains the stitching image data stream of each functional area, and these data will reflect the defects and deviations that may occur in different areas during the stitching process, ensure that the stitching conditions of each area are accurately recorded, and provide a basis for subsequent defect detection.

[0020] Furthermore, before the present application provides area-by-area stitching image acquisition of the division result based on the high-precision visible light camera, it includes:

[0021] Extract material features in the material feature library based on the material information of the material-functional area mapping diagram to obtain multiple material features of the target coat; configure the multi-angle adjustment strategy of the high-precision visible light camera based on the multiple material features; perform multi-angle light adaptability optimization of the high-precision visible light camera based on the multi-angle adjustment strategy.

[0022] Preferably, the system terminal extracts the material information of each functional area of the coat from the obtained material-functional area mapping diagram, and uses this material information as an index to input it into the material feature library for matching, and extracts the features of each material, including reflectivity, light absorption ability, texture roughness, transparency, etc., providing a basis for camera configuration to ensure that different material requirements can be met during shooting. Among them, this material feature library is constructed based on big data retrieval and includes various materials and their corresponding characteristics. After obtaining multiple material features of the coat, the system terminal configures a multi-angle adjustment strategy for the high-precision visible light camera based on these features. The core of this strategy is to adjust the shooting angle and parameters of the high-precision visible light camera according to the characteristics of different materials. For example, for materials with high reflectivity (such as silk), the camera is configured to avoid direct strong light reflection and adjusted to a side-angle shooting to reduce reflection interference. For materials with strong light absorption (such as wool or cotton cloth), the camera is configured to increase the exposure time and collect light from multiple angles to ensure that the details of the material texture can be clearly presented. For materials with rough textures (such as coarse cotton cloth), the contrast is increased to enhance the camera's capture of material texture details. The system terminal configures a camera angle adjustment strategy for each functional area according to the material characteristics to ensure that the material characteristics are accurately reflected. Subsequently, according to the configured multi-angle adjustment strategy, the system terminal optimizes the multi-angle light adaptability of the high-precision visible light camera. Specifically, the top view angle and side view angle of the camera are dynamically adjusted according to the multi-angle adjustment strategy, so that it can shoot each material from different angles to adapt to the reflection characteristics and texture details of each material and reduce the deviation caused by single light. The ambient light or the light source built into the camera is automatically adjusted according to the multi-angle adjustment strategy to ensure sufficient light during shooting in each functional area and avoid over-bright or over-dark phenomena. In addition, the focus and exposure parameters of the camera are dynamically adjusted during shooting at different angles to ensure that the material details at each shooting angle can be clearly displayed. Through this optimization process, the camera can automatically adjust the light and angle according to the material characteristics of each functional area of the coat to ensure the acquisition of high-quality image data and provide the best image basis for subsequent defect detection.

[0023] Activate the defect feature extraction model, perform channel matching according to the material-functional area mapping diagram, and input the suture image data stream of each functional area according to the matching result to extract defect characteristics, and obtain the suture deviation stream and suture deviation feature set of different materials.

[0024] In one embodiment, after obtaining the suture image data stream of each functional area, the system terminal activates the defect feature extraction model. The defect feature extraction model contains multiple defect feature extraction channels inside, and each defect feature extraction channel corresponds to a kind of material, which is used to identify and analyze the defect features in the overcoat stitching process. Subsequently, according to the material information of each functional area recorded in the material-functional area mapping diagram, defect feature extraction channel matching is performed, and each defect feature extraction channel is marked with a functional area. After that, the system terminal inputs the suture image data stream of each functional area according to the mark of each channel, and the defect feature extraction channel will extract defect characteristics according to the received suture image data stream, generating suture deviation streams and suture deviation feature sets of different materials, and specifically describing the possible defects and deviation problems in the stitching of each area.

[0025] Further, as Figure 2 shown, this application provides a method for constructing a target three-dimensional model, including:

[0026] Connect to the historical suture database, perform similarity matching in the historical suture database using the suture image data stream of each functional area, and obtain the first root node. The first root node includes the historical suture image data stream, historical suture deviation stream, and historical suture deviation feature set of different functional areas; perform first-level expansion in the historical suture database based on the first root node to obtain first-level expansion leaf nodes; based on the first-level expansion leaf nodes, perform second-level expansion in the historical suture database to obtain second-level expansion leaf nodes; expand layer by layer to the H-level expansion leaf nodes to obtain multi-level expansion leaf nodes, where H≥2; in the direction from the H-level to the first root node, perform reverse aggregation on the multi-level expansion leaf nodes to construct clustering clusters for each functional area; perform data integration on the clustering clusters for each functional area, perform material division according to the integration result to obtain a sample data set; construct a defect feature extraction model based on the sample data set.

[0027] Preferably, the system terminal is connected to the historical stitching database through a reserved port. This historical stitching database stores a large amount of data on historical stitching processes, including historical stitching image data streams, historical stitch deviation streams, and historical stitch deviation feature sets. Among them, the historical stitch deviation stream is a stitching image with defects, and the historical stitch deviation feature set corresponds one-to-one with the historical stitch deviation stream. Subsequently, the system terminal uses the stitching image data streams of each functional area to perform similarity matching in the historical stitching database. Taking the collar as an example, the system terminal calculates multiple similarity degrees between the stitching image data stream of the collar and the historical stitching image data streams in the historical stitching database. Then, these similarity degrees are compared with a similarity threshold, and the similarity degrees greater than or equal to the similarity threshold are selected. The historical stitching image data stream corresponding to the selected similarity degree, as well as the historical stitch deviation stream and the historical stitch deviation feature set corresponding to this historical stitching image data stream, are added to the first root node. After that, the system terminal performs a first-level expansion in the historical stitching database based on the first root node, that is, searches for similar historical records based on the first root node to generate first-level expansion leaf nodes. On this basis, continue to perform a second-level expansion on the first-level expansion leaf nodes to find more similar data and generate second-level expansion leaf nodes. Through such an expansion process, it is expanded layer by layer to the H-level expansion leaf nodes, where H is greater than or equal to 2. Each layer of expansion adds more historical data, making the sample set of the model more abundant. For the stitching image data streams of the remaining functional areas, the system terminal uses the same method as above to determine the first root node for the stitching image data streams of these functional areas, and based on the first root node, performs layer-by-layer expansion to obtain the H-level expansion leaf nodes of each functional area. After expanding to the H level, the system terminal starts to trace back from the H-level expansion leaf nodes to the first root node for reverse aggregation. This process integrates all the data of the expanded leaf nodes together to construct clustering clusters for different functional areas. For these clustering clusters of functional areas, the system terminal performs data integration, that is, aggregates all the data in the clustering clusters together, and then performs material classification on the aggregated result according to the material, differentiates the historical stitching image data streams, historical stitch deviation streams, and historical stitch deviation feature sets corresponding to different materials, and generates a sample data set. Then, the system terminal trains multiple defect feature extraction channels based on multiple material sample data in the sample data set, and each defect feature extraction channel corresponds to one material. Finally, multiple defect feature extraction channels are integrated to construct a defect feature extraction model for subsequent overcoat stitching defect detection to ensure detection accuracy and adaptability.

[0028] Taking wool material as an example, the system terminal divides the wool material sample data in the sample dataset into a training set and a validation set, and trains the constructed initial defect feature extraction channel with the training set. After the training is completed, the validation set is used for verification to evaluate the performance of the channel. Among them, the initial defect feature extraction channel can be constructed based on convolutional neural networks, support vector machines, etc. Taking convolutional neural networks as an example, a structure of an initial defect feature extraction channel is constructed using a convolutional neural network, including an input layer, an output layer, a convolutional layer, a pooling layer, a fully connected layer, etc. Then, the weights of the initial defect feature extraction channel are initialized using random numbers, etc., and the training set is input into the initialized initial defect feature extraction channel for forward propagation, and is sequentially transmitted through the input layer, convolutional layer, pooling layer, fully connected layer, output layer, etc., to calculate a prediction result including a stitch deviation flow and a stitch deviation feature set. Subsequently, the cross-entropy loss function is used to calculate the loss value between the prediction result and the sample data, and the gradient of the loss with respect to the weights of each layer is calculated layer by layer through backpropagation. Then, the gradient descent algorithm is used to optimize the channel parameters and adjust the weights to minimize the value of the loss function. The above process is repeated until the maximum number of iterations is reached. After the training is completed, the validation set is used to test the channel performance, and the accuracy rate of the channel in the classification tasks of the stitch deviation flow and the stitch deviation feature set is evaluated. If the accuracy rate meets the expected expectation, the current initial defect feature extraction channel is output as the final wool material defect feature extraction channel. Otherwise, hyperparameters such as the learning rate and the number of training batches are adjusted to further improve the defect characterization extraction effect of the wool material defect feature extraction channel.

[0029] Further, the present application provides a first-level expansion in the historical suture database based on the first root node to obtain first-level expansion leaf nodes, including:

[0030] Traversing the historical suture database based on the first root node to obtain a first expansion neighborhood; evaluating the similarity distance between the first root node and the first expansion neighborhood to obtain a second expansion neighborhood; randomly sampling based on the second expansion neighborhood to obtain first-level expansion leaf nodes.

[0031] Optionally, after obtaining the first root node, the system terminal traverses the historical suture database according to the characteristics of the first root node and searches for other historical data with similar characteristics to form a first expansion neighborhood. The first expansion neighborhood includes all historical suture data with a high similarity to the first root node. After obtaining the first expansion neighborhood, the system terminal performs a similarity measurement on the first root node and all the data in the first expansion neighborhood. Common similarity measurement methods include Euclidean distance, cosine similarity, dynamic time warping, etc. Taking Euclidean distance as an example, the system terminal calculates the similarity distance between the first root node and all the data in the neighborhood, and then quantifies the similarity between the first root node and the remaining historical suture data using 1 divided by the sum of 1 and the similarity distance. Subsequently, based on the calculated similarity, the historical stitching data in the first extended neighborhood is filtered, and the historical stitching data with a similarity greater than or equal to the similarity threshold to the first root node is selected to form a second extended neighborhood. This step ensures that the extended nodes are all historical stitching data highly relevant to the first root node. Then, the system terminal performs random sampling on the second extended neighborhood according to a preset data volume. The purpose of random sampling is to select representative and diverse historical data from the second extended neighborhood, rather than simply selecting the nearest nodes. Random sampling can prevent the channel from overfitting to a certain type of feature, thereby enhancing the generalization ability of the channel. In addition, uniform sampling, similarity-weighted sampling, etc. can also be used to ensure that different types of data are included in the sampling results. Through random sampling, the system terminal generates first-level extended leaf nodes. These first-level extended leaf nodes represent the historical stitching data related to the first root node and its extended neighborhood and will serve as important data for subsequent expansion or model training.

[0032] Furthermore, the present application provides a method for reversely aggregating the multi-level extended leaf nodes in the direction from the H-level to the first root node to construct a sample data set, including:

[0033] Performing a central tendency analysis on the historical stitching image data stream, historical stitch deviation stream, and historical stitch deviation feature set in the H-level extended leaf nodes to obtain an H-level historical stitching image aggregated data stream, an H-level historical stitch aggregated deviation stream, and an H-level historical stitch aggregated deviation feature set; adding the H-level historical stitching image aggregated data stream, the H-level historical stitch aggregated deviation stream, and the H-level historical stitch aggregated deviation feature set to the H-1 level extended leaf nodes for central tendency analysis, and tracing back level by level until the first-level historical stitching image aggregated data stream, the first-level historical stitch aggregated deviation stream, and the first-level historical stitch aggregated deviation feature set are added to the first similar root node to generate a sample training data set.

[0034] Optionally, after obtaining the H-level extended leaf nodes, the system terminal performs an aggregation operation on the historical stitched image data stream of each H-level extended leaf node to extract its common features, that is, in the way of mean image, the pixels of multiple similar stitched images are averaged to obtain the aggregated H-level historical stitched image data stream. This step helps to remove accidental interference and noise and highlight the main features. Similar to the historical stitched image data stream, the system terminal performs the same operation on the historical suture deviation stream of each H-level extended leaf node to obtain the H-level historical suture aggregated deviation stream. By averaging the features in the historical suture deviation feature set of each H-level extended leaf node, the H-level historical suture aggregated deviation feature set is obtained. Subsequently, the system terminal adds these H-level historical stitched image data streams, H-level historical suture aggregated deviation streams, and H-level historical suture aggregated deviation feature sets representing the entire H-level extended leaf nodes to the H-1 level extended leaf nodes, and then performs the same central tendency analysis on the H-1 level extended leaf nodes. Through step-by-step backtracking, the data of each level (such as H-2 level, H-3 level) extended leaf nodes is merged with the data of the previous level, and the central tendency analysis and aggregation operations are repeated until backtracking to the first-level extended leaf nodes. After the final aggregation operation is completed on the first-level extended leaf nodes, the data has undergone multiple trend analyses and aggregations, forming a highly integrated data set. In the first-level extended leaf nodes, the system terminal finally integrates the highly integrated data set with the historical stitched image data stream, historical suture deviation stream, and historical suture deviation feature set of the first root node. This integration result represents the comprehensive data between different levels of extended leaf nodes, thus obtaining the sample data set. This sample data set retains the core features of historical data at different levels and can be used for channel training to ensure that the channel can capture various possible suture defects and deviation patterns.

[0035] Collect the real-time process parameters of the stitching device, and combine with the suture deviation feature set to perform time-series matching of the homologous functional regions between the real-time process parameters of the stitching device and the suture deviation stream, and generate a process-deviation association data set for different functional regions.

[0036] In one embodiment, the system terminal collects the process parameters of the sewing device in real time. These parameters include stitch speed, fabric feed speed, tension setting value, stitch pitch setting value, etc. These process parameters reflect the specific working conditions of the sewing device when performing sewing operations in different functional areas (such as collars, cuffs, pockets, etc.). Subsequently, these process parameters are combined with the stitch deviation feature set. The stitch deviation feature set contains the stitch quality problems in each functional area, such as stitch deviation, uneven stitches, exposed thread ends, etc. To ensure the effective association between the process parameters and the stitch deviation features, the system terminal performs a homologous functional area time series matching on these data. This means that the system terminal will match the process parameters at each time point with the corresponding stitch deviation at the same time point according to the time line, ensuring that the device state at the same moment corresponds to the sewing result. Through this matching, the system terminal generates a process-deviation association data set, which associates the sewing process parameters of each functional area with the corresponding stitch deviation features. Different functional areas may have different sewing processes and deviation patterns. Such an association data set can help further analyze the sewing quality problems in each area and identify the process parameters that may cause these problems, providing a basis for subsequent defect detection.

[0037] Perform multi-scale fitting on the process-deviation association data sets of the different functional areas to identify the key process-deviation association data sets.

[0038] In one embodiment, after obtaining the process-deviation association data sets of different functional areas, the system terminal performs multi-scale fitting on these data sets. The process of multi-scale fitting is to gradually fit the relationship between the process parameters and the stitch deviation in different process parameter dimensions. Specifically, the system terminal analyzes the influence of the stitch speed, fabric feed speed, tension setting value, and stitch pitch setting value in the process parameters on the stitch deviation for the sewing processes of different functional areas. By performing fitting at different scales, the system terminal can more comprehensively capture the complex relationship between the process parameters and the stitch deviation, and obtain the process parameters that have the greatest impact on the sewing quality. This process helps to eliminate the interference of noise and abnormal data and more accurately identify which process parameters are highly correlated with the stitch deviation. Finally, the system terminal identifies the key process-deviation association data set, which contains the process parameters and sewing features that have a significant impact on the stitch deviation and is used for subsequent quality assessment.

[0039] Furthermore, the present application provides multi-scale fitting of the process-deviation association data sets of the different functional areas to identify the key process-deviation association data sets of different functional areas, including:

[0040] Perform homologous matching on the process-deviation correlation datasets of the different functional regions to obtain a homologous process-deviation set; decompose the process parameters of the homologous process-deviation set to obtain multiple single-parameter deviation control groups; perform dynamic analysis of the correlation degree on the multiple single-parameter deviation control groups based on a preset sliding window to obtain multiple single-parameter correlation degrees; perform single-parameter matching on the process-deviation correlation datasets of the different functional regions based on the multiple single-parameter correlation degrees to obtain a key process-deviation correlation dataset.

[0041] Preferably, the system terminal performs homologous matching on the process-deviation correlation datasets of different functional regions, that is, matches the process parameters and stitch deviation data of the same coat. This means that the system terminal will ensure that all process parameters and stitch deviation characteristics come from the stitching process of the same coat. By matching the data of coats from the same source, a homologous process-deviation set can be obtained, which can more accurately reflect the process and deviation relationship of the coat during the entire stitching process. Subsequently, the process parameters in the homologous process-deviation set are decomposed. The process parameters consist of multiple variables, such as stitch speed, fabric feed speed, tension setting value, stitch length setting value, etc. Taking a coat as an example, the system terminal decomposes the process parameters of this coat separately and correlates them with the stitch deviation characteristics of each functional region of this coat to form multiple single-parameter deviation control groups. Each control group will correspond to a process parameter and the stitch deviation characteristics of each functional region. After that, a preset sliding window is used to intercept each single-parameter deviation control group to obtain multiple single-parameter deviation sequences of each single-parameter deviation control group. Taking the stitch speed deviation sequence as an example, the system terminal performs normalization processing by the maximum-minimum method, that is, calculates the ratio of the difference between the current value and the minimum value of this parameter to the difference between the maximum value and the minimum value of this parameter to obtain a single-parameter normalized deviation sequence. Then, a stitch speed standard sequence is obtained as a reference sequence for subsequent calculations, and this stitch speed standard sequence is set based on historical experience. Then, the absolute difference is calculated for each time point between the stitch speed deviation sequence and the reference sequence, the minimum difference and the maximum difference are statistically obtained, and then the correlation degree at each time point is calculated according to the dynamic analysis formula of the correlation degree. The specific dynamic analysis formula of the correlation degree is as follows: Among them, ε i (k) is the correlation degree at the kth time point, k is the time point, i is the sequence number of the process parameter, Δ min is the minimum difference, Δ max is the maximum difference, ρ is the resolution coefficient, usually taking a value of 0.5, Δ i(k) is the absolute difference between the deviation sequence of the i-th process parameter and the reference sequence at the k-th time point. After calculating the correlation degree at each time point, the system terminal accumulates all the correlation degrees of the pin speed deviation sequence, and then calculates the mean value to obtain the correlation degree of the deviation sequence of the pin speed deviation sequence. Finally, the same calculation is performed on the remaining deviation sequences to obtain multiple correlation degrees of the deviation sequences, and then the correlation degrees of the deviation sequences in the same single-parameter deviation control group are averaged to obtain multiple single-parameter correlation degrees. After obtaining multiple single-parameter correlation degrees, the system terminal matches the multiple single-parameter correlation degrees with the process-deviation correlation data sets in different functional regions, that is, integrates the single-parameter correlation degrees belonging to the same functional region, and then selects the single-parameter correlation degree with the largest correlation degree from them to extract parameters from the process-deviation correlation data set of the functional region, obtaining the key process-deviation correlation data set. Through this operation, the system terminal can identify the process parameters that have the most significant impact on the sewing quality, providing an important basis for subsequent quality assessment.

[0042] Further, before the system provided by the present application performs dynamic analysis of the correlation degree on the multiple single-parameter deviation control groups based on a preset sliding window, it includes:

[0043] Determine the hysteresis effect of the multiple single-parameter deviation control groups to obtain the hysteresis effect sequence of each single-parameter deviation control group; verify the hysteresis effect sequence based on the hysteresis effect threshold to determine the hysteresis time; if the hysteresis time is not 0, perform time series translation on the single-parameter deviation control group corresponding to the hysteresis time.

[0044] Optionally, for each single-parameter deviation control group, the system terminal analyzes whether there is a time-lag effect in the relationship between the process parameter and the suture deviation. The lag effect can be determined by various methods. Commonly used methods include cross-correlation analysis, that is, by calculating the correlation between the process parameter sequence and the suture deviation sequence at different time lags to determine whether there is a significant lag. To accurately evaluate the time-lag effect, the system terminal determines the lag effect of multiple single-parameter deviation control groups through the constructed lag effect evaluation operator, calculates the lag correlation at each time point, and obtains the lag effect sequence of each single-parameter deviation control group. Subsequently, according to the time sequence relationship, the lag correlation and the lag effect threshold of each lag effect sequence are compared. If there is a lag correlation greater than or equal to the lag effect threshold, the system terminal extracts the first lag correlation greater than or equal to the lag effect threshold, and uses the time point corresponding to this lag correlation as the lag time. If not, the lag time is set to 0, indicating that there is no lag effect. After that, based on the non-zero lag time, the system terminal performs time-series adjustment (translation) on the single-parameter deviation control group of the process parameter, that is, moves the time series of the process parameter forward or backward by the corresponding lag time, so that the process parameter and the suture deviation can synchronously reflect the actual situation. The single-parameter deviation control groups after time-series translation will be more matched, making the subsequent correlation analysis more accurate.

[0045] Furthermore, the present application provides a method for determining the lag effect of the multiple single-parameter deviation control groups to obtain the lag effect sequence of each single-parameter deviation control group, including:

[0046] Constructing a lag effect evaluation operator, the lag effect evaluation operator includes a lag effect evaluation function, and the lag effect evaluation function is specifically as follows:

[0047] Optionally, the lag effect evaluation operator includes an input layer, an output layer, and a calculation layer. The calculation layer is built-in with a lag effect evaluation function, and this lag effect evaluation function is used to analyze the lag effect between process parameters (such as stitch speed, tension, etc.) and suture deviation. The lag effect evaluation function is specifically as follows: Among them, CCF(k) is the result of the hysteresis effect, representing the correlation between the process parameter sequence and the suture deviation sequence at the hysteresis time k. k is the hysteresis time, representing the hysteresis time step between time series. n is the length of the time series, that is, the number of data points in the single-parameter deviation control group. X(t) is the process parameter sequence, representing the process parameter value at the time point t, including but not limited to the stitch speed, fabric feed speed, tension setting value, and stitch length setting value. Y(t + k) is the suture deviation at the time t + k in the suture deviation sequence, representing the suture deviation value at the time point t + k. X is the average value of the process parameter sequence, and Y is the average value of the suture deviation sequence.

[0048] Input the key process-deviation association dataset into the stitching defect association analyzer for quality assessment, and combine the stitching quality tolerance to generate a stitching defect detection report.

[0049] In one embodiment, the system terminal inputs multiple key process-deviation association datasets into the stitching defect association analyzer. The stitching defect association analyzer also includes multiple stitching defect association analysis channels, and each stitching defect association analysis channel corresponds to a kind of material. The system terminal matches the stitching defect association analysis channels through the material identifier in the key process-deviation association dataset and inputs the corresponding key process-deviation association dataset. Each stitching defect association analysis channel will conduct a comprehensive quality assessment on the input data to determine the predicted stitching quality of the functional area corresponding to each key process-deviation association dataset. Among them, the construction method of the stitching defect association analyzer is the same as that of the aforementioned defect feature extraction model. Subsequently, in combination with the stitching quality tolerance, that is, the set acceptable defect range, the obtained predicted stitching quality is judged. If the predicted stitching quality exceeds the preset tolerance range, the system terminal marks that there is a quality problem in the functional area and generates a stitching defect detection report, which includes the stitching quality situation, specific defect description, and the process parameter with the greatest impact degree of the functional area with quality problems. This report can help understand the problems in the stitching process, optimize the process flow, and improve the overall quality.

[0050] In summary, the embodiments of the present application have at least the following technical effects:

[0051] In the embodiments of the present application, by obtaining the design drawings of the target coat, dividing the functional areas, and matching the material information to generate a material-functional area mapping diagram. Then, using a high-precision visible light camera to collect stitching image data for each functional area one by one, activating the defect feature extraction model, and extracting the stitch deviation flow and feature set for different material areas through channel matching. At the same time, collecting the real-time process parameters of the stitching device, and generating a process-deviation correlation data set through temporal matching of the homologous functional areas of the stitch deviation flow. Subsequently, identifying the key process-deviation correlation data set through multi-scale fitting, and inputting it into the stitching defect correlation analyzer, and finally generating a stitching defect detection report in combination with the stitching quality tolerance. In addition, the defect feature extraction model is extended by connecting to the historical stitching database, constructing similar nodes based on historical data and performing multi-level aggregation to generate a sample data set. This sample data set is used to construct a defect feature extraction model with prediction and detection capabilities. For further precise analysis, multi-scale fitting involves decomposing the process parameters of the homologous matching data, analyzing the sliding window correlation degree, and performing temporal adjustment for the hysteresis effect. The evaluation of the hysteresis effect is completed by constructing an evaluation operator and applying relevant evaluation functions, and finally realizing the intelligent detection of potential defects in the stitching process. These technical effects jointly solve the technical problems of inaccurate defect detection and insufficient real-time performance caused by fluctuations in process parameters in different functional areas during the coat stitching process, and achieve the effect of improving the detection accuracy and adaptability by combining synchronous analysis and multi-scale fitting of process parameters and stitch deviation data.

[0052] It should be noted that the above sequence of the embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above description of specific embodiments of this specification is given. The processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0053] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0054] This specification and the drawings are only exemplary descriptions of the present application and are considered to have covered any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. An intelligent detection method for coat stitching defects, characterized in that: The method comprises: Obtaining a target coat design drawing, dividing the functional areas based on the target coat design drawing, matching material information according to the division result, and generating a material-functional area mapping diagram; Based on a high-precision visible light camera, stitching images are collected area by area on the division results to obtain stitching image data streams of each functional area; Activate the defect feature extraction model, perform channel matching according to the material-functional area mapping diagram, input the stitched image data stream of each functional area according to the matching result to perform defect characterization extraction, and obtain the suture deviation stream and suture deviation feature set of different materials; Collecting real-time process parameters of suturing equipment, combining the suture deviation feature set, performing time series matching of homologous functional areas on the real-time process parameters of suturing equipment and the suture deviation flow, and generating process-deviation association data sets of different functional areas; Performing multi-scale fitting on the process-deviation correlation data sets of the different functional areas to identify key process-deviation correlation data sets; The key process-deviation association data set is input into the stitching defect association analyzer for quality assessment, and combined with the stitching quality tolerance, a stitching defect detection report is generated.

2. The intelligent detection method for coat stitching defects according to claim 1, characterized in that: Construct a defect feature extraction model, the method includes: Connecting to a historical stitching database, using the stitched image data streams of each functional area to perform similarity matching in the historical stitching database, to obtain a first root node, wherein the first root node includes historical stitched image data streams of different functional areas, historical stitch deviation streams, and historical stitch deviation feature sets; Performing a first-level expansion in the historical stitching database based on the first root node to obtain a first-level expanded leaf node; Based on the first-level expanded leaf node, performing a second-level expansion in the historical stitching database to obtain a second-level expanded leaf node; By expanding layer by layer to H-level expansion leaf nodes, a multi-level expansion leaf node is obtained, where H ≥ 2; Taking level H to the first root node as the direction, reversely aggregate the multi-level extended leaf nodes to construct clusters of each functional area; Integrate the data of the functional area clusters, divide the materials according to the integration results, and obtain a sample data set; A defect feature extraction model is constructed based on the sample data set.

3. The intelligent detection method for coat stitching defects according to claim 2, characterized in that: Based on the first root node, a first-level expansion is performed in the history stitching database to obtain a first-level expansion leaf node, the method comprising: Traversing the historical stitching database based on the first root node to obtain a first extended neighborhood; Performing similarity distance evaluation on the first root node and the first extended neighborhood to obtain a second extended neighborhood; Random sampling is performed based on the second extended neighborhood to obtain a first-level extended leaf node.

4. The intelligent detection method for coat stitching defects according to claim 2, characterized in that: Taking level H to the first root node as the direction, the multi-level extended leaf nodes are reversely aggregated to construct a sample data set, and the method includes: Performing a central tendency analysis on the historical stitched image data stream, the historical suture deviation stream, and the historical suture deviation feature set in the H-level extended leaf nodes to obtain an H-level historical stitched image aggregate data stream, an H-level historical suture aggregate deviation stream, and an H-level historical suture aggregate deviation feature set; The H-level historical stitched image aggregate data stream, the H-level historical seam aggregate deviation stream and the H-level historical seam aggregate deviation feature set are added to the H-1-level extended leaf node for central tendency analysis, and then traced back level by level until the first-level historical stitched image aggregate data stream, the first-level historical seam aggregate deviation stream and the first-level historical seam aggregate deviation feature set are added to the first similarity root node to generate a sample training data set.

5. The intelligent detection method for coat stitching defects according to claim 1, characterized in that: Performing multi-scale fitting on the process-deviation correlation data sets of the different functional areas to identify key process-deviation correlation data sets of the different functional areas, the method comprising: Performing homology matching on the process-deviation association data sets of the different functional areas to obtain homology process-deviation sets; Decomposing the homologous process-deviation set by process parameters to obtain a plurality of single parameter deviation control groups; Performing a dynamic correlation analysis on the multiple single parameter deviation control groups based on a preset sliding window to obtain multiple single parameter correlations; Single parameter matching is performed on the process-deviation correlation data sets of the different functional areas based on the multiple single parameter correlation degrees to obtain a key process-deviation correlation data set.

6. The intelligent detection method for coat stitching defects according to claim 5, characterized in that: Before dynamically analyzing the correlation of the multiple single parameter deviation control groups based on a preset sliding window, the method includes: Performing hysteresis effect determination on the multiple single parameter deviation control groups to obtain a hysteresis effect sequence of each single parameter deviation control group; Verifying the hysteresis effect sequence based on a hysteresis effect threshold to determine a hysteresis time; If the lag time is not 0, the single parameter deviation control group corresponding to the lag time is time-shifted.

7. The intelligent detection method for coat stitching defects according to claim 6, characterized in that: Determining the hysteresis effect of the multiple single parameter deviation control groups to obtain the hysteresis effect sequence of each single parameter deviation control group includes: A hysteresis effect evaluation operator is constructed, wherein the hysteresis effect evaluation operator includes a hysteresis effect evaluation function, and the hysteresis effect evaluation function is specifically as follows: Wherein, CCF(k) is the hysteresis effect result, k is the hysteresis time, n is the length of the time series, X(t) is the process parameter sequence, including but not limited to stitch speed, fabric feed speed, tension setting value, stitch length setting value, Y(t+k) is the stitch deviation at time t+k in the stitch deviation sequence, is the average value of the process parameter sequence, is the average value of the suture deviation series.

8. The intelligent detection method for coat stitching defects according to claim 6, characterized in that: Based on the high-precision visible light camera, the image of the segmentation result is stitched region by region, and before the image of the target coat is stitched by the high-precision visible light camera, the method includes: Extracting material features from a material feature library based on the material information of the material-functional area mapping diagram to obtain a plurality of material features of the target coat; Configuring a multi-angle adjustment strategy for the high-precision visible light camera based on the multiple material features; The multi-angle light adaptability of the high-precision visible light camera is optimized based on the multi-angle adjustment strategy.

Images