A machine vision-based method and system for quality inspection of gravity blankets

Abstract

This invention provides a machine vision-based method and system for quality inspection of gravity blankets, relating to the field of machine vision. The method includes: matching the input standard values ​​of multi-layer detection indicators of the gravity blanket according to its type to generate multi-layer detection indicator standard values; obtaining pre-application scenario parameters, including user-customized parameters and predetermined usage parameters, and adjusting the multi-layer detection indicator standard values ​​to generate multi-layer detection indicator benchmark values; traversing the multi-layer detection indicators of the gravity blanket to generate multi-layer indicator weight distribution results; acquiring images of the gravity blanket to be inspected to generate multi-layer image acquisition results; and performing quality inspection on the gravity blanket to be inspected based on the multi-layer indicator weight distribution results and the multi-layer detection indicator benchmark values, generating quality inspection results. This invention solves the technical problem in existing technologies where the uniformly set quality inspection standards for gravity blankets result in insufficient compatibility with application scenarios.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, specifically to a machine vision-based method and system for quality inspection of gravity blankets. Background Technology

[0002] Gravity blankets utilize the principle of contact pressure to increase the secretion levels of serotonin and melatonin in the human body, lower blood pressure, and improve sleep quality. They have become quite popular in recent years. Like many products, quality testing is an important guarantee for the performance of gravity blankets. However, due to the multi-layered structure of gravity blankets and their complex composition, the quality inspection process is complicated, making the quality testing process particularly important.

[0003] Currently, quality inspection of gravity blankets is mostly semi-automated, meaning that quality inspection is carried out according to set rules. Different factories need to set different quality inspection rules for gravity blankets, and the standard values ​​for testing are set uniformly by the manufacturers, which has a low degree of integration with the application scenarios.

[0004] In existing technologies, the uniform quality inspection standards for gravity blankets lead to technical problems such as insufficient compatibility with application scenarios. Summary of the Invention

[0005] This application provides a machine vision-based method and system for quality inspection of gravity blankets, which solves the technical problem in the prior art where the uniform quality inspection standards for gravity blankets are not well-suited to application scenarios.

[0006] In view of the above problems, this application provides a machine vision-based method and system for detecting the quality of gravity blankets.

[0007] In a first aspect, this application provides a machine vision-based method for quality inspection of gravity blankets. The method employs a machine vision-based gravity blanket quality inspection system, which is communicatively connected to an image acquisition device. The method includes: matching multi-layer detection indicators of the gravity blanket according to its type; inputting the multi-layer detection indicators into a standard value matching table to generate standard values ​​for the multi-layer detection indicators; obtaining pre-application scenario parameters, including user-customized parameters and predetermined usage parameters; adjusting the standard values ​​of the multi-layer detection indicators according to the user-customized parameters and the predetermined usage parameters to generate benchmark values ​​for the multi-layer detection indicators; traversing the multi-layer detection indicators of the gravity blanket and performing weight distribution to generate a multi-layer indicator weight distribution result; acquiring images of the gravity blanket to be inspected using the image acquisition device to generate multi-layer image acquisition results; and performing quality inspection on the gravity blanket to be inspected based on the multi-layer indicator weight distribution results and the multi-layer detection indicator benchmark values, thereby generating a quality inspection result.

[0008] On the other hand, this application provides a machine vision-based gravity blanket quality inspection system, wherein the system is communicatively connected to an image acquisition device, and the system includes: a detection index matching module, used to match multi-layer detection indexes of the gravity blanket according to the type of the gravity blanket; an index standard value generation module, used to input the multi-layer detection indexes of the gravity blanket into a standard value matching table to generate multi-layer detection index standard values; a pre-application scenario parameter matching module, used to obtain pre-application scenario parameters, wherein the pre-application scenario parameters include user-customized parameters and predetermined purpose parameters; an index benchmark value generation module, used to adjust the multi-layer detection index standard values ​​according to the user-customized parameters and the predetermined purpose parameters to generate multi-layer detection index benchmark values; an index weight distribution module, used to traverse the multi-layer detection indexes of the gravity blanket to perform weight distribution and generate multi-layer index weight distribution results; an image acquisition module, used to acquire images of the gravity blanket to be inspected through the image acquisition device and generate multi-layer image acquisition results; and a quality inspection module, used to perform quality inspection on the gravity blanket to be inspected based on the multi-layer index weight distribution results and the multi-layer detection index benchmark values, based on the multi-layer image acquisition results, and generate quality inspection results.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] This technical solution employs a method that matches multi-layered detection indicators based on the type of gravity blanket; assigns standard values ​​to each multi-layered detection indicator using a standard value matching table; adjusts the standard values ​​of the detection indicators by loading user-customized parameters and pre-defined usage parameters, resulting in multi-layered detection indicator benchmark values ​​that are highly compatible with the application scenario; distributes the detection indicators by weight; and then retrieves the image acquisition device to perform quality inspection based on the multi-layered detection indicator benchmark values ​​and weight distribution results. By adjusting the standard values ​​uniformly set by the manufacturer using user-customized parameters and pre-defined usage parameters, and then performing quality inspection based on machine vision, the technical effectiveness of the quality inspection and its integration with the application scenario is improved.

[0011] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0012] Figure 1 This application provides a schematic flowchart of a gravity blanket quality inspection method based on machine vision.

[0013] Figure 2 This application provides a schematic diagram of the quality inspection result flow in a machine vision-based gravity blanket quality inspection method according to an embodiment of the present application.

[0014] Figure 3 This application provides a schematic diagram of a gravity blanket quality inspection system based on machine vision.

[0015] Figure reference numerals: Image acquisition device 001, detection index matching module 11, index standard value generation module 12, pre-application scenario parameter matching module 13, index benchmark value generation module 14, index weight distribution module 15, image acquisition module 16, quality detection module 17. Detailed Implementation

[0016] The overall concept of the technical solution provided in this application is as follows:

[0017] This application provides a machine vision-based method and system for quality inspection of gravity blankets. The method employs a technique that matches multiple layers of inspection indicators according to the type of gravity blanket; assigns standard values ​​to each layer of inspection indicators using a standard value matching table; adjusts the standard values ​​of the inspection indicators by loading user-customized parameters and pre-defined usage parameters, resulting in multi-layered inspection indicator benchmark values ​​with high relevance to the application scenario; distributes the inspection indicators by weight; and then retrieves the image acquisition device to perform quality inspection based on the multi-layered inspection indicator benchmark values ​​and weight distribution results. By adjusting the standard values ​​uniformly set by the manufacturer using user-customized parameters and pre-defined usage parameters, and thus performing quality inspection based on machine vision, the technical effectiveness of the quality inspection and its relevance to the application scenario is improved.

[0018] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0019] Example 1

[0020] like Figure 1 As shown in the figure, this application provides a machine vision-based method for detecting the quality of a gravity blanket. The method utilizes a machine vision-based gravity blanket quality detection system, which is communicatively connected to an image acquisition device. The method includes the following steps:

[0021] S100: Match the multi-layer test index of gravity blanket according to the type of gravity blanket;

[0022] Specifically, the type of gravity blanket refers to the information that characterizes the model of the gravity blanket. Different models of gravity blankets correspond to different composition structures. Gravity blankets are composed of multiple layers. For example, the structure is as follows: four-layer structure: surface fabric - glass beads - filling cotton - surface fabric; five-layer structure: surface fabric - filling cotton - glass beads - filling cotton - surface fabric; six-layer structure: surface fabric - lining / inner fabric - glass beads - filling cotton - lining / inner fabric - surface fabric; seven-layer structure: surface fabric - lining / inner fabric - filling cotton - glass beads - filling cotton - lining / inner fabric - surface fabric. As can be seen from the above structural examples, different gravity blankets are usually composed of multiple different layers, so each layer has different testing indicators. The multi-layer testing indicators of gravity blankets refer to the testing indicators corresponding to each of the different layers, including at least the material type, material content, and material distribution characteristics of each layer. Among them, the material distribution characteristics are exemplified by indicators such as: the stitch distance of the fabric, the pattern shape of the fabric, and the thickness of the fabric; the thickness and pattern of the lining / inner fabric; the density and size of the glass beads; and the weight and thickness of the filling cotton.

[0023] Furthermore, this includes overall inspection indicators after assembly, such as overall weight and dimensional characteristics. These indicators are easy to inspect and will not be elaborated on here. Before the gravity blanket is assembled into multiple layers, quality inspection is required. After assembly, the multiple layers are sewn together to form the gravity blanket. The inspection of the gravity blanket is then carried out along with the manufacturing process. Various indicators can be directly or indirectly detected using machine vision. Direct inspection indicators include, for example, the stitch distance, pattern shape, thickness, and material type of the fabric—indicators that can be directly observed and determined. Indirect indicators include, for example, weight and material content, which can be indirectly determined based on the material's density and dimensional characteristics. Therefore, machine vision is used for quality inspection. By using machine vision along the process flow for quality inspection, traditional sampling inspection is replaced, improving inspection coverage and combining the inspection process with the production process, thus improving inspection efficiency.

[0024] S200: Input the multi-layer test index of the gravity blanket into the standard value matching table to generate the standard value of the multi-layer test index;

[0025] Specifically, the standard value matching table refers to a cloud database that uses big data to construct quality inspection index values ​​for each layer of traditional gravity blankets. Preferably, the standard value matching table can be based on blockchain, with multiple gravity blanket manufacturers located on blockchain nodes providing specific characteristic values ​​of various component layers' inspection indicators. The index values ​​are preferably determined by collecting the values ​​with the highest repetition frequency. If there are no repetition values, the average value is taken, thus obtaining the standard value matching table.

[0026] The multi-layer test index standard value refers to the standard value determined by inputting the multi-layer test index of the gravity blanket into the standard value matching table. It is the minimum requirement index value that production needs to meet. The standard value matching table built on blockchain breaks down data silos and improves the objectivity of setting quality test index values.

[0027] S300: Obtain pre-application scenario parameters, wherein the pre-application scenario parameters include user-customized parameters and predetermined purpose parameters;

[0028] Specifically, pre-application scenario parameters refer to scenario data that characterizes the application scenario and are used to adjust the standard values ​​of multi-layer testing indicators. Pre-application scenario parameters include user-customized parameters, which characterize user-customized indicator parameters, such as customized thickness, weight, customized added materials, added content, customized patterns, etc.; and predetermined usage parameters, which refer to the scenarios in which the weighted blanket will be applied, such as in the bedroom, on the sofa, during lunch break, or outdoor camping, and the age, gender, weight, etc. of the applicable population. By using user-customized parameters and predetermined usage parameters, the standard values ​​of multi-layer testing indicators can be adjusted to obtain benchmark values ​​with a high degree of fit with the application scenario, providing a reference benchmark for subsequent quality testing and improving the scenario-based nature of quality testing.

[0029] S400: Adjust the standard value of the multi-layer detection index according to the user-customized parameters and the predetermined purpose parameters to generate the multi-layer detection index benchmark value;

[0030] Furthermore, the step S400, which involves adjusting the standard values ​​of the multi-layer detection indexes based on the user-customized parameters and the predetermined usage parameters to generate multi-layer detection index benchmark values, includes the following steps:

[0031] S410: Based on the user-customized parameters, obtain the added material type parameters and added material content parameters;

[0032] S420: Obtain the user type parameter and the predetermined location parameter according to the predetermined purpose parameter;

[0033] S430: Input the added material type parameter and the added material content parameter into the adjustment parameter matching model to generate the first adjustment parameter;

[0034] S440: Input the user type parameter and the predetermined location parameter into the adjustment parameter matching model to generate the second adjustment parameter;

[0035] S450: Adjust the standard value of the multi-layer detection index according to the first adjustment parameter and the second adjustment parameter to generate the benchmark value of the multi-layer detection index.

[0036] Furthermore, step S430 includes the following steps:

[0037] S431: Construct the first half-model training data, wherein the first half-model training data includes a dataset with added material type records, a dataset with added material content records, and a dataset with first adjustment parameter identifiers;

[0038] S432: Construct the second half-model training data, wherein the second half-model training data includes a user type record dataset, a predetermined location record dataset, and a second adjustment parameter identifier dataset;

[0039] S433: Train a first parameter-adjusted matching half-model based on the first half-model training data;

[0040] S434: Train a second parameter-adjusted matching half-model based on the training data of the second half-model;

[0041] S435: Merge the first adjustment parameter matching semi-model and the second adjustment parameter matching semi-model to generate the adjustment parameter matching model.

[0042] Furthermore, based on the training data of the first half-model, training a first parameter-adjusted matching half-model, step S433 includes the following steps:

[0043] S433-1: Using the added material type record dataset and the added material content record dataset as input data, and the first adjustment parameter identifier dataset as output identifier data, construct the first sub-model;

[0044] S433-2: Extract the training data of the first half-model whose output accuracy of the first sub-model does not meet the preset accuracy, and add it to the first training data;

[0045] S433-3: Train the second sub-model based on the first training data;

[0046] S433-4: Merge the first sub-model and the second sub-model to generate the first adjusted parameter matching half-model.

[0047] Furthermore, the step S434 of training a second parameter-adjusted matching half-model based on the second half-model training data includes the following steps:

[0048] S434-1: Using the user type record dataset and the predetermined location record dataset as input data, and the second adjustment parameter identifier dataset as output identifier data, construct a third sub-model;

[0049] S434-2: Extract the training data of the second half-model whose output accuracy of the third sub-model does not meet the preset accuracy, and add it to the second training data;

[0050] S434-3: Train the fourth sub-model based on the second training data;

[0051] S434-4: Merge the third sub-model and the fourth sub-model to generate the second adjustment parameter matching semi-model.

[0052] Specifically, the benchmark value of multi-layer detection index refers to the result determined after adjusting the standard value of multi-layer detection index through user-customized parameters and predetermined application parameters. However, the process of adjusting the user-customized parameters and predetermined application parameters has a strong non-linear relationship. Therefore, the preferred adjustment method is to use an adjustment parameter matching model built based on artificial intelligence.

[0053] The user-customized parameters in this application embodiment include: adding material type parameters and adding material content parameters; the intended use parameters include: user type parameters (preferably: user's age, gender, and weight parameters) and intended location parameters (preferably using intended location, such as: in the bedroom, on the sofa, during a nap, or while camping outdoors).

[0054] First, the added material type parameter and added material content parameter are input into the adjustment parameter matching model to obtain the first adjustment parameter, which represents the direction and degree of adjustment of the multi-layer testing index standard value determined based on the user-customized parameters. Second, the user type parameter and predetermined location parameter are input into the adjustment parameter matching model to obtain the second adjustment parameter, which represents the direction and degree of adjustment of the multi-layer testing index standard value determined based on the predetermined application parameter. Finally, the multi-layer testing index standard values ​​are adjusted sequentially according to the first and second adjustment parameters to obtain the multi-layer testing index baseline value.

[0055] Furthermore, the parameter matching model construction process was adjusted as follows:

[0056] The parameter matching model is an intelligent model built on random decision forest, which includes a first parameter matching semi-model for processing user-customized parameters and a second parameter matching semi-model for processing parameters for a predetermined purpose.

[0057] Construction of the first adjustment parameter matching semi-model: The training data for the first semi-model refers to the training data used to train the first adjustment parameter matching semi-model. The preferred acquisition method is to obtain the data from multiple gravity blanket manufacturers based on the aforementioned blockchain, including datasets with added material type records, datasets with added material content records, and datasets with first adjustment parameter identifiers. The preferred training process is as follows:

[0058] The training data for the first half-model is divided into k parts, and the data is extracted k times with replacement to obtain the first set of training data. This process is repeated M times to obtain M sets of training data. Further, a decision tree is constructed using each set of data, with M sets of data corresponding to M decision trees. When constructing any decision tree, the datasets containing added material type records and added material content records are used as input data, and the dataset containing the first adjustment parameter identifier is used as the output identifier data. Supervised training is performed, and convergence is considered achieved when the output stabilizes. The M decision trees are then merged to obtain the first sub-model. To ensure the generalization ability of the model's output, a certain amount of error is allowed. Another model is used to fit the model to address this error, with a preset accuracy rate representing the highest accuracy rate of the selected training data. The training data of the first half-model whose output accuracy is lower than the preset accuracy rate is denoted as the first training data. Using the first training data, the datasets containing added material type records and added material content records are used as input data, and the dataset containing the first adjustment parameter identifier is used as the output identifier data. Supervised training is then performed, and a second decision tree is constructed based on the decision tree framework. Finally, the first and second sub-models are merged, meaning the optimal output is the average of the two, generating the first adjusted parameter matched semi-model. The random forest model construction process ensures the generalization ability of the model's output, while the setting of the second sub-model improves the accuracy of the first adjusted parameter matched semi-model's output.

[0059] Construction of the second adjustment parameter matching semi-model: The training data for the second semi-model refers to the training data used to train the first adjustment parameter matching semi-model. The preferred acquisition method is to obtain the data from multiple gravity blanket manufacturers using the aforementioned blockchain, including user type parameters, predetermined position parameters, a first adjustment parameter identifier dataset, and a second adjustment parameter identifier dataset. The preferred training process is as follows:

[0060] The training data for the second half-model is divided into k parts, and the first set of training data is obtained by sampling k times with replacement. This process is repeated M times to obtain M sets of training data. Further, a decision tree is constructed using each set of data, with M sets of data corresponding to M decision trees. When constructing any decision tree, the user type record dataset and the predetermined location record dataset are used as input data, and the second adjustment parameter identifier dataset is used as output identifier data. Supervised training is performed, and convergence is considered achieved when the output stabilizes. The M decision trees are then merged to obtain the third sub-model. To ensure the generalization ability of the model's output, a certain amount of error is allowed. Another model is used to fit the model to address the error, with a preset accuracy rate representing the highest accuracy rate of the selected training data. The training data for the second half-model, where the output accuracy of the third sub-model is lower than the preset accuracy rate, is denoted as the second training data. Using the second training data, the user type record dataset and the predetermined location record dataset are used as input data, and the second adjustment parameter identifier dataset is used as output identifier data. Supervised training is performed, and a decision tree is constructed based on the decision tree framework, denoted as the fourth sub-model. Finally, the third and fourth sub-models are merged, meaning the optimal output is the average of the two, generating the second adjusted parameter matching semi-model. The random forest model construction process ensures the model's output generalization ability, while the fourth sub-model improves the accuracy of the second adjusted parameter matching semi-model's output. Finally, the first and second adjusted parameter matching semi-models are merged to obtain the adjusted parameter matching model, which is then set to a pending response state, awaiting subsequent calls.

[0061] S500: Traverse the multi-layer detection indicators of the gravity blanket and perform weight distribution to generate multi-layer indicator weight distribution results;

[0062] Furthermore, the step of traversing the multiple layers of detection indicators of the gravity blanket and performing weight distribution to generate multi-layer indicator weight distribution results includes the following steps:

[0063] S510: Obtain the first importance rating participant, the second importance rating participant, and so on up to the Nth importance rating participant, wherein the information exchange between any two participants is isolated.

[0064] S520: Input the multi-layer detection index of the gravity blanket into the first importance scoring participant, the second importance scoring participant, and so on up to the Nth importance scoring participant, and obtain the scoring results of the first participant, the second participant, and so on up to the Nth participant;

[0065] S530: Obtain the formula for indicator weight distribution:

[0066]

[0067] Among them, w iLet x be the weight distribution result of the i-th indicator. ik For the score of the k-th parameter of the i-th indicator, x j The score for the j-th indicator on the k-th parameter side;

[0068] S540: Input the scoring results of the first participant, the scoring results of the second participant, up to the scoring results of the Nth participant into the indicator weight distribution formula to generate the multi-level indicator weight distribution results.

[0069] Specifically, since the various indicators of the gravity blanket have different priorities, the overall evaluation of the gravity blanket needs to distribute the importance of each indicator according to its weight. The preferred indicator process for any layer is as follows:

[0070] The first-importance scoring participant, the second-importance scoring participant, and so on up to the Nth-importance scoring participant refer to multiple participants retrieved based on the aforementioned blockchain. Each participant represents a gravity blanket manufacturing enterprise. During weight distribution, the participants are in a state of information isolation. By inputting multi-layered detection indicators of the gravity blanket into the first-importance scoring participant, the second-importance scoring participant, and so on up to the Nth-importance scoring participant, the scoring results of the first participant, the second participant, and so on up to the Nth participant are obtained. A higher score indicates a more important corresponding indicator. Furthermore, the indicator weight distribution formula is retrieved:

[0071]

[0072] The sum of scores representing a certain indicator. The sum of all indicator scores is represented by the input of the first participant's score, the second participant's score, and so on up to the Nth participant's score. The corresponding level of indicator weight distribution results are obtained by traversing all gravity blanket layers. The multi-level indicator weight distribution results are then obtained and set to a pending response state, waiting for subsequent calls.

[0073] S600: The image acquisition device acquires images of the gravity blanket to be tested and generates multi-layer image acquisition results.

[0074] S700: Based on the multi-layer index weight distribution results and the multi-layer detection index benchmark values, perform quality detection on the gravity blanket to be detected based on the multi-layer image acquisition results, and generate quality detection results.

[0075] Furthermore, such as Figure 2 As shown, step S700 involves performing quality detection on the gravity blanket to be detected based on the multi-layer index weight distribution results and the multi-layer detection index benchmark values, and generating a quality detection result. The step S700 includes the following steps:

[0076] S710: Based on the multi-layer detection index of the gravity blanket, perform feature analysis on the multi-layer image acquisition results to generate multi-layer detection index feature values;

[0077] S720: Determine whether the feature value of the multi-layer detection index meets the benchmark value of the multi-layer detection index;

[0078] S730: If not met, generate a non-conforming indicator and add it to the quality inspection results;

[0079] S740: If satisfied, generate a comprehensive feature value of the gravity blanket based on the multi-layer index weight distribution result and the feature value of the multi-layer detection index;

[0080] S750: Determine whether the comprehensive characteristic value of the gravity blanket meets the preset characteristic value;

[0081] S760: If the conditions are met, generate qualified identification information to identify the gravity blanket to be tested.

[0082] Specifically, the multi-layer image acquisition result refers to the image acquisition result of any layer after the production of the weighted blanket to be inspected is completed. The image acquisition device is preferably a movable industrial camera, such as a CCD industrial camera. Based on the detection indicators of the corresponding layer, the corresponding feature values ​​are extracted from the image acquisition results, such as the pin distance, glass bead size, thickness and other feature values, which are denoted as multi-layer detection indicator feature values.

[0083] The process involves determining whether the characteristic values ​​of multi-layered testing indicators fall within the range defined by the benchmark values. If not, unqualified indicators and their differences are generated and added to the quality inspection results for further adjustment. If they do fall within the range, the blanket is considered qualified and an overall evaluation can be performed. This involves weighting and summing the corresponding layer's indicator weight distribution results and the corresponding layer's test indicator characteristic values, and recording the result as the comprehensive characteristic value of the gravity blanket. The preset characteristic value refers to the ideal comprehensive characteristic value of the gravity blanket. The process also involves determining whether the comprehensive characteristic value of the gravity blanket falls within the range defined by the preset characteristic value. If it does, the gravity blanket is considered a qualified, excellent-quality product; otherwise, it is considered a qualified, good-quality product. Based on the corresponding comparison results, qualified identification information for good-quality products or excellent-quality products is generated to identify the gravity blanket to be tested and group it. This completes the quality inspection of the corresponding layer of the gravity blanket to be tested.

[0084] In summary, the machine vision-based gravity blanket quality inspection method and system provided in this application have the following technical effects:

[0085] 1. This technical solution employs a method that matches multi-layered detection indicators based on the type of gravity blanket; then applies a standard value matching table to assign standard values ​​to each multi-layered detection indicator; further adjusts the standard values ​​of the detection indicators by loading user-customized parameters and pre-defined usage parameters, resulting in multi-layered detection indicator benchmark values ​​with high integration with the application scenario; then distributes the detection indicators by weight; and finally retrieves the image acquisition device to perform quality inspection based on the multi-layered detection indicator benchmark values ​​and weight distribution results. By adjusting the standard values ​​uniformly set by the manufacturer through user-customized parameters and pre-defined usage parameters, and then performing quality inspection based on machine vision, the technical effect of improving the integration of quality inspection with the application scenario is enhanced.

[0086] Example 2

[0087] Based on the same inventive concept as the machine vision-based gravity blanket quality detection method in the foregoing embodiments, such as Figure 3 As shown in the figure, this application provides a machine vision-based gravity blanket quality inspection system, wherein the system and the image acquisition device 001 are communicatively connected, and the system includes:

[0088] The detection index matching module 11 is used to match the multi-layer detection index of the gravity blanket according to the type of gravity blanket;

[0089] The index standard value generation module 12 is used to input the multi-layer detection index of the gravity blanket into the standard value matching table and generate multi-layer detection index standard values.

[0090] The pre-application scenario parameter matching module 13 is used to obtain pre-application scenario parameters, wherein the pre-application scenario parameters include user-customized parameters and predetermined purpose parameters;

[0091] The indicator benchmark value generation module 14 is used to adjust the multi-layer detection indicator standard value according to the user-customized parameters and the predetermined purpose parameters, and generate multi-layer detection indicator benchmark values.

[0092] The indicator weight distribution module 15 is used to traverse the multi-layer detection indicators of the gravity blanket to perform weight distribution and generate multi-layer indicator weight distribution results.

[0093] Image acquisition module 16 is used to acquire images of the gravity blanket to be tested through image acquisition device 001 and generate multi-layer image acquisition results;

[0094] The quality inspection module 17 is used to perform quality inspection on the gravity blanket to be inspected based on the multi-layer index weight distribution results and the multi-layer inspection index benchmark values, and generate quality inspection results.

[0095] Furthermore, the steps performed by the indicator benchmark value generation module 14 include:

[0096] Based on the user-customized parameters, obtain the added material type parameters and added material content parameters;

[0097] Based on the predetermined purpose parameters, obtain the user type parameters and predetermined location parameters;

[0098] Input the added material type parameter and the added material content parameter into the adjustment parameter matching model to generate the first adjustment parameter;

[0099] The user type parameter and the predetermined location parameter are input into the adjustment parameter matching model to generate the second adjustment parameter;

[0100] The standard values ​​of the multi-layer detection index are adjusted according to the first adjustment parameter and the second adjustment parameter to generate the benchmark values ​​of the multi-layer detection index.

[0101] Furthermore, the steps performed by the indicator benchmark value generation module 14 include:

[0102] Construct the first half-model training data, which includes a dataset with added material type records, a dataset with added material content records, and a dataset with first adjustment parameter identifiers;

[0103] Construct the training data for the second half of the model, which includes a user type record dataset, a predetermined location record dataset, and a second adjustment parameter identifier dataset;

[0104] Based on the training data of the first half-model, train the first parameter-adjusted matching half-model;

[0105] Based on the training data of the second half-model, train the second parameter-adjusted matching half-model;

[0106] The first adjustment parameter matching semi-model and the second adjustment parameter matching semi-model are merged to generate the adjustment parameter matching model.

[0107] Furthermore, the steps performed by the indicator benchmark value generation module 14 include:

[0108] Using the added material type record dataset and the added material content record dataset as input data, and the first adjustment parameter identifier dataset as output identifier data, a first sub-model is constructed.

[0109] Extract the training data of the first half-model whose output accuracy does not meet the preset accuracy and add it to the first training data;

[0110] Based on the first training data, train the second sub-model;

[0111] The first sub-model and the second sub-model are merged to generate the first adjusted parameter matching semi-model.

[0112] Furthermore, the steps performed by the indicator benchmark value generation module 14 include:

[0113] Using the user type record dataset and the predetermined location record dataset as input data, and the second adjustment parameter identifier dataset as output identifier data, a third sub-model is constructed.

[0114] Extract the training data of the second half-model whose output accuracy of the third sub-model does not meet the preset accuracy, and add it to the second training data;

[0115] Train the fourth sub-model based on the second training data;

[0116] The third sub-model and the fourth sub-model are merged to generate the second adjustment parameter matching semi-model.

[0117] Furthermore, the indicator weight distribution module 15 performs the following steps:

[0118] Obtain the first-importance rating participant, the second-importance rating participant, and so on up to the Nth-importance rating participant, wherein the information exchange between any two participants is isolated.

[0119] The multi-layer detection index of the gravity blanket is input into the first importance scoring participant, the second importance scoring participant, and so on up to the Nth importance scoring participant, to obtain the scoring results of the first participant, the second participant, and so on up to the Nth participant;

[0120] Formula for obtaining indicator weight distribution:

[0121]

[0122] Among them, w i Let x be the weight distribution result of the i-th indicator. ik For the score of the k-th parameter of the i-th indicator, x j The score for the j-th indicator on the k-th parameter side;

[0123] The scoring results of the first participant, the scoring results of the second participant, and up to the scoring results of the Nth participant are input into the indicator weight distribution formula to generate the multi-level indicator weight distribution results.

[0124] Furthermore, the quality inspection module 17 performs the following steps:

[0125] Based on the multi-layer detection index of the gravity blanket, feature analysis is performed on the multi-layer image acquisition results to generate multi-layer detection index feature values;

[0126] Determine whether the feature values ​​of the multi-layer detection index meet the benchmark values ​​of the multi-layer detection index;

[0127] If the requirements are not met, generate non-compliant indicators and add them to the quality inspection results.

[0128] If satisfied, a comprehensive feature value of the gravity blanket is generated based on the multi-level index weight distribution results and the multi-level detection index feature values.

[0129] Determine whether the comprehensive characteristic value of the gravity blanket meets the preset characteristic value;

[0130] If the conditions are met, a qualified identification information is generated to identify the gravity blanket to be tested.

[0131] In summary, any step of the method described above can be stored as a computer instruction or program in an unrestricted computer memory, and can be called and identified by an unrestricted computer processor to implement any method in the embodiments of this application, without any additional restrictions.

[0132] Furthermore, the "first" or "second" mentioned above may not only represent a sequential relationship, but may also represent a specific concept, and / or refer to the individual or collective selection of multiple elements. Clearly, those skilled in the art can make various modifications and variations to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A machine vision-based method for quality inspection of gravity blankets, characterized in that, The method employs a machine vision-based gravity blanket quality inspection system, which is communicatively connected to an image acquisition device. The method includes: Based on the type of gravity blanket, match the multi-layer test indicators for the gravity blanket; Input the multi-layer test indexes of the gravity blanket into the standard value matching table to generate standard values ​​for the multi-layer test indexes; Obtain pre-application scenario parameters, wherein the pre-application scenario parameters include user-customized parameters and predetermined purpose parameters; The standard values ​​of the multi-layer detection index are adjusted according to the user-customized parameters and the predetermined purpose parameters to generate the multi-layer detection index benchmark values; The weight distribution of the multi-layer detection indicators of the gravity blanket is performed by iterating through the multi-layer indicator weight distribution results. The image acquisition device is used to acquire images of the gravity blanket to be tested, generating multi-layer image acquisition results. Based on the multi-layer index weight distribution results and the multi-layer detection index benchmark values, the quality of the gravity blanket to be tested is performed based on the multi-layer image acquisition results to generate quality detection results; the step of adjusting the multi-layer detection index standard values ​​according to the user-customized parameters and the predetermined usage parameters to generate multi-layer detection index benchmark values ​​includes: Based on the user-customized parameters, obtain the added material type parameters and added material content parameters; Based on the predetermined purpose parameters, obtain the user type parameters and predetermined location parameters; Input the added material type parameter and the added material content parameter into the adjustment parameter matching model to generate the first adjustment parameter; The user type parameter and the predetermined location parameter are input into the adjustment parameter matching model to generate the second adjustment parameter; Adjusting the standard value of the multi-layer detection index according to the first adjustment parameter and the second adjustment parameter to generate the benchmark value of the multi-layer detection index; the step of adjusting the standard value of the multi-layer detection index according to the user-customized parameter and the predetermined purpose parameter to generate the benchmark value of the multi-layer detection index further includes: Construct the first half-model training data, which includes a dataset with added material type records, a dataset with added material content records, and a dataset with first adjustment parameter identifiers; Construct the training data for the second half of the model, which includes a user type record dataset, a predetermined location record dataset, and a second adjustment parameter identifier dataset; Based on the training data of the first half-model, train the first parameter-adjusted matching half-model; Based on the training data of the second half-model, train the second parameter-adjusted matching half-model; The first adjustment parameter matching semi-model and the second adjustment parameter matching semi-model are merged to generate the adjustment parameter matching model.

2. The method as described in claim 1, characterized in that, The step of training a first parameter-adjusted matching semi-model based on the first semi-model training data includes: Using the added material type record dataset and the added material content record dataset as input data, and the first adjustment parameter identifier dataset as output identifier data, a first sub-model is constructed. Extract the training data of the first half-model whose output accuracy does not meet the preset accuracy and add it to the first training data; Based on the first training data, train the second sub-model; The first sub-model and the second sub-model are merged to generate the first adjusted parameter matching semi-model.

3. The method as described in claim 1, characterized in that, The step of training the second parameter-adjusted matching half-model based on the second half-model training data includes: Using the user type record dataset and the predetermined location record dataset as input data, and the second adjustment parameter identifier dataset as output identifier data, a third sub-model is constructed. Extract the training data of the second half-model whose output accuracy of the third sub-model does not meet the preset accuracy, and add it to the second training data; Train the fourth sub-model based on the second training data; The third sub-model and the fourth sub-model are merged to generate the second adjustment parameter matching semi-model.

4. The method as described in claim 1, characterized in that, The step of traversing the multiple layers of detection indicators of the gravity blanket and performing weight distribution to generate multi-layer indicator weight distribution results includes: Obtain the first-importance rating participant, the second-importance rating participant, and so on up to the Nth-importance rating participant, wherein the information exchange between any two participants is isolated. The multi-layer detection index of the gravity blanket is input into the first importance scoring participant, the second importance scoring participant, and so on up to the Nth importance scoring participant, to obtain the scoring results of the first participant, the second participant, and so on up to the Nth participant; Formula for obtaining indicator weight distribution: in, This represents the weight distribution result for the i-th indicator. The score for the k-th parameter of the i-th indicator. The score for the j-th indicator on the k-th parameter side; The scoring results of the first participant, the scoring results of the second participant, and up to the scoring results of the Nth participant are input into the indicator weight distribution formula to generate the multi-level indicator weight distribution results.

5. The method as described in claim 1, characterized in that, The step of performing quality detection on the gravity blanket to be detected based on the multi-layer index weight distribution results and the multi-layer detection index benchmark values, and generating quality detection results, includes: Based on the multi-layer detection index of the gravity blanket, feature analysis is performed on the multi-layer image acquisition results to generate multi-layer detection index feature values; Determine whether the feature values ​​of the multi-layer detection index meet the benchmark values ​​of the multi-layer detection index; If the requirements are not met, generate non-compliant indicators and add them to the quality inspection results. If satisfied, a comprehensive feature value of the gravity blanket is generated based on the multi-level index weight distribution results and the multi-level detection index feature values. Determine whether the comprehensive characteristic value of the gravity blanket meets the preset characteristic value; If the conditions are met, a qualified identification information is generated to identify the gravity blanket to be tested.

6. A machine vision-based gravity blanket quality inspection system, characterized in that, The system is used to implement the machine vision-based gravity blanket quality detection method according to any one of claims 1-5, the system is communicatively connected to an image acquisition device, and the system includes: The detection index matching module is used to match the multi-layer detection index of the gravity blanket according to the type of gravity blanket; The index standard value generation module is used to input the multi-layer detection index of the gravity blanket into the standard value matching table and generate multi-layer detection index standard values. A pre-application scenario parameter matching module is used to obtain pre-application scenario parameters, wherein the pre-application scenario parameters include user-customized parameters and predetermined purpose parameters; The indicator benchmark value generation module is used to adjust the multi-layer detection indicator standard value according to the user-customized parameters and the predetermined purpose parameters, and generate multi-layer detection indicator benchmark values. The indicator weight distribution module is used to traverse the multi-layer detection indicators of the gravity blanket to distribute the weights and generate multi-layer indicator weight distribution results. The image acquisition module is used to acquire images of the gravity blanket to be tested through the image acquisition device and generate multi-layer image acquisition results; The quality inspection module is used to perform quality inspection on the gravity blanket to be inspected based on the multi-layer index weight distribution results and the multi-layer inspection index benchmark values, and generate quality inspection results.

Images