A computer vision-based method and system for quality inspection of stamped parts

Abstract

This invention provides a computer vision-based method and system for quality inspection of stamped parts. The method includes: extracting a set of appearance features from high-definition images using computer vision technology; processing and analyzing the appearance feature set using a first quality inspection model to obtain a first pass / fail assessment value; if the first pass / fail assessment value is higher than a first assessment threshold, generating and outputting a quality inspection pass / fail conclusion; if the first pass / fail assessment value is lower than a second assessment threshold, generating and outputting a quality inspection fail / fail conclusion; if the first pass / fail assessment value is lower than the first assessment threshold but higher than the second assessment threshold, inputting the appearance feature set into a second quality inspection model and outputting a second pass / fail assessment value; if the second pass / fail assessment value is higher than a third assessment threshold, generating and outputting a quality inspection pass / fail conclusion; otherwise, generating and outputting a quality inspection fail / fail conclusion. This invention enables the identification of both obvious and inconspicuous defects in stamped parts.

Description

Technical Field

[0001] This invention relates to the field of quality inspection technology, and more specifically, to a method and system for quality inspection of stamped parts based on computer vision. Background Technology

[0002] Quality inspection of stamped parts is a crucial step, directly impacting the final quality and reliability of the product. Currently, some solutions utilize computer vision processing technology for quality inspection of stamped parts. These existing methods primarily extract the contour features of the stamped area, then calculate the deviation between the extracted contour features and the designed contour features. If the deviation exceeds a certain threshold, the part is considered unqualified.

[0003] However, this method of quality inspection of stamped parts based on a single factor only inspects those stamped parts with obvious stamping quality problems. For those stamped parts with non-obvious stamping quality problems, it is easy to mistakenly judge them as qualified. This will result in an excessively high percentage of unqualified products in the stamped parts product line, making it difficult to meet actual production needs. Summary of the Invention

[0004] To address the technical problems mentioned above, this invention provides a computer vision-based method, system, electronic device, and computer storage medium for inspecting the quality of stamped parts.

[0005] This invention provides a computer vision-based method for quality inspection of stamped parts, comprising the following steps:

[0006] A high-resolution image of the stamped part after stamping is obtained, and an appearance feature set of the stamped part is extracted from the high-resolution image based on computer vision technology; wherein, the appearance feature set includes contour information, depth information, color information, and texture information;

[0007] The first quality inspection model is used to process and analyze the appearance feature set to obtain a first qualified assessment value. If the first qualified assessment value is higher than the first assessment threshold, a qualified quality inspection conclusion information is generated and output. If the first qualified assessment value is lower than the second assessment threshold, a unqualified quality inspection conclusion information is generated and output.

[0008] If the first qualified assessment value is lower than the first assessment threshold but higher than the second assessment threshold, then the appearance feature set is input into the second quality inspection model, and the second quality inspection model outputs the second qualified assessment value.

[0009] If the second qualified assessment value is higher than the third assessment threshold, then the quality inspection qualified conclusion information is generated and output; otherwise, the quality inspection unqualified conclusion information is generated and output.

[0010] The first quality inspection model has a lower ability to identify non-obvious defects in stamped parts than the second quality inspection model.

[0011] Optionally, the first quality inspection model is constructed based on any one of support vector machine, Gaussian Naive Bayes, hidden Markov model, and random forest; the second quality inspection model is constructed based on a deep learning algorithm, wherein the deep learning algorithm is any one of convolutional neural network, recurrent neural network, long short-term memory network, and Transformer.

[0012] Optionally, the step of extracting the appearance feature set of the stamped part from the high-definition image based on computer vision technology includes:

[0013] The high-definition image is subjected to orientation correction processing, and preset stamping area distribution information is retrieved. Based on the stamping area distribution information, each stamping area is determined from the high-definition image.

[0014] Based on computer vision technology, sub-appearance feature sets are extracted from the image corresponding to the stamping area. The sub-appearance feature sets are then transformed and fused into a matrix to obtain the appearance feature set.

[0015] Optionally, the second quality inspection model is trained in the following manner:

[0016] The local network node uses the deep learning algorithm to construct the base model of the second quality inspection model, and publishes the base model to the collaborative network;

[0017] The collaborative network nodes in the collaborative network receive the base model and collect the corresponding training datasets themselves according to the model application scenario information in the published information, and use the training datasets to train the base model;

[0018] During the training of the base model by each of the collaborative network nodes, the local network node publishes training strategy update information to each of the collaborative network nodes through the collaborative network. Each of the collaborative network nodes adjusts its training method according to the training strategy update information until the base model reaches the training termination condition. The local network node then extracts the key parameters of each base model and feeds them back to the local network node.

[0019] The local network node receives the key parameters of each model, imports the key parameters of each model into the target base model, and tests the target base model using a test dataset. Based on the test results, it determines the key parameters of the target model from the key parameters of each model. The target base model is trained by the local network node.

[0020] By importing the key parameters of the target model into the target base model, the second quality inspection model is obtained.

[0021] Optionally, the local network node publishes training policy update information to each of the cooperative network nodes through the cooperative network, including:

[0022] The local network node obtains, through the collaborative network, a first number of training data types contained in the training dataset of each collaborative network node, and a second number of training data types that each collaborative network node has already used for training.

[0023] The severity of the training termination condition is determined based on the first quantity and the second quantity. The training strategy update information is generated based on the severity of the training termination condition and is published to each of the collaborative network nodes through the collaborative network.

[0024] Optionally, the severity is positively correlated with the first quantity and negatively correlated with the second quantity.

[0025] The present invention also provides a computer vision-based stamping part quality inspection system, including a feature extraction module, a first processing module, and a second processing module;

[0026] The feature extraction module is used to acquire high-definition images of stamped parts after stamping, and extract the appearance feature set of the stamped parts from the high-definition images based on computer vision technology; wherein, the appearance feature set includes contour information, depth information, color information, and texture information;

[0027] The first processing module is used to perform a first processing analysis on the appearance feature set using a first quality inspection model to obtain a first qualified assessment value. If the first qualified assessment value is higher than a first assessment threshold, a qualified quality inspection conclusion information is generated and output. If the first qualified assessment value is lower than a second assessment threshold, a unqualified quality inspection conclusion information is generated and output.

[0028] The second processing module is configured to input the appearance feature set into the second quality inspection model if the first qualified assessment value is lower than the first assessment threshold but higher than the second assessment threshold, and the second quality inspection model outputs the second qualified assessment value; if the second qualified assessment value is higher than the third assessment threshold, generate and output quality inspection qualified conclusion information, otherwise generate and output quality inspection unqualified conclusion information.

[0029] The first quality inspection model has a lower ability to identify non-obvious defects in stamped parts than the second quality inspection model.

[0030] The present invention also provides an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to perform the method as described in any of the preceding claims.

[0031] The present invention also provides a computer storage medium storing a computer program that, when executed by a processor, performs the method described in any of the preceding claims.

[0032] The present invention also provides a computer program product comprising computer code, which, when executed by a processor of an electronic device, implements the method described in any of the preceding claims.

[0033] The beneficial effects of this invention are as follows:

[0034] This invention employs two quality inspection models: a first model and a second model. These two models have different processing capabilities; the first model is weaker and uses simpler identification methods to handle stamped parts with obvious quality problems. The second model, however, uses more complex methods to identify stamped parts without obvious quality problems. Therefore, by using two quality inspection models with different processing capabilities in combination, this invention enables the identification of stamped parts with both obvious and subtle quality problems, thereby significantly improving the yield rate of stamped parts. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic flowchart of a computer vision-based method for inspecting the quality of stamped parts, as disclosed in an embodiment of the present invention.

[0037] Figure 2 This is a schematic diagram of a computer vision-based stamping part quality inspection system disclosed in an embodiment of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0039] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0040] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0041] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, or the orientation or positional relationships commonly used when the product is in use, are merely for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0042] like Figure 1 As shown in the figure, a computer vision-based method for inspecting the quality of stamped parts according to an embodiment of the present invention includes the following steps:

[0043] A high-resolution image of the stamped part after stamping is obtained, and an appearance feature set of the stamped part is extracted from the high-resolution image based on computer vision technology; wherein, the appearance feature set includes contour information, depth information, color information, and texture information;

[0044] The first quality inspection model is used to perform a first processing analysis on the appearance feature set to obtain a first qualified evaluation value. If the first qualified evaluation value is higher than the first evaluation threshold, a qualified quality inspection conclusion information is generated and output. If the first qualified evaluation value is lower than the second evaluation threshold, a unqualified quality inspection conclusion information is generated and output.

[0045] If the first qualified assessment value is lower than the first assessment threshold but higher than the second assessment threshold, then the appearance feature set is input into the second quality inspection model, and the second quality inspection model outputs the second qualified assessment value.

[0046] If the second qualified assessment value is higher than the third assessment threshold, then the quality inspection qualified conclusion information is generated and output; otherwise, the quality inspection unqualified conclusion information is generated and output.

[0047] The first quality inspection model has a lower ability to identify non-obvious defects in stamped parts than the second quality inspection model.

[0048] This invention employs two quality inspection models: a first model and a second model. These two models differ in their ability to identify non-obvious defects in stamped parts. The first model has a weaker identification capability than the second model, using simpler methods to handle stamped parts with obvious defects. The second model, however, uses more complex methods to identify stamped parts with non-obvious defects. Therefore, by using two quality inspection models with different processing capabilities in combination, this invention enables the identification of stamped parts with both obvious and non-obvious defects, thereby significantly improving the yield rate of stamped parts.

[0049] Specifically, this invention first uses a high-definition camera to capture high-definition images of the stamped parts. Computer vision technology is then used to extract contour, depth, color, and texture information from these high-definition images, forming an appearance feature set. A first quality inspection model is used to analyze each appearance element in the feature set using a simple method. If the stamped part has obvious defects (e.g., cracks, significant positional deviations), the first quality inspection model outputs a first pass assessment value lower than a second assessment threshold, resulting in a quality inspection failure conclusion. Conversely, if the stamped part has no defects, the first quality inspection model outputs a first pass assessment value higher than the first assessment threshold, resulting in a quality inspection pass conclusion. However, when the stamped part has non-obvious defects (e.g., minor cracks, mainly concentrated inside or on the back of the stamped part), the first quality inspection model cannot distinguish them, and its output first pass assessment value falls between the first and second assessment thresholds. In this case, a second quality inspection model with stronger processing capabilities is used for further processing. If the second qualified assessment value output by the second quality inspection model is higher than the third assessment threshold (the third assessment threshold can be the same as the first assessment threshold, or it can be a threshold set separately for the second quality inspection model), then the qualified quality inspection conclusion information is generated and output; otherwise, the unqualified quality inspection conclusion information is generated and output.

[0050] Optionally, the first quality inspection model is constructed based on any one of support vector machine, Gaussian Naive Bayes, hidden Markov model, and random forest; the second quality inspection model is constructed based on a deep learning algorithm, wherein the deep learning algorithm is any one of convolutional neural network, recurrent neural network, long short-term memory network, and Transformer.

[0051] In this embodiment of the invention, Support Vector Machines, Gaussian Naive Bayes, Hidden Markov Models, and Random Forests are traditional machine learning algorithms whose architectures have been thoroughly studied. Currently, it is practically difficult to achieve a qualitative breakthrough in these algorithms, resulting in a low upper limit to the processing power of the first quality inspection model built upon them. It can basically only identify very obvious or frequently occurring quality defects in stamped parts, while failing to accurately identify non-obvious defects (mainly internal defects) or new defects that have not been trained / identified before. Deep learning algorithms, which have become increasingly popular in recent years, are a class of algorithms inspired by the structure of the human brain. They use multi-layered neural networks to learn complex representations of data, making them particularly suitable for identifying the aforementioned non-obvious defects or new defects that have not been trained / identified before in stamped parts. Of course, the processing power of the first and second quality inspection models can be artificially altered by adjusting the training intensity (the amount of training data, training strategy, etc.). Generally, the classification / prediction accuracy of a model trained with less data is significantly lower than that of a model trained with more data. The differences in training strategies will be explained in the following sections.

[0052] Optionally, the step of extracting the appearance feature set of the stamped part from the high-definition image based on computer vision technology includes:

[0053] The high-definition image is subjected to orientation correction processing, and preset stamping area distribution information is retrieved. Based on the stamping area distribution information, each stamping area is determined from the high-definition image.

[0054] Based on computer vision technology, sub-appearance feature sets are extracted from the image corresponding to the stamping area. The sub-appearance feature sets are then transformed and fused into a matrix to obtain the appearance feature set.

[0055] In this embodiment of the invention, when the stamped part is transferred to the quality inspection area, its orientation may be irregular. In this case, it is necessary to first identify the key feature information, determine the current orientation of the stamped part based on this key feature information, and then compare the current orientation with the standard orientation. If they are inconsistent, the angle of the high-definition image is adjusted to correct the orientation. Simultaneously, the stamping scheme information for each stamped part is preset, which includes detailed stamping area distribution information. This stamping area distribution information includes the sub-regions where the stamping head will exert stamping pressure on the stamping template, thereby identifying the stamping areas from the high-definition image. Then, image processing is performed on the sub-images of each stamping area using computer vision technology to extract corresponding sub-appearance feature sets. The features contained in each sub-appearance feature set are organized into a sub-matrix, and then the sub-matrices are fused to obtain the appearance feature set used for analysis by the first quality inspection model.

[0056] By first identifying the stamping area, the entire high-definition image can be processed using computer vision, thus reducing unnecessary image processing load and improving quality inspection efficiency.

[0057] It should be noted that when the first pass assessment value for the classification and identification of the sub-appearance features corresponding to each stamping area in the appearance feature set by the first quality inspection model is lower than the second assessment threshold, a quality inspection failure conclusion is generated and output; only when all first pass assessment values ​​are higher than the first assessment threshold is a quality inspection pass conclusion generated and output. Of course, the first quality inspection model can also process each group of sub-features in the appearance feature set separately and obtain the corresponding first pass assessment value, and then perform threshold comparison; this invention does not limit this.

[0058] Optionally, the second quality inspection model is trained in the following manner:

[0059] The local network node uses the deep learning algorithm to construct the base model of the second quality inspection model, and publishes the base model to the collaborative network;

[0060] The collaborative network nodes in the collaborative network receive the base model and collect the corresponding training datasets themselves according to the model application scenario information in the published information, and use the training datasets to train the base model;

[0061] During the training of the base model by each of the collaborative network nodes, the local network node publishes training strategy update information to each of the collaborative network nodes through the collaborative network. Each of the collaborative network nodes adjusts its training method according to the training strategy update information until the base model reaches the training termination condition. The local network node then extracts the key parameters of each base model and feeds them back to the local network node.

[0062] The local network node receives the key parameters of each model, imports the key parameters of each model into the target base model, and tests the target base model using a test dataset. Based on the test results, it determines the key parameters of the target model from the key parameters of each model. The target base model is trained by the local network node.

[0063] By importing the key parameters of the target model into the target base model, the second quality inspection model is obtained.

[0064] In this embodiment of the invention, a collaborative distributed training method is used to train the aforementioned second quality inspection model. Specifically, a basic model of the second quality inspection model is first constructed using any of the aforementioned deep learning algorithms. This basic model is then published in a collaborative network, which includes local nodes and other collaborative network nodes. Some of the other collaborative network nodes will respond to the collaborative training information (generally based on their remaining training resources, the reward for collaborative training, etc.). The published information will also include the model application scenario information of the basic model (i.e., the identification of non-obvious defects in stamped parts in this invention). These responding collaborative network nodes will collect data related to the quality inspection of stamped parts to form corresponding training data for training the basic model published by the local network node.

[0065] In addition to employing a collaborative distributed training method, another improvement of this invention is that the local network node can also publish training strategy update information to each collaborative network node, enabling them to adjust their training methods. That is, the local network node can appropriately intervene in the training process of the collaborative network nodes, thereby facilitating the achievement of optimal training results. After each collaborative network node completes training, it extracts the key parameters of the trained base model and feeds them back to the local network node. Simultaneously, while the collaborative network nodes are training, the local network node also trains the base model based on its own collected training data to obtain the target base model. It then imports each set of key parameters into the target base model, and uses test data to test the target base model. Based on the test results, the optimal set of key parameters can be determined, and importing it into the target base model yields the aforementioned second quality control model.

[0066] The test dataset is a portion of the training data collected by the local network nodes. In other words, the local network nodes set one part of the collected training data as the training dataset and the other part as the test dataset. Furthermore, the training data in the training datasets determined by both the local and collaborative network nodes consists of data pairs between the appearance features of the stamped parts and probability labels indicating they are non-obvious defects. These probability labels are assigned manually or by machine annotation.

[0067] Optionally, the local network node publishes training policy update information to each of the cooperative network nodes through the cooperative network, including:

[0068] The local network node obtains, through the collaborative network, a first number of training data types contained in the training dataset of each collaborative network node, and a second number of training data types that each collaborative network node has already used for training.

[0069] The severity of the training termination condition is determined based on the first quantity and the second quantity. The training strategy update information is generated based on the severity of the training termination condition and is published to each of the collaborative network nodes through the collaborative network.

[0070] In this embodiment of the invention, during the training process of each collaborative network node, the local network node can obtain two types of information: a first number of training data types contained in the training dataset of each collaborative network node, and a second number of training data types that each collaborative network node has already used for training. Metal stamping parts include various types such as stainless steel, cold-rolled steel, aluminum, hot-dip galvanized steel, copper, carbon steel, low-titanium alloy, and magnesium alloy. The information published by the local network node only includes the stamping scenario, without specifically specifying which type of sheet metal to stamp (because stamping machines generally need to stamp multiple types of sheet metal). Therefore, each collaborative network node will collect training data corresponding to different sheet metal types based on its own experience and understanding, leading to the difference in the aforementioned first number. During training, each collaborative network node also inputs the corresponding training data grouped according to the sheet metal type, for example, first using training data of aluminum sheet metal to train the basic model, and then using training data of hot-dip galvanized steel sheet metal to train the basic model, which leads to the difference in the aforementioned second number.

[0071] Based on the first and second quantities mentioned above, the stringency of the training termination condition can be determined. The stringency of the training termination condition refers to the threshold value of the objective function in the training termination condition; that is, when the objective function value is below this threshold, the training is considered successful and training can be terminated. The higher the stringency, the smaller the corresponding threshold, and vice versa. The stringency can be a coefficient near 1. Multiplying this coefficient by the initial threshold adjusts the initial threshold. Each collaborative network node adjusts its training termination condition based on the published training strategy update information.

[0072] The aforementioned collaborative network preferably uses a blockchain network, such as a public blockchain network, a private blockchain network, or a hybrid blockchain network, without any specific limitation.

[0073] Optionally, the severity is positively correlated with the first quantity and negatively correlated with the second quantity.

[0074] In this embodiment of the invention, the more training data types included in the training dataset, the more stable the analytical capabilities of the base model need to be. In this case, a higher level of stringency is set, i.e., the threshold in the training termination condition is increased. Conversely, the less training data types included, the higher the stringency is set, i.e., the threshold in the training termination condition is appropriately decreased. Simultaneously, stringency is negatively correlated with the second quantity; that is, the more data types already used for training, the higher the training completion progress. In this case, the stringency of the base model can be appropriately reduced to avoid failing to meet the threshold in the training termination condition. Conversely, the fewer data types already used for training, the lower the training completion progress. In this case, stringent training of the base model needs to be maintained to avoid prematurely ending the training.

[0075] The formulas for calculating the severity and the first and second quantities are as follows:

[0076] ;

[0077] In the formula, For stringency, As the first quantity, For the second quantity, , Let be the coefficient, and .For example =0.05, =0.025. The above is merely an example and is not intended to limit the scope of protection of this invention.

[0078] like Figure 2 As shown, this embodiment of the invention also discloses a computer vision-based stamping part quality inspection system, including a feature extraction module, a first processing module, and a second processing module;

[0079] The feature extraction module is used to acquire high-definition images of stamped parts after stamping, and extract the appearance feature set of the stamped parts from the high-definition images based on computer vision technology; wherein, the appearance feature set includes contour information, depth information, color information, and texture information;

[0080] The first processing module is used to perform a first processing analysis on the appearance feature set using a first quality inspection model to obtain a first qualified assessment value. If the first qualified assessment value is higher than a first assessment threshold, a qualified quality inspection conclusion information is generated and output. If the first qualified assessment value is lower than a second assessment threshold, a unqualified quality inspection conclusion information is generated and output.

[0081] The second processing module is configured to input the appearance feature set into the second quality inspection model if the first qualified assessment value is lower than the first assessment threshold but higher than the second assessment threshold, and the second quality inspection model outputs the second qualified assessment value; if the second qualified assessment value is higher than the third assessment threshold, generate and output quality inspection qualified conclusion information, otherwise generate and output quality inspection unqualified conclusion information.

[0082] The first quality inspection model has a lower ability to identify non-obvious defects in stamped parts than the second quality inspection model.

[0083] This invention also discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to execute the method described in the foregoing embodiments.

[0084] This invention also discloses a computer storage medium storing a computer program, which is executed by a processor to perform the methods described in the foregoing embodiments.

[0085] This invention also discloses a computer program product containing computer code, which, when executed by a processor of an electronic device, implements the method described in the foregoing embodiments.

[0086] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0087] The methods and related apparatuses provided in this application are described with reference to the method flowcharts and / or structural diagrams provided in this application. Specifically, each block of the method flowchart and / or structural diagram, as well as combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to create a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 A process or multiple processes and / or structures illustrate the steps of the functions specified in one or more boxes.

[0088] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A computer vision-based method for quality inspection of stamped parts, characterized in that, Includes the following steps: A high-resolution image of the stamped part after stamping is obtained, and an appearance feature set of the stamped part is extracted from the high-resolution image based on computer vision technology; wherein, the appearance feature set includes contour information, depth information, color information, and texture information; The first quality inspection model is used to process and analyze the appearance feature set to obtain a first qualified assessment value. If the first qualified assessment value is higher than the first assessment threshold, a qualified quality inspection conclusion information is generated and output. If the first qualified assessment value is lower than the second assessment threshold, a unqualified quality inspection conclusion information is generated and output. If the first qualified assessment value is lower than the first assessment threshold but higher than the second assessment threshold, then the appearance feature set is input into the second quality inspection model, and the second quality inspection model outputs the second qualified assessment value. If the second qualified assessment value is higher than the third assessment threshold, then the quality inspection qualified conclusion information is generated and output; otherwise, the quality inspection unqualified conclusion information is generated and output. The first quality inspection model has a lower ability to identify non-obvious defects in stamped parts than the second quality inspection model. The second quality inspection model was trained in the following way: Local network nodes use deep learning algorithms to construct the base model of the second quality inspection model, and publish the base model to the collaborative network, which uses a blockchain network. The collaborative network nodes in the collaborative network receive the base model and collect the corresponding training datasets themselves according to the model application scenario information in the published information, and use the training datasets to train the base model; During the training of the base model by each of the collaborative network nodes, the local network node publishes training strategy update information to each of the collaborative network nodes through the collaborative network. Each of the collaborative network nodes adjusts its training method according to the training strategy update information until the base model reaches the training termination condition. The local network node then extracts the key parameters of each base model and feeds them back to the local network node. The local network node receives the key parameters of each model, imports the key parameters of each model into the target base model, and tests the target base model using a test dataset. Based on the test results, it determines the key parameters of the target model from the key parameters of each model. The target base model is trained by the local network node. By importing the key parameters of the target model into the target base model, the second quality inspection model is obtained. The local network node publishes training policy update information to each of the cooperative network nodes through the cooperative network, including: The local network node obtains, through the collaborative network, a first number of training data types contained in the training dataset of each collaborative network node, and a second number of training data types that each collaborative network node has already used for training. The severity of the training termination condition is determined based on the first quantity and the second quantity, and the training strategy update information is generated based on the severity of the training termination condition. The training strategy update information is published to each node of the collaborative network through the collaborative network. The severity is positively correlated with the first quantity and negatively correlated with the second quantity.

2. The method for quality inspection of stamped parts based on computer vision according to claim 1, characterized in that: The first quality inspection model is constructed based on any one of Support Vector Machine, Gaussian Naive Bayes, Hidden Markov Model, and Random Forest; the second quality inspection model is constructed based on a deep learning algorithm, wherein the deep learning algorithm is any one of Convolutional Neural Network, Recurrent Neural Network, Long Short-Term Memory Network, and Transformer.

3. The computer vision-based method for inspecting the quality of stamped parts according to claim 2, characterized in that: The appearance feature set of the stamped part is extracted from the high-definition image based on computer vision technology, including: The high-definition image is subjected to orientation correction processing, and preset stamping area distribution information is retrieved. Based on the stamping area distribution information, each stamping area is determined from the high-definition image. Based on computer vision technology, sub-appearance feature sets are extracted from the image corresponding to the stamping area. The sub-appearance feature sets are then transformed and fused into a matrix to obtain the appearance feature set.

4. A computer vision-based stamping part quality inspection system, the system being based on the method described in any one of claims 1-3, comprising a feature extraction module, a first processing module, and a second processing module; characterized in that: The feature extraction module is used to acquire high-definition images of stamped parts after stamping, and extract the appearance feature set of the stamped parts from the high-definition images based on computer vision technology; wherein, the appearance feature set includes contour information, depth information, color information, and texture information; The first processing module is used to perform a first processing analysis on the appearance feature set using a first quality inspection model to obtain a first qualified assessment value. If the first qualified assessment value is higher than a first assessment threshold, a qualified quality inspection conclusion information is generated and output. If the first qualified assessment value is lower than a second assessment threshold, a unqualified quality inspection conclusion information is generated and output. The second processing module is configured to input the appearance feature set into the second quality inspection model if the first qualified assessment value is lower than the first assessment threshold but higher than the second assessment threshold, and the second quality inspection model outputs the second qualified assessment value; if the second qualified assessment value is higher than the third assessment threshold, generate and output quality inspection qualified conclusion information, otherwise generate and output quality inspection unqualified conclusion information. The first quality inspection model has a lower ability to identify non-obvious defects in stamped parts than the second quality inspection model.

5. An electronic device, comprising: Memory containing executable program code; A processor coupled to the memory; characterized in that: the processor calls the executable program code stored in the memory to perform the method as described in any one of claims 1-3.

6. A computer storage medium storing a computer program, characterized in that: The computer program is executed by the processor to perform the method as described in any one of claims 1-3.

7. A computer program product, characterized in that: The computer program product includes computer code, which, when executed by a processor of an electronic device, implements the method as described in any one of claims 1-3.

Images