A method for identifying formed wheels based on machine learning

Abstract

The present invention discloses a method for identifying a formed wheel hub based on machine learning. To solve the problems of being unable to distinguish wheel hub codes and the non-uniformity of wheel hub structures, the method includes the following steps: S1: Collect wheel hub code and wheel hub structure data and construct a wheel hub database; S2: Collect wheel hub codes and images of both side structures, and extract image features based on a neural network; S3: Compare the extracted image features with the wheel hub codes and wheel hub structures in the database, verify the wheel hub codes with the corresponding wheel hub structures, and obtain a matching result; S4: Identify the wheel hub type according to the matching result. The beneficial effects of the present invention are as follows: This method identifies the wheel hub based on the convolutional neural network of machine learning, which is convenient for determining the wheel hub type; this method realizes the identification of the non-uniformity of wheel hub codes and wheel hub structures through two different identification methods for wheel hub codes and wheel hub structures, and can also avoid the occurrence of inaccurate identification caused by identification errors.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and particularly to a method for identifying formed wheels based on machine learning. Background Art

[0002] With the development of the automotive industry, the automotive parts manufacturing industry, especially the automotive wheel manufacturing industry, has also developed rapidly. In the automotive wheel industry, it is particularly crucial to understand the model parameters of wheels. For example, the diameter and width of the wheel determine the size of the tire that can be matched and installed, the center hole diameter of the wheel needs to match the bearing size of the vehicle, and parameters such as the offset, number of bolt holes, and pitch circle diameter of the wheel all need to meet the original design conditions of the vehicle. During vehicle production, both the vehicle and the wheels installed on it have set model parameters. During the repair and replacement of wheels, the parameters of the newly replaced wheels also need to conform to the original wheel parameter settings. If the gap between the two is too large, it will greatly affect the driving performance of the vehicle and even cause accidents. Therefore, we need to find a certain method to facilitate obtaining the model parameters of the wheels.

[0003] A "Method and System for Detecting and Positioning Wheels on a Roller Path Based on a Camera and Machine Learning" disclosed in a Chinese patent document, with the publication number CN110992339A, the system includes two parts: hardware and software. The hardware part includes a main control unit and a vision detection unit. The main control unit is responsible for connecting devices such as cameras, conveyor belts, and PLCs and running related programs. The vision detection unit, including multiple industrial RGB cameras, is responsible for collecting images of the wheels on the roller path in real time. The software part includes a main control software and a vision detection software. The main control software sends motion control signals to the PLC according to the operation process and the signal status of each hardware device. The vision detection software is responsible for detecting, identifying, and positioning the wheels on the roller path. Its disadvantages are: it cannot distinguish the situation where the wheel codes and wheel structures are not unified. Summary of the Invention

[0004] The present invention mainly aims to solve the problem of not being able to distinguish the situation where the wheel codes and wheel structures are not unified, and provides a method for identifying formed wheels based on machine learning, which can distinguish the situation where the wheel codes and wheel structures are not unified through two different wheel identification methods.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for identifying formed wheels based on machine learning, comprising the following steps:

[0007] S1: Collect wheel code and wheel structure data and construct a wheel database;

[0008] S2: Collect the wheel hub code and images of the structures on both sides, and extract image features based on a neural network;

[0009] S3: Compare the extracted image features with the wheel hub codes and wheel hub structures in the database, verify the wheel hub code and the corresponding wheel hub structure, and obtain a matching result;

[0010] S4: Identify the wheel hub type according to the matching result.

[0011] The wheel hub database includes an interrelated wheel hub code database and a wheel hub structure database, which is convenient for comparing between the identified wheel hub code and the wheel hub type determined by the wheel hub structure.

[0012] Collecting the structures on both sides of the wheel hub can more accurately determine the wheel hub type corresponding to the wheel hub structure, which is convenient for improving the accuracy of the result.

[0013] This method can identify the situation where the wheel hub code and the wheel hub structure are not unified through two different wheel hub identification methods of the wheel hub code and the wheel hub structure, and can also avoid the occurrence of inaccurate identification caused by incorrect identification.

[0014] Preferably, the step of collecting the wheel hub code and wheel hub structure data and constructing the wheel hub database in step S1 includes the following steps:

[0015] S11: Analyze the wheel hub code structure according to the wheel hub code rule;

[0016] S12: Store the meaning corresponding to each code and construct a code database;

[0017] S13: Store and construct a structure database according to the wheel hub structures of different wheel hub models.

[0018] The unified coding data of auto parts is represented in two ways: basic data or basic data plus extended data. The basic data consists of the Global Trade Item Number GTIN (composed of the manufacturer identification code, the commodity item code, and the check digit) and the part lot number or part serial number. Among them, the application identifier 01 is mandatory, and the application identifiers 10 and 21 are in three items. In addition to the basic data, when additional data information is required for auto parts, extended data can be used.

[0019] Storing the meaning corresponding to each code in the constructed code database is convenient for the wheel hub code identified later to match the corresponding wheel hub type.

[0020] Storing the structure parameters of different types of wheel hubs in the structure database is convenient for the wheel hub structure identified later to match the corresponding wheel hub type.

[0021] Preferably, the hub structure described in step S13 includes a bolt hole structure, a window structure, a valve hole structure, a spoke structure, a mounting surface structure, a rib back structure, a rim structure, an offset, a hanging structure, and a vehicle loading curve.

[0022] The bolt hole structure includes straight hole bolt holes, tapered bolts, spherical bolt holes, etc.; the valve hole structure includes ordinary valve holes, valve holes with back boring, etc.; the mounting surface structure includes mounting surface flash escapes, etc.; the rib back structure includes rib back flash escapes, etc., and the rib back flash escape includes rib tips, slopes, and rib middles, etc.; the rim structure includes rim A, rim B, rim C, rim D, etc.; the offset refers to the distance from the center line of the rim to the mounting surface, including positive offset, negative offset, and zero offset, etc.

[0023] Determining the hub type through the above hub structure can improve the accuracy of this method.

[0024] Preferably, the hub code includes a prefix code, a manufacturer identification code, a commodity item code, and a check code. The commodity item code includes a wheel material code, a forming process code, a wheel structure code, and a wheel surface condition code.

[0025] The prefix code consists of 2 - 3 digits and is assigned by the International Article Numbering Association. The prefix digits 690 - 697 have been assigned by the International Coding Association (GS1) to the China Article Numbering Center.

[0026] The manufacturer identification code consists of 8 digits (including the prefix code) and is responsible for being assigned and managed by the China Article Numbering Center. In view of the particularity of the auto parts industry, the coding center has opened a green channel, and all auto parts enterprises are assigned 8 - digit manufacturer identification codes.

[0027] The commodity item code consists of 4 digits.

[0028] The wheel material code is the first digit. The codes 1 - 8 represent wheel materials, including: code 1: aluminum alloy, code 2: steel, code 3: magnesium alloy, code 4: carbon fiber, code 5: steel - aluminum composite, code 6: aluminum - magnesium composite, code 7: aluminum - carbon fiber composite, code 8: magnesium - carbon fiber composite.

[0029] The forming process code is the second digit. The codes 1 - 8 represent wheel forming processes, including: code 1: low - pressure casting, code 2: gravity casting, code 3: differential pressure casting, code 4: gravity pressure casting, code 5: liquid die forging, code 6: casting and spinning, code 7: forging, code 8: forging and spinning.

[0030] The wheel structure code is the third digit. The codes 1 - 3 represent wheel structures, including: code 1: one - piece integral wheel, code 2: two - piece combined wheel, code 3: three - piece combined wheel.

[0031] The wheel surface status code is the fourth digit, and the codes 1 - 8 represent the wheel surface status, including: Code 1: Full painting, Code 2: Painted bright surface, Code 3: Hydroplating, Code 4: Vacuum plating, Code 5: Polishing, Code 6: Electrophoresis, Code 7: Color matching, Code 8: Water transfer printing.

[0032] The check code is a single digit used to verify the correctness of the entire code.

[0033] Preferably, the acquisition of the hub code and the structure images on both sides in step S2 includes the following steps:

[0034] S21: Locate the position of the hub code and collect the hub code information;

[0035] S22: Collect the structural image information of the front, back, and side of the hub.

[0036] In step S21, through the existing positioning algorithm, the approximate position of the hub code is determined for image acquisition, which helps to reduce the difficulty of subsequent image processing and make the results more accurate.

[0037] In step S22, collecting the structural image information of all aspects of the hub can accurately obtain the image information of all structures, which is convenient for accurately identifying the hub type subsequently.

[0038] Preferably, the extraction of image features based on the neural network in step S2 includes the following steps:

[0039] S23: Perform image preprocessing on the collected images based on the convolutional neural network;

[0040] S24: Extract features from the processed images.

[0041] Image processing is carried out through the convolutional neural network of machine learning. Due to the learnability of the convolutional neural network, the recognition results of this method can be more accurate.

[0042] Preferably, step S23 includes the following steps:

[0043] S231: Perform image enhancement by means of rotation, adding Gaussian noise, and image jitter to construct a hub image dataset;

[0044] S232: Use 70% of the hub image datasets of different types as the training set and 30% of the datasets as the test set for network performance testing to obtain the final basic network type;

[0045] S233: Adjust the network feature extraction part and the feature classification part of the final basic network type;

[0046] S234: Set the general parameters and hyperparameters of the convolutional neural network, train the general parameters and continuously update them; determine the values of the hyperparameters by the method of controlling variables.

[0047] S235: Complete the preprocessing of the hub image based on the convolutional neural network.

[0048] The rotation in step S231 includes up-down flipping, left-right flipping, etc., and randomly generates a rotation angle, which can simulate the position images of the hub with different relative nozzle hole structures collected by the image system in the pipeline.

[0049] The purpose of adding Gaussian noise to the dataset is to enhance the network's learning ability for images containing noise, and the method is implemented through the existing Gaussian noise density function formula.

[0050] The purpose of image jitter is to improve the network's recognition ability for hub images collected under different lighting conditions.

[0051] The purpose of image data regularization is to make the input image data of the same size. The original collected images have different resolution sizes and most of them have too high resolution. If directly used for learning, the computational amount is too large. Therefore, first adjust the resolution of the image. The adjustment criterion is: while ensuring the aspect ratio of the image remains unchanged, set the maximum width of the image to 480 and the maximum height to 640. Then, in order to meet the requirement that the input image size in the network structure is the same, use the common "Reshape" operation in image processing to regularize the image size to 224×224. Finally, normalize the image.

[0052] Step S233 is a prior art to make the network meet the requirements of the hub image classification task.

[0053] Step S234 is the design of general parameters and hyperparameters in the prior art.

[0054] Preferably, the comparison of the extracted image features with the hub codes and hub structures in the database in step S3 includes the following steps:

[0055] S31: Compare the extracted image features with the hub codes in the database, analyze the meaning of the hub codes in the image features, and determine the hub type.

[0056] S32: Compare the extracted image features with the hub structures in the database, analyze the hub type corresponding to the hub structure in the image features, and determine the hub type.

[0057] By comparing the extracted image features, the hub codes and the hub structures in the database, the hub types determined by the two recognition methods can be obtained, which can improve the accuracy of this method.

[0058] Preferably, the verification of the wheel hub code and the corresponding wheel hub structure in step S3 includes the following steps:

[0059] S33: Compare the wheel hub type corresponding to the wheel hub code with the wheel hub type corresponding to the wheel hub structure;

[0060] S34: If the two wheel hub types are the same, output "matching result is the same"; if the two wheel hub types are different, jump to step S1 until the two wheel hub types are the same.

[0061] Steps S33 and S34 achieve the purpose of accurate identification by comparing the wheel hub types obtained by two different identification methods, can distinguish the situation where the wheel hub code and the wheel hub structure are not unified, and can also avoid the occurrence of inaccurate identification caused by misidentification.

[0062] Preferably, the identification of the wheel hub type according to the matching result in step S4 includes the following steps:

[0063] S41: Identify the matching result output in step S34;

[0064] S42: When the matching result is "matching result is the same", output the wheel hub type.

[0065] Step S4 can only output the wheel hub type when the matching result is "matching result is the same", otherwise, this method will continue to loop, which can improve the accuracy of the result.

[0066] The beneficial effects of the present invention are:

[0067] (1) This method identifies the wheel hub based on the convolutional neural network of machine learning, which is convenient for determining the wheel hub type.

[0068] (2) This method realizes the discrimination of the situation where the wheel hub code and the wheel hub structure are not unified through two different identification methods of the wheel hub code and the wheel hub structure.

[0069] (3) This method avoids the occurrence of inaccurate identification caused by misidentification through two different identification methods of the wheel hub code and the wheel hub structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a flow diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0071] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0072] As Figure 1 shown, a method for identifying a formed wheel hub based on machine learning includes the following steps:

[0073] S1: Collect the hub code and hub structure data and build a hub database;

[0074] S2: Collect the hub code and the structure images on both sides, and extract the image features based on a neural network;

[0075] S3: Compare the extracted image features with the hub code and hub structure in the database, verify the hub code and the corresponding hub structure, and obtain a matching result;

[0076] S4: Identify the hub type according to the matching result.

[0077] The hub database includes an interrelated hub code database and a hub structure database, which is convenient for comparing between the identified hub code and the hub type determined by the hub structure.

[0078] Collecting the structure on both sides of the hub can more accurately determine the hub type corresponding to the hub structure, which is convenient for improving the accuracy of the result.

[0079] This method can identify the situation where the hub code and the hub structure are not unified through two different hub identification methods of the hub code and the hub structure, and can also avoid the occurrence of inaccurate identification caused by incorrect identification.

[0080] In step S1, collecting the hub code and hub structure data and building the hub database includes the following steps:

[0081] S11: Analyze the hub code structure according to the hub code rule;

[0082] S12: Store the meaning corresponding to each code and build a code database;

[0083] S13: Store and build a structure database according to the hub structures of different hub models.

[0084] The unified coding data of automotive parts is represented in two ways: basic data or basic data plus extended data. The basic data consists of the Global Trade Item Number GTIN (composed of the manufacturer identification code, item code, and check digit) and the part lot number or part serial number. Among them, the application identifier 01 is mandatory, and the application identifiers 10 and 21 are three items. In addition to the basic data, when additional data information is required for automotive parts, extended data can be used.

[0085] Storing the meaning corresponding to each code in the built code database is convenient for the hub code identified subsequently to match the corresponding hub type.

[0086] Storing the structure parameters of different types of hubs in the structure database is convenient for the hub structure identified subsequently to match the corresponding hub type.

[0087] In step S13, the hub structure includes a bolt hole structure, a window structure, a valve hole structure, a spoke structure, a mounting surface structure, a rib back structure, a rim structure, an offset, a hanging structure, and a loading curve.

[0088] The bolt hole structure includes straight hole bolt holes, tapered bolts, spherical bolt holes, etc.; the valve hole structure includes ordinary valve holes, valve holes with back boring, etc.; the mounting surface structure includes mounting surface flash escape, etc.; the rib back structure includes rib back flash escape, etc., and the rib back flash escape includes rib tips, slopes, and rib middles, etc.; the rim structure includes Rim A, Rim B, Rim C, Rim D, etc.; the offset refers to the distance from the rim center line to the mounting surface, including positive offset, negative offset, and zero offset, etc.

[0089] Determining the hub type through the above hub structure can improve the accuracy of this method.

[0090] The hub code includes a prefix code, a manufacturer identification code, a commodity item code, and a check code. The commodity item code includes a wheel material code, a forming process code, a wheel structure code, and a wheel surface condition code.

[0091] The prefix code consists of 2 - 3 digits and is assigned by the International Article Numbering Association. The prefix digits 690 - 697 have been assigned by the International Coding Association (GS1) to the China Article Numbering Center.

[0092] The manufacturer identification code consists of 8 digits (including the prefix code) and is responsible for being assigned and managed by the China Article Numbering Center. In view of the particularity of the auto parts industry, the coding center has opened a green channel, and all 8 - digit manufacturer identification codes are assigned to auto parts enterprises.

[0093] The commodity item code consists of 4 digits.

[0094] The wheel material code is the first digit. The codes 1 - 8 represent wheel materials, including: Code 1: Aluminum alloy, Code 2: Steel, Code 3: Magnesium alloy, Code 4: Carbon fiber, Code 5: Steel - aluminum composite, Code 6: Aluminum - magnesium composite, Code 7: Aluminum - carbon fiber composite, Code 8: Magnesium - carbon fiber composite.

[0095] The forming process code is the second digit. The codes 1 - 8 represent wheel forming processes, including: Code 1: Low - pressure casting, Code 2: Gravity casting, Code 3: Differential pressure casting, Code 4: Gravity pressure casting, Code 5: Liquid die forging, Code 6: Cast spinning, Code 7: Forging, Code 8: Forging spinning.

[0096] The wheel structure code is the third digit. The codes 1 - 3 represent wheel structures, including: Code 1: One - piece integral wheel, Code 2: Two - piece combined wheel, Code 3: Three - piece combined wheel.

[0097] The wheel surface status code is the fourth digit, and the codes 1-8 represent the wheel surface status, including: Code 1: full painting, Code 2: painted bright surface, Code 3: electroplating, Code 4: vacuum plating, Code 5: polishing, Code 6: electrophoresis, Code 7: color matching, Code 8: water transfer printing.

[0098] The check code is a single-digit number used to verify the correctness of the entire code.

[0099] The steps of collecting the hub code and the structure images on both sides in step S2 include the following steps:

[0100] S21: Locate the position of the hub code and collect the hub code information;

[0101] S22: Collect the structure image information of the front, back, and side of the hub.

[0102] In step S21, through the existing positioning algorithm, the approximate position of the hub code is determined for image acquisition, which helps to reduce the difficulty of subsequent image processing and make the result more accurate.

[0103] In step S22, collecting the structure image information of the hub in all directions can accurately obtain the image information of all structures, which is convenient for accurately identifying the hub type subsequently.

[0104] The steps of extracting image features based on a neural network in step S2 include the following steps:

[0105] S23: Perform image preprocessing on the collected images based on a convolutional neural network;

[0106] S24: Extract features from the processed images.

[0107] By using the convolutional neural network of machine learning for image processing, due to the learnability of the convolutional neural network, the recognition result of this method can be more accurate.

[0108] Step S23 includes the following steps:

[0109] S231: Perform image enhancement by rotating, adding Gaussian noise, and image jittering to construct a hub image dataset;

[0110] S232: Use 70% of the hub image datasets of different types as the training set and 30% of the datasets as the test set for network performance testing to obtain the final basic network type;

[0111] S233: Adjust the network feature extraction part and the feature classification part of the final basic network type;

[0112] S234: Set the general parameters and hyperparameters of the convolutional neural network, train the general parameters and continuously update them; determine the values of the hyperparameters by the method of controlling variables.

[0113] S235: Complete the preprocessing of the hub image based on the convolutional neural network.

[0114] The rotation in step S231 includes up-down flipping, left-right flipping, etc., and randomly generates a rotation angle, which can simulate the position images of the hub with different relative nozzle hole structures collected by the image system in the pipeline.

[0115] The purpose of adding Gaussian noise to the dataset is to enhance the network's learning ability for images containing noise, and the method is implemented through the existing Gaussian noise density function formula.

[0116] The purpose of image jitter is to improve the network's recognition ability for hub images collected under different lighting conditions.

[0117] The purpose of image data regularization is to make the sizes of the input image data consistent. The resolutions of the originally collected images are different and most of them are too high, and the computational cost is too large if directly used for learning. Therefore, first adjust the resolution of the image. The adjustment criterion is: set the maximum width of the image to 480 and the maximum height to 640 while keeping the aspect ratio of the image unchanged. Then, in order to meet the requirement that the sizes of the input images in the network structure are consistent, use the common "Reshape" operation in image processing to regularize the image size to 224×224. Finally, normalize the image.

[0118] Step S233 is a prior art to make the network meet the requirements of the hub image classification task.

[0119] Step S234 is the design of general parameters and hyperparameters in the prior art.

[0120] In step S3, comparing the extracted image features with the hub codes and hub structures in the database includes the following steps:

[0121] S31: Compare the extracted image features with the hub codes in the database, analyze the meaning of the hub codes in the image features, and determine the hub type.

[0122] S32: Compare the extracted image features with the hub structures in the database, analyze the hub type corresponding to the hub structure in the image features, and determine the hub type.

[0123] By comparing the extracted image features, the hub codes and the hub structures in the database, the hub types determined by the two recognition methods can be obtained, which can improve the accuracy of this method.

[0124] In step S3, verifying the wheel hub code against the corresponding wheel hub structure includes the following steps:

[0125] S33: Compare the wheel hub type corresponding to the wheel hub code with the wheel hub type corresponding to the wheel hub structure;

[0126] S34: If the two wheel hub types are the same, output "matching result is the same"; if the two wheel hub types are different, jump to step S1 until the two wheel hub types are the same.

[0127] Steps S33 and S34 achieve the purpose of accurate identification by comparing the wheel hub types obtained by two different identification methods, can identify the situation where the wheel hub code and the wheel hub structure are not unified, and can also avoid the occurrence of inaccurate identification due to incorrect identification.

[0128] In step S4, identifying the wheel hub type according to the matching result includes the following steps:

[0129] S41: Identify the matching result output in step S34;

[0130] S42: When the matching result is "matching result is the same", output the wheel hub type.

[0131] In step S4, the wheel hub type can be output only when the matching result is "matching result is the same"; otherwise, this method continues to loop, which can improve the accuracy of the result.

[0132] It should be understood that this embodiment is only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of this application.

Claims

1. A method for identifying formed wheels based on machine learning, characterized in that, It includes the following steps: S1: Collect hub codes and hub structure data, construct a hub database, establish a coding rule, and form a commodity item code consisting of 4 digits; S2: Collect hub codes and structural images on both sides, and extract image features based on a neural network; S3: Compare the extracted image features with the hub codes and hub structures in the database to obtain the hub types determined by two identification methods, verify the hub codes and the corresponding hub structures, and compare the hub types obtained by the two different identification methods to obtain a matching result; S4: Identify the hub type according to the matching result.

2. The method for identifying a formed wheel hub based on machine learning according to claim 1, wherein The hub code includes a prefix code, a manufacturer identification code, a commodity item code, and a check code. The commodity item code consists of 4 digits, including the wheel material code as the first digit, with codes 1-8 representing wheel materials, the forming process code as the second digit, with codes 1-8 representing wheel forming processes, the wheel structure code as the third digit, with codes 1-3 representing wheel structures, and the wheel surface state code as the fourth digit, with codes 1-8 representing wheel surface states.

3. The method for identifying a formed hub based on machine learning according to claim 1, wherein The steps of collecting hub codes and hub structure data and constructing a hub database in step S1 include the following steps: S11: Analyze the hub code structure according to the hub coding rule; S12: Store the meanings corresponding to each code and construct a coding database; S13: Store and construct a structure database according to the hub structures of different hub models.

4. The method for identifying a formed wheel hub based on machine learning according to claim 3, wherein The hub structure in step S13 includes a bolt hole structure, a window structure, a valve hole structure, a spoke structure, a mounting surface structure, a rib back structure, a rim structure, an offset, a hanging structure, and a loading curve.

5. A method for identifying a formed wheel hub based on machine learning according to claim 1, characterized in that The steps of collecting hub codes and structural images on both sides in step S2 include the following steps: S21: Locate the hub code position and collect hub code information; S22: Collect structural image information of the front, back, and side of the hub.

6. The method for identifying a formed wheel hub based on machine learning according to claim 1, characterized in that The steps of extracting image features based on a neural network in step S2 include the following steps: S23: Perform image preprocessing on the collected images based on a convolutional neural network; S24: Extract features from the processed images.

7. A method for identifying a formed wheel hub based on machine learning according to claim 6, characterized in that, Step S23 includes the following steps: S231: Perform image enhancement by means of rotation, adding Gaussian noise, and image jittering to construct a hub image dataset; S232: Use 70% of the hub image datasets of different types as the training set and 30% of the datasets as the test set for network performance testing to obtain the final basic network type; S233: Adjust the network feature extraction part and the feature classification part of the final basic network type; S234: Set the general parameters and hyperparameters of the convolutional neural network, train the general parameters and continuously update them; determine the values of the hyperparameters by the control variable method; S235: Complete hub image preprocessing based on the convolutional neural network.

8. The method for identifying a formed wheel hub based on machine learning according to claim 1, wherein The steps of comparing the extracted image features with the hub codes and hub structures in the database in step S3 include the following steps: S31: Compare the extracted image features with the hub codes in the database, analyze the meanings of the hub codes in the image features, and determine the hub type; S32: Compare the extracted image features with the wheel hub structures in the database, analyze the wheel hub types corresponding to the wheel hub structures in the image features, and determine the wheel hub types.

9. A method for identifying a formed wheel hub based on machine learning according to claim 8, characterized in that, The verification of associating the wheel hub codes with the corresponding wheel hub structures described in step S3 includes the following steps: S33: Compare the wheel hub types corresponding to the wheel hub codes and the wheel hub types corresponding to the wheel hub structures; S34: If the two wheel hub types are the same, output "The matching results are the same"; if the two wheel hub types are different, jump to step S1 until the two wheel hub types are the same.

10. The method for identifying a formed wheel hub based on machine learning according to claim 9, wherein, The identification of the wheel hub type according to the matching result described in step S4 includes the following steps: S41: Identify the matching result output in step S34; S42: When the matching result is "The matching results are the same", output the wheel hub type.

Images