CN112258472B - Automatic scoring method for automobile exterior shape - Google Patents
Automatic scoring method for automobile exterior shape Download PDFInfo
- Publication number
- CN112258472B CN112258472B CN202011127986.0A CN202011127986A CN112258472B CN 112258472 B CN112258472 B CN 112258472B CN 202011127986 A CN202011127986 A CN 202011127986A CN 112258472 B CN112258472 B CN 112258472B
- Authority
- CN
- China
- Prior art keywords
- automobile
- model
- grade
- database
- modeling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0282—Rating or review of business operators or products
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- Evolutionary Computation (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Computer Hardware Design (AREA)
- Accounting & Taxation (AREA)
- Strategic Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Automation & Control Theory (AREA)
- Entrepreneurship & Innovation (AREA)
- Quality & Reliability (AREA)
- Mathematical Optimization (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Aviation & Aerospace Engineering (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an automatic scoring method for automobile exterior shapes, which comprises the steps of establishing a multi-view automobile picture database for automobile angle identification, automobile grade identification and automobile brand identification; creating an automobile modeling multi-view database for modeling scoring; creating an automobile user grading database which corresponds to the automobile modeling multi-view database one by one; and carrying out comprehensive grading processing such as outlier processing, user attribute analysis, user attribute weighted average, Bayesian statistical algorithm and the like on the user evaluation database. Dividing the automobile modeling multi-view database into a plurality of sub-databases according to automobile grade standards, establishing a mapping relation between multi-angle automobile pictures and reasonable modeling scores by using a deep learning regression method, and obtaining a plurality of different training models according to different training data sets; and (4) establishing an automatic evaluation machine for the external shape of the automobile to realize the automation of the evaluation of the external shape of the automobile.
Description
Technical Field
The invention relates to a big data-oriented automobile appearance scoring method considering multi-dimensional attributes of a user, and relates to an automatic quantitative scoring method based on multi-view automobile pictures.
Background
The quality of the automobile model is the embodiment of the design content of the modeler, meanwhile, the automobile model is an important factor considered when a consumer buys the automobile, automobile enthusiasts and related practitioners can appreciate and judge the automobile model, the automobile design conforming to the mass aesthetic sense can instantly catch the eyeball of the user, so that the psychology of the consumer is grasped, and the design of the model meeting the requirements of the user is very important for automobile manufacturers. Different automobile brands mostly have competitive automobile models on the same level, and the configuration, power and reliability of the automobile models are similar, so that the appearance modeling becomes the central importance of the differentiated design.
In the fierce market competition and the digital intelligent transformation of the automobile industry, the automobile modeling design is taken as an important ring in research and development design, and the technical innovation is urgently needed. In the research and development process, a reliable scoring model is required to score the automobile model designed by a designer to design the model satisfied by a consumer, and the reliable scoring model can be established through a big data driving and deep learning method to help and guide the design of the model.
Previous automobile model evaluation research work mostly relies on artificial feature definition and extraction, and although many advances are made in automobile model analysis methods using predefined features, these methods are highly dependent on artificial extraction. In the big data era, machine learning and deep learning develop rapidly, data mining and artificial intelligence are widely applied, machine learning represented by deep learning is greatly developed, and the computer vision direction is developed rapidly. The deep learning method has the advantages that the feature extraction, the feature selection, the classifier and the regressor are integrated, the full automation of the modeling scoring can be realized, and the subjectivity of the manually specified features is shielded as much as possible. The development of the internet provides greater leeway and possibility for designing the automobile model. The invention provides an external shape scoring method considering multi-dimensional attributes of users and different automobile grades, automobile brands and automobile angles, and an automatic quantitative scoring method of multi-view automobile pictures based on deep learning, so that the external shape of the automobile is reasonably and automatically scored.
Disclosure of Invention
Aiming at the existing problems, the invention provides an external shape automatic scoring method considering multi-dimensional attributes of users and different automobile grades, automobile brands and automobile angles.
The technical scheme of the invention is as follows:
an automatic scoring method for an automobile exterior shape comprises the following steps:
(1) creating a multi-view automobile picture database for identifying automobile angles, automobile grades and automobile brands: arranging multi-view automobile pictures of different brands and different models, wherein the total number of samples is not less than 15000; the database comprises automobiles of all grades, the automobiles are classified according to the grades, the cars are classified into A, B, C, D four grades, and the SUVs are classified into compact, medium and large SUVs; the number of the covered automobile brands is not less than 30, the number of the single brand automobile types is not less than 10, and 50 pictures are uniformly sampled around a circle at the head-up angle of each automobile type; respectively labeling according to automobile angles, automobile grades and automobile brands, and dividing a data set into a 70% training set and a 30% verification set;
(2) respectively carrying out angle recognition training, brand recognition training and grade recognition training on data in the multi-view automobile picture database to correspondingly obtain an automobile angle recognition model, an automobile brand recognition model and an automobile grade recognition model so as to recognize the angle, the grade and the brand of any given automobile picture;
(3) creating an automobile model multi-view database, collecting large-scale automobile model multi-view pictures of different grades and different brands, wherein the database is divided into 8 types according to cars (A, B, C, D four grades) and SUVs (compact, medium-sized and large-sized), the number of the types of the cars under each automobile grade is not less than 30, and 30 pictures are uniformly sampled when each type of the cars surrounds a circle at a head-up angle; marking the database picture by using an automobile angle identification model, an automobile grade identification model and an automobile brand identification model;
(4) creating an evaluation database corresponding to the modeling multi-view automobile picture database, and collecting comment samples of valid users corresponding to the modeling multi-view automobile picture database, wherein the comment samples comprise modeling scores and user information such as vehicle information, appearance evaluation scores, user specialties, user objectivity and the like.
(5) Performing data outlier filtering processing on the collected grading data; in the scoring process, outlier data inevitably occurs for various reasons, and if the score fairness is affected without preprocessing, reasonable outlier preprocessing is necessary to select. The invention adopts the Grabas method to remove the outlier, checks whether the data has the lower side outlier, processes the outlier and then calculates.
(6) In order to accurately analyze the credibility of the user participating in evaluation, quantize the attributes of the user, examine and quantize the multi-dimensional attributes of the user, and convert the measurement of various dimensional attributes into weight factors by adopting a multi-dimensional weighted Gaussian distribution function; then, calculating the weight of the user after quantization, and further calculating the weighted average score of the corresponding vehicle type;
(7) and (4) carrying out comprehensive weighted average on the results in the step (6), wherein the more the number of the evaluated persons of different vehicle types in the same vehicle grade is, the closer the Bayesian average is to the arithmetic average, the smaller the influence on the score is, and then, the Bayesian average algorithm is adopted to calculate the final weighted score. And (4) marking the obtained final weighting scores of the automobile models on the multi-view database of the automobile models one by one, and dividing a data set into a 70% training set and a 30% verification set.
(8) The method comprises the steps of establishing a mapping relation between automobile model scores and automobile model multi-view pictures by adopting a deep learning regression method, normalizing automobile model score labels to be 0-1, connecting a Sigmoid function to the last layer of a network to perform logistic regression calculation, extracting picture features, and fitting to model score points on a Sigmoid function curve, so that a trained model can predict and output the automobile model scores. In order to obtain a network model with higher expression capability, the automobile model multi-view database picture is divided into eight sub data sets according to the automobile grade, and different models are obtained through training.
(9) And (3) establishing an automatic evaluation machine for the external shape of the automobile, and calculating automobile grading adjustment parameters under different grades after obtaining the trained shape grading model. The method comprises the steps of carrying out scoring prediction on automobile model multi-view database pictures under a certain grade, measuring the accuracy of the prediction scoring by using MAE (mean absolute error), obtaining the scoring data statistics of different angles of each automobile model under the same automobile grade, and sorting and manufacturing an automobile model scoring statistical table. Selecting the lowest partial angle of MAE of each vehicle type score, for example, selecting the lowest angle, counting the angle in a table to obtain the occurrence frequency of each angle, calculating score adjusting parameters of each angle, namely the frequency of occurrence of each angle, wherein the angle with the highest frequency shows that the score credibility at the position is higher, the vehicle score adjusting parameters under different grades are different, and predicting and calculating the pictures of each sub-database respectively.
(10) By utilizing the automobile outer shape evaluation machine, the scoring prediction can be made for the given automobile picture. After a single-view-angle automobile model picture is given, identifying the grade of an automobile, selecting a corresponding scoring model, testing to obtain the model score x of the automobile picture, and obtaining a standard deviation sigma by using an automobile model score statistical table under the grade to obtain the automobile model score, wherein for example, the score confidence interval when the confidence coefficient reaches 95% is [ x-2 sigma, x +2 sigma ]; after a plurality of automobile modeling pictures are given, the angles of the automobile model pictures are identified, the automobile grades of the automobile are identified, corresponding scoring prediction models are selected to obtain modeling scores of different angle pictures, and the scores are adjusted by selecting scoring adjustment parameters of corresponding angles according to a scoring adjustment parameter set of corresponding automobile grades, so that the pictures falling into an angle area with higher credibility play a greater role in scoring, and the modeling scores with higher confidence are obtained.
The invention has the beneficial effects that:
1) an automatic automobile exterior shape user scoring machine is established.
2) An automobile modeling multi-view database and a user scoring sample data set corresponding to the database are established.
3) And the automobile angle, the automobile grade and the automobile brand are accurately identified.
4) And (4) carrying out parameter calibration on the automobile pictures with different grades and different angles to obtain a high-confidence modeling score, wherein the result can be referred by designers and consumers.
Compared with the prior art, the invention has the following remarkable advantages:
1) the method adopts large-scale automobile modeling multi-view pictures of different styles, collects scoring samples of effective users corresponding to an automobile modeling multi-view database, comprehensively considers factors such as multi-dimensional attributes of the users and the number of evaluation of single automobile models, and ensures the reasonability of automobile scoring training data.
2) Aiming at the problem that small sample data and subjective evaluation forms are mainly adopted in the traditional automobile external shape evaluation process, multi-dimensional attributes of users are considered for big data, reasonable shape scores are obtained based on analysis and quantification of different user attributes, and certain reference significance is provided for designers to master the psychology of the users.
3) The models are respectively trained according to the automobile grades by adopting a deep learning regression method, the mapping relation between the automobile model score and the automobile model multi-view image is established, the dependence on artificial feature definition and extraction is avoided, and compared with other methods, the method is higher in robustness and better in precision.
4) The method can calibrate parameters of automobile modeling pictures of different automobile grades and different angles, construct an automatic automobile outer modeling evaluation machine, realize the automation of automobile outer modeling evaluation, and obtain a modeling score with high confidence level.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a diagram of a ResNet network residual block.
Fig. 3 is a schematic diagram of a weighted composite scoring mechanism based on user attribute analysis.
Fig. 4 is a weighted gaussian distribution valued weight factor.
Fig. 5 is a two-dimensional user attribute weight distribution diagram.
Fig. 6 is a graph of Sigmoid function.
Detailed Description
The following further describes a specific embodiment of the present invention with reference to the drawings and technical solutions.
A scoring method for the external shape of an automobile is shown in figure 1 and comprises the following steps:
(1) creating a multi-view automobile picture database for identifying automobile angles, automobile grades and automobile brands: arranging multi-view automobile pictures of different brands and different models, wherein the total number of samples is not less than 15000; the database comprises automobiles of all grades, the automobiles are classified according to the grades, the cars can be classified into A, B, C, D four grades, and the SUV can be classified into compact, medium and large SUV; the number of the covered automobile brands is not less than 30, the number of the single brand automobile types is not less than 10, and 50 pictures are uniformly sampled around a circle at the head-up angle of each automobile type. And respectively labeling according to the automobile angle, the automobile grade and the automobile brand, and dividing the data set into a 70% training set and a 30% verification set.
(2) And carrying out angle, brand and grade recognition training by adopting a deep learning classification network. For example, using a ResNet50 network, the full name of ResNet is ResidualNetwork, called residual network, and the residual blocks in the network are shown in fig. 2, and the residual network can be understood as adding short-cut connections (shortcutconnections) to the forward network. The design of the shortcut connection enables the original data to pass through the processing layers directly while the original data is processed by the following layers. The whole model does not increase the complexity due to the design of shortcut connection. And respectively carrying out angle recognition training, brand recognition training and grade recognition training on the data in the multi-view automobile picture database to obtain an automobile angle recognition model, an automobile brand recognition model and an automobile grade recognition model.
(3) The method comprises the steps of creating an automobile model multi-view database, collecting large-scale automobile model multi-view pictures of different grades and different brands, dividing the database into 8 types according to cars (A, B, C, D four grades) and SUVs (compact, medium-sized and large-sized), enabling the number of models of each automobile grade to be not less than 30, and enabling each automobile type to uniformly sample 30 pictures around a circle at a head-up angle. And marking the database picture by using an automobile angle identification model, an automobile grade identification model and an automobile brand identification model.
(4) And creating an evaluation database corresponding to the modeling multi-view database, and collecting a comment sample of the effective user corresponding to the modeling multi-view database. Obtaining the model score and user information of the model multi-view database vehicle model, such as vehicle information, appearance evaluation score, user speciality, user objectivity and the like, and collecting user score samples corresponding to the automobile model multi-view database one by one.
(5) And processing the collected user scoring database to obtain a reasonable comprehensive scoring mechanism (refer to fig. 3). The collected appearance assessment scores are first subjected to a data outlier filtering process. The Grabbs test method is a method for removing abnormal values of user evaluation data, which is commonly used in statistics, and the statistics of the Grabbs test method are as follows:
where s is the standard deviation of the samples,for sampling, the average value is calculated according to a formula to obtain each sampling value x i Comparing the G value with the Grabbs critical table determined according to the sample number and the confidence coefficient, and if the obtained G value is larger than the table median value, considering the x i It is excluded as an outlier.
(6) And quantifying the attributes of the user, deciding that the user has multiple dimension attributes, and converting the measurement of the various dimension attributes into a weight factor. The weight dereferencing principle is adopted as follows: the higher the user dimension attribute value is, the higher the weight factor is; the weight factor can select a discrete rational number and can also select a continuous function value; values outside 3 σ (standard deviation) are 0; it is ensured as far as possible that the lowest level 3 sigma is not 0. For example, the confidence weight and the expertise of the user satisfy a normal distribution, and the weight factor is considered to be obtained by sampling the probability distribution density function, so that a one-dimensional weighted gaussian distribution function is defined:
wherein W is an initial weighting factor; x is a user attribute quantization value; σ is the standard deviation; mu is the highest user professional quantization factor; a is the amplitude elastic coefficient;b controls the degree of flatness of the function. As shown in fig. 4, assume that the parameters σ -1, μ -3,b ═ 2ln1.5, the abscissa represents the user attribute, and the ordinate represents the calculated weight. Assuming that the weighting factor is selected as a discrete value, the parameters are {0,0.5,2,3}, which are the points of circles in fig. 4. Considering a plurality of independent user attribute dimensions, a one-dimensional weighted gaussian distribution function can be generalized to a multi-dimensional situation:
parameter values can be set, weight factors are calculated through formulas, and attribute quantization is carried out on the user. In practical application, taking two independent user attribute dimensions as an example, the following formula can be adopted:
desirable parameter sigma 1 =σ 2 =1,μ 1 =μ 2 =3,B 1 =B 2 The user attribute weight factor is calculated as 2ln1.5, and fig. 5 is a two-dimensional user attribute weight distribution diagram. For a certain vehicle model, the number of estimated users is assumed to be n, and any user X i The weight after attribute quantization is:
the weighted average score R for this vehicle type is:wherein P is i Is a user X i And (5) scoring the vehicle type.
(7) And (4) performing comprehensive weighted average on the results in the step (6), wherein the more the number of the evaluated persons of different vehicle types in the same vehicle grade is, the closer the Bayesian average is to the arithmetic average, and the smaller the influence on the score is. The formula is adopted:
wherein WR is the final weighted score, and n is the number of people evaluated in a certain vehicle type; m is the minimum number of people to be evaluated in all vehicle types under the same grade; r is the average score (obtained by weight calculation) of the vehicle type; and C is the average score calculated after the weighted average scores of all the vehicle types under the same vehicle grade. And marking the obtained final weighting scores of the automobile models on the multi-view database of the automobile models one by one, and dividing the data set into a 70% training set and a 30% verification set.
(8) And (3) extracting the features by adopting a feature extraction network, normalizing the automobile model score labels to be between 0 and 1, connecting a Sigmoid function at the last layer of the network to perform logistic regression calculation, wherein the Sigmoid function is as shown in figure 6, and fitting the image features to model score points on a Sigmoid function curve after the image features are extracted, so that the trained model can predict and output automobile model scores. In order to obtain a network model with stronger expression capability, the automobile model multi-view database picture is divided into eight sub-databases according to the automobile grade, and deep learning regression training is carried out to obtain different models. For example, shallow neural network ResNet8 is used for feature extraction.
The parameters are as follows:
the settings for the training hyper-parameters are chosen as follows: epoch is 50; training batch is 16; the initial learning rate is 0.01, and when the epoch is in the range of 31-40, the learning rate is adjusted to 0.001, and when the epoch is in the range of 41-50, the learning rate is adjusted to 0.0001; momentum (momentum) is 0.9 and attenuation (weight) is 0.0005.
(9) And (3) establishing an automatic evaluation machine for the external shape of the automobile, and calculating automobile grading adjustment parameters under different grades after obtaining the trained shape grading model. The method comprises the steps of carrying out scoring prediction on automobile model multi-view database pictures under a certain grade, measuring the accuracy of the prediction scoring by using MAE (mean absolute error), obtaining scoring data statistics of different angles of all automobile models under the same automobile grade, and sorting and manufacturing an automobile model scoring statistical table. Selecting the part of angles with the lowest MAE of each vehicle type score, for example, selecting the lowest angle, counting in a table, obtaining the times of occurrence of each angle, and expressing the times in a set formCalculating a grading adjustment parameter of each angle, namely the frequency of occurrence of each angle:the angle with the highest frequency shows that the scoring credibility is higher at the position, so the automobile scoring adjustment parameter set at the grade is combinedThe automobile grading adjustment parameters under different grades are different, and the pictures of each sub-database are predicted and calculated respectively.
(10) By utilizing the automobile outer shape evaluation machine, the scoring prediction can be made for the given automobile picture. After a single-view automobile model picture is given, the grade of an automobile is identified, a corresponding scoring model is selected, the model score of the automobile picture is obtained through testing and is x, a standard deviation sigma is obtained through an automobile model score statistical table under the grade, the automobile model score is obtained, for example, the score confidence interval when the confidence coefficient reaches 95% is [ x-2 sigma, x +2 sigma ]](ii) a After a plurality of automobile model pictures are given, the angle of the automobile model picture is identified (the picture angle is marked as theta) 1 ,θ 2 ,…,θ k ) Identifying the automobile grade of the automobile, selecting a corresponding grade prediction model, and obtaining the shape grades of the pictures with different anglesThe average score can be calculated:
the grading adjustment parameter set corresponding to the automobile grade is utilized, the grading adjustment parameter corresponding to the angle is selected to adjust the grading, so that the picture falling into the angle area with higher credibility plays a greater role in grading, and the formula is as follows:
where X is the higher confidence build score.
Claims (1)
1. An automatic scoring method for an automobile exterior shape is characterized by comprising the following steps:
(1) creating a multi-view automobile picture database for identifying automobile angles, automobile grades and automobile brands: arranging multi-view automobile pictures of different brands and different models, wherein the total number of samples is not less than 15000; the database comprises automobiles of all grades, the automobiles are classified according to the grades, the cars are classified into A, B, C, D four grades, and the SUVs are classified into compact, medium and large SUVs; the number of the covered automobile brands is not less than 30, the number of the single brand automobile types is not less than 10, and 50 pictures are uniformly sampled around a circle at the head-up angle of each automobile type; respectively labeling according to automobile angles, automobile grades and automobile brands, and dividing a data set into a 70% training set and a 30% verification set;
(2) respectively carrying out angle recognition training, brand recognition training and grade recognition training on data in the multi-view automobile picture database to correspondingly obtain an automobile angle recognition model, an automobile brand recognition model and an automobile grade recognition model so as to recognize the angle, the grade and the brand of any given automobile picture;
(3) creating an automobile model multi-view database, collecting large-scale automobile model multi-view pictures of different grades and different brands, wherein the database is divided into 8 types according to saloon cars (A, B, C, D four grades) and SUVs (compact, medium-sized and large-sized), the number of models of each automobile grade is not less than 30, and 30 pictures are uniformly sampled in each automobile type around a circle at a head-up angle; marking the database picture by using an automobile angle identification model, an automobile grade identification model and an automobile brand identification model;
(4) establishing an evaluation database corresponding to the modeling multi-view automobile picture database, and collecting comment samples of effective users corresponding to the modeling multi-view automobile picture database, wherein the comment samples comprise modeling scores and user information;
(5) performing data outlier filtering processing on the collected modeling scoring data; removing outliers by adopting a Grabas method, checking whether the data have lower side outliers, processing the outliers and then calculating;
(6) in order to accurately analyze the credibility of a user participating in evaluation, quantize the attributes of the user, examine and quantize the multidimensional attributes of the user, and convert the measurement of various dimensional attributes into weight factors by adopting a multidimensional weighting Gaussian distribution function; then, calculating the weight of the user after quantization, and further calculating the weighted average score of the corresponding vehicle type;
(7) performing comprehensive weighted average on the result obtained in the step (6), and calculating a final weighted score by adopting a Bayesian average algorithm; marking the obtained final automobile model weighting scores on an automobile model multi-view database one by one, and dividing a data set into a 70% training set and a 30% verification set;
(8) adopting a deep learning regression method to establish a mapping relation between automobile modeling weighted score and an automobile modeling multi-view picture, normalizing an automobile modeling score label to be between 0 and 1, connecting a Sigmoid function on the last layer of a network to perform logistic regression calculation, extracting picture characteristics, and fitting to a modeling score point on a Sigmoid function curve to enable a trained model to predict and output an automobile modeling score; in order to obtain a network model with stronger expression capability, dividing an automobile model multi-view database picture into eight sub data sets according to automobile grades, and training to obtain different models;
(9) establishing an automatic evaluation machine for the external shape of the automobile, and calculating automobile grading adjustment parameters under different grades after obtaining a trained shape grading model; grading and predicting the automobile model multi-view database pictures at a certain grade, measuring the precision of the prediction grading by using MAE (maximum intensity extraction), obtaining the statistics of grading data of different angles of each automobile model at the same automobile grade, and sorting and manufacturing an automobile model grading statistical table; selecting the lowest part of angles of MAE of each vehicle type score, counting the angles in a table to obtain the occurrence frequency of each angle, calculating score adjusting parameters of each angle, namely the occurrence frequency of each angle, wherein the angle with the highest frequency shows that the score credibility at the position is higher, the vehicle score adjusting parameters under different grades are different, and predicting and calculating the pictures of each sub-database respectively;
(10) utilizing an automobile external modeling evaluation machine to make scoring prediction on a given automobile picture; after a single-view automobile model picture is given, identifying the grade of an automobile, selecting a corresponding scoring model, testing to obtain the model score x of the automobile picture, and obtaining a standard deviation sigma by using an automobile model score statistical table under the grade to obtain the automobile model score; after a plurality of automobile model pictures are given, the angles of the automobile model pictures are identified, the automobile grade of the automobile is identified, a corresponding grade prediction model is selected to obtain the model grades of the different angle pictures, the grade adjustment parameter set corresponding to the automobile grade is utilized, the grade adjustment parameters corresponding to the angles are selected to adjust the grade, so that the pictures falling into the angle area with higher credibility play a greater role in the grade, and the model grade with higher confidence coefficient is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011127986.0A CN112258472B (en) | 2020-10-20 | 2020-10-20 | Automatic scoring method for automobile exterior shape |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011127986.0A CN112258472B (en) | 2020-10-20 | 2020-10-20 | Automatic scoring method for automobile exterior shape |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112258472A CN112258472A (en) | 2021-01-22 |
CN112258472B true CN112258472B (en) | 2022-09-20 |
Family
ID=74245230
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011127986.0A Active CN112258472B (en) | 2020-10-20 | 2020-10-20 | Automatic scoring method for automobile exterior shape |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112258472B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113434628B (en) * | 2021-05-14 | 2023-07-25 | 南京信息工程大学 | Comment text confidence detection method based on feature level and propagation relation network |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105975941B (en) * | 2016-05-31 | 2019-04-12 | 电子科技大学 | A kind of multi-direction vehicle detection identifying system based on deep learning |
CN106096144B (en) * | 2016-06-13 | 2019-04-09 | 大连理工大学 | A kind of automobile brand genetic analysis method based on preceding face moulding |
CN109408809A (en) * | 2018-09-25 | 2019-03-01 | 天津大学 | A kind of sentiment analysis method for automobile product comment based on term vector |
CN109766808A (en) * | 2018-12-29 | 2019-05-17 | 南通贝法瑞信息科技有限公司 | A kind of AI algorithm carrying out automobile mark and identification based on two-dimensional tag |
CN110415071B (en) * | 2019-07-03 | 2024-02-27 | 西南交通大学 | Automobile competitive product comparison method based on viewpoint mining analysis |
CN110533093A (en) * | 2019-08-24 | 2019-12-03 | 大连理工大学 | A kind of automobile front face brand family analysis method |
-
2020
- 2020-10-20 CN CN202011127986.0A patent/CN112258472B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN112258472A (en) | 2021-01-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108765412B (en) | Strip steel surface defect classification method | |
CN109597968B (en) | SMT big data-based solder paste printing performance influence factor analysis method | |
CN110543616B (en) | SMT solder paste printing volume prediction method based on industrial big data | |
CN110135459B (en) | Zero sample classification method based on double-triple depth measurement learning network | |
CN109977808A (en) | A kind of wafer surface defects mode detection and analysis method | |
CN111738363B (en) | Alzheimer disease classification method based on improved 3D CNN network | |
CN111338950A (en) | Software defect feature selection method based on spectral clustering | |
CN110598299A (en) | Surface integrity evaluation method based on sensor fusion and deep learning | |
CN116883393B (en) | Metal surface defect detection method based on anchor frame-free target detection algorithm | |
CN113601261B (en) | Monitoring method of online rapid optimization model for cutter | |
CN112258472B (en) | Automatic scoring method for automobile exterior shape | |
CN114897802A (en) | Metal surface defect detection method based on improved fast RCNN algorithm | |
CN112529638A (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
CN117315380B (en) | Deep learning-based pneumonia CT image classification method and system | |
CN115456693A (en) | Automatic evaluation method for automobile exterior design driven by big data | |
CN112200252A (en) | Joint dimension reduction method based on probability box global sensitivity analysis and active subspace | |
CN117392450A (en) | Steel material quality analysis method based on evolutionary multi-scale feature learning | |
CN116738332A (en) | Aircraft multi-scale signal classification recognition and fault detection method combining attention mechanism | |
CN117078960A (en) | Near infrared spectrum analysis method and system based on image feature extraction | |
CN112804650B (en) | Channel state information data dimension reduction method and intelligent indoor positioning method | |
CN115630332A (en) | Wheat flour quality characteristic prediction method | |
CN115206455A (en) | Deep neural network-based rare earth element component content prediction method and system | |
CN114596296A (en) | High-sensitivity hot-rolled steel coil end surface defect identification system and method | |
WO2024113340A1 (en) | Part defect detection method based on unsupervised transfer learning | |
CN113343787B (en) | Deep learning-based medium-level assessment method suitable for map contrast scene |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |