A vehicle year model screening method, device, and storage medium

Abstract

The application provides a vehicle year model screening method and device, a storage medium, and a vehicle year model screening method. The method comprises the following steps: obtaining vehicle attributes, calculating a vehicle attribute loss function according to the vehicle attributes; obtaining labeled year model data, calculating the features of the labeled year model data; generating a year model pseudo label according to the year model branch features, and calculating a loss function of the year model branch; constructing a vehicle year model recognition model based on a deep neural network according to the vehicle attribute function and the loss function of the year model branch; obtaining a vehicle image to be recognized, and screening the vehicle image according to the vehicle year model recognition model. The application combines vehicle-specific attributes to predict the pseudo label of an unlabeled vehicle image, thereby improving the accuracy of vehicle year model screening.

Description

Technical Field

[0001] This application relates to the technical field, specifically to a method, device, and storage medium for selecting vehicle model year. Background Technology

[0002] Structured analysis of vehicles requires filtering out different models from different years from captured vehicle images to update the model year recognition. Existing literature includes methods for filtering new categories using clustering. This involves supervised training of a feature extraction model on a labeled set, then extracting features from an unlabeled dataset to perform clustering and filter new categories. However, this approach weakens the strong semantic similarity between labeled and unlabeled classes.

[0003] Existing technologies also include methods for training a single network to perform new class selection. These methods involve performing two different enhancements on an unlabeled image to generate two views, using the predicted value from one view as the pseudo-label for the other, thus training the classification network. While this method is suitable for distinguishing various categories of data, it is less effective for vehicle model year selection tasks due to the large number of model year categories, resulting in computational difficulties and a need for improvement in recognition accuracy. Summary of the Invention

[0004] To address the aforementioned issues, embodiments of this application provide a method, apparatus, and storage medium for screening vehicle model years. By combining vehicle-specific attributes to predict pseudo-labels for unlabeled vehicle images, the accuracy of screening new vehicle model year categories is improved.

[0005] Therefore, one aspect of this application provides a method for selecting vehicle model year, the method comprising the following steps:

[0006] Obtain vehicle attributes and calculate the vehicle attribute loss function based on the vehicle attributes;

[0007] Acquire labeled model year data and calculate the characteristics of the labeled model year data;

[0008] Generate pseudo-labels for the year based on the year branch features, and calculate the loss function for the year branch;

[0009] Based on the vehicle attribute function and the loss function of the model year branch, a vehicle model year recognition model based on a deep neural network is constructed.

[0010] Obtain an image of the vehicle to be identified, and then filter the vehicle image by year using the vehicle year recognition model.

[0011] Furthermore, the calculation of the features of the marked year data specifically involves,

[0012] Acquire training data, which are pre-generated pseudo-labels; pass the training data through a first image enhancement module and a second image enhancement module to obtain a first training view and a second training view; input the first training view and the second training view into the backbone network to obtain a first year feature and a second year feature.

[0013] Further, in step S3, a pseudo-label for the year is generated based on the year branch features, and the loss function for the year branch is calculated, specifically as follows:

[0014] Obtain a first training view and a second training view, and use the first training view and the second training view to calculate and generate a first pseudo label and a second pseudo label through an iterative sinkhorn-knopp algorithm;

[0015] Calculate the loss function for the year branch based on the first pseudo-label and the second pseudo-label.

[0016] Furthermore, the loss function of the year branch includes a first view training loss function and a second training view loss function, wherein the first view loss function is related to the first pseudo-label and the second view loss function is related to the second pseudo-label.

[0017] Furthermore, the loss function of the deep neural network is the sum of the vehicle attribute loss function and the model year branch loss function.

[0018] Furthermore, the vehicle attributes include vehicle orientation, model, and brand; the calculation of the vehicle attribute loss function based on the vehicle attributes specifically involves,

[0019] Acquire training data, which is a dataset labeled with model year, vehicle orientation, model, and brand;

[0020] The dataset, labeled with model year, vehicle orientation, model, and brand, is sequentially processed through the image enhancement module to obtain a view, which is then input into the backbone network to obtain features.

[0021] The features are input into a fully connected layer, and the loss function is calculated for the labeled year, vehicle orientation, model, and brand.

[0022] Furthermore, the first year feature f1 m for:

[0023] f1 m =M(z1)

[0024] Where z1 is a feature, z1 = B(D) a1 ), D a1 This is the first training view.

[0025] Furthermore, the formula for calculating the loss function of the first training view is as follows:

[0026]

[0027] in, Let k be the first pseudo-label, and k be the category of the feature. Let i be the i-th dimension feature in the first year's features; and Where t is a constant, This is the i-th dimension of the features in the second year's designation;

[0028] The formula for calculating the loss function of the second training view is:

[0029]

[0030] In another aspect of this application, an apparatus is provided that includes a plurality of processors, a memory, and a computer program stored in the memory and executable on the processors, wherein the processors, when executing the computer program, implement a vehicle model year selection method as described above.

[0031] In another aspect, this application provides a storage medium having a computer program stored thereon, which, when executed, implements a vehicle model year selection method as described above.

[0032] As described above, this application provides a method, device, and storage medium for screening vehicle model years. It employs a single network for screening and combines vehicle attributes and vehicle attribute loss functions with model year features and model year feature loss functions to construct a vehicle recognition model based on a deep neural network. By using the above vehicle recognition model to screen new model year categories, it reduces a significant amount of manual screening costs, simplifies the screening process, and achieves high accuracy in screening new vehicle model year categories.

[0033] V. Scope of application of the present invention.

[0034] The above summary provides a simplified overview of some concepts, which will be further described in detail in the following specific embodiments. The above summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to help determine the scope of the claimed subject matter. The claimed subject matter is not limited to embodiments that address any or all the shortcomings pointed out in the background art. Attached Figure Description

[0035] The accompanying drawings, incorporated in and forming part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without any creative effort. These drawings and textual descriptions are not intended to limit the scope of the concept of this application in any way, but rather to illustrate the concepts of this application to those skilled in the art by referring to specific embodiments.

[0036] Figure 1 This is a flowchart illustrating the embodiments provided in this application;

[0037] Figure 2 The deep neural network diagram provided in this application. Detailed Implementation

[0038] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0039] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0040] It should be understood that although the terms first, second, third, etc., may be used herein to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this document, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if," as used herein, may be interpreted as "when," "when," or "in response to determination." Furthermore, as used herein, the singular forms "a," "an," and "the" are intended to also include the plural forms unless the context indicates otherwise. It should be further understood that the terms "comprising," "including," indicate the presence of the stated feature, step, operation, element, component, item, kind, and / or group, but do not exclude the presence, occurrence, or addition of one or more other features, steps, operations, elements, components, items, kinds, and / or groups. The terms "or," "and / or," "including at least one of the following," etc., as used in this application, may be interpreted as inclusive, or mean any one or any combination thereof. An exception to this definition will only occur if the combination of elements, functions, steps, or operations is inherently mutually exclusive in some way.

[0041] It should be understood that although the steps in the flowcharts of this application's embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0042] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0043] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0044] Please see Figure 1 , 2This application provides a method for screening vehicle model years, which combines vehicle-specific attributes to predict pseudo-labels for unlabeled vehicle images, thereby improving the accuracy of screening new vehicle model year categories.

[0045] Specifically, this method includes the following steps:

[0046] Step S1: Obtain vehicle attributes and calculate the vehicle attribute loss function based on the vehicle attributes.

[0047] Vehicle attributes include vehicle orientation, vehicle type, and brand. Vehicle type refers to the vehicle's model, such as sedan, SUV, or MPV. Vehicle orientation refers to the direction the vehicle is pointing, and brand refers to the vehicle's series or model number. Identifying vehicle attributes assists in the classification of vehicle recognition models and improves the accuracy of identifying vehicle model years. Calculating the vehicle attribute loss function requires separately calculating loss functions for vehicle orientation, vehicle type, and brand.

[0048] First, obtain training data D, which is a dataset labeled with the year, vehicle orientation, model, and brand.

[0049] The training data D, which consists of datasets labeled with model year, vehicle orientation, vehicle type, and brand, is processed through the augmentation image enhancement module to obtain view D. a The input is fed into the backbone network backone and labeled as B, resulting in feature z, as shown in formula (1):

[0050] z = B(D) a (1)

[0051] Input the feature z into a fully connected layer and calculate the loss function for the labeled year, vehicle orientation, model, and brand.

[0052] Specifically, taking a brand as an example, the feature z is input to obtain f. b As shown in formula (2):

[0053] f b =B(z) (2)

[0054] The vehicle's brand is labeled as data BL. n For category n, calculate the cross-entropy loss using the following formula:

[0055]

[0056] This leads to the brand's loss function, denoted as l. b As shown in formula (3):

[0057] l b =CrossEntropyLoss(fb ,l b (3)

[0058] Similarly, the loss function l for vehicle orientation is calculated using the method described above. d The loss function of the vehicle model l t .

[0059] Step S2: Obtain the marked year data and calculate the characteristics of the marked year data.

[0060] The model year data consists of pre-acquired vehicle model year data, labeled on the vehicle images. This labeled model year data is used as training data D, which contains k classes. It is worth noting that in this application, during the training process, the training data consists of generated pseudo-labels, without using real labels. The pre-generated pseudo-labels contain generated vehicle model year data, with a complete and even distribution of types and quantities, which is more conducive to training this vehicle model year selection model.

[0061] The training data D is processed through the first image enhancement module of augmentation1 to obtain the first training view D. a1 The input is fed into the backbone network backone to obtain feature z1, as shown in formula (4):

[0062] z1=B(D a1 (4)

[0063] Inputting feature z1 into the year-specific fully connected layer (Model Unlabeled Head), with category k, yields the first year-specific feature f1. m As shown in formula (5):

[0064] f1 m =M(z1) (5)

[0065] Similarly, the training data D is processed through the augmentation2 second image enhancement module to obtain the second training view D. a2 The second-year model features are obtained by connecting the backbone network backone and the model year fully connected layer.

[0066] Step S3: Generate pseudo-labels for the year based on the year branch features, and calculate the loss function for the year branch.

[0067] The loss function of the year branch includes a first training view loss function and a second training view loss function. Therefore, the first training view and the second training view obtained in the above steps are obtained, and the first training view and the second training view are used to generate a first pseudo-label and a second pseudo-label through an iterative sinkhorn-knopp algorithm; the first training view D a1 The pseudo-tags are from After calculation by the sinkhorn-knopp module, it is marked as

[0068] First pseudo-label The calculation formulas are shown in formulas (6) and (7):

[0069]

[0070]

[0071] This is the first pseudo-label, where t is a constant. For the i-th dimension of the second year's features, as shown in the above formula, the first pseudo-label... Features of the second year Generate, marked as Features of the first year model f1 m Relatedly, the first training view D a1 The loss function is labeled as l a1 The calculation formula is shown in formula (8):

[0072]

[0073] View D a2 The pseudo-tags are from f1 m After calculation by the sinkhorn-knopp module, it is marked as The calculation method is the same as in view D. a1 Second training view D a2 The loss function is labeled as l a2 The calculation formula is shown in (9):

[0074]

[0075] Step S4: Construct a vehicle year recognition model based on a deep neural network according to the loss function of the vehicle attribute function and the year branch;

[0076] like Figure 2 The figure shows the deep neural network block diagram of the vehicle model year recognition model. As shown in the figure, the loss function of the deep neural network is the sum of the vehicle attribute loss function and the loss function of the model year branch, as shown in formula (10):

[0077] l = l a1 +l a2 +l b +l d +l t (10)

[0078] Step S5: Obtain the image of the vehicle to be identified, and filter the vehicle image by year using the vehicle year recognition model.

[0079] The vehicle model year to be identified is used to filter the vehicle model year based on the vehicle's orientation, brand, model, and other features, and outputs a new model year category for the vehicle.

[0080] This application provides a method for screening vehicle model years. It employs a single network for screening and combines vehicle attributes and vehicle attribute loss functions with model year features and model year feature loss functions to construct a vehicle recognition model based on a deep neural network. By using the above vehicle recognition model to screen new model year categories, it reduces a significant amount of manual screening costs, simplifies the screening process, and achieves high accuracy in screening new model year categories.

[0081] Based on the same inventive concept, the present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the aforementioned method for batch setting of smart home devices.

[0082] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in computer-readable media can be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0083] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0084] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0085] In this application, the same or similar terms, concepts, technical solutions and / or application scenario descriptions are generally described in detail only when they appear for the first time. When they appear again, they are generally not repeated for the sake of brevity. When understanding the technical solutions and other contents of this application, the same or similar terms, concepts, technical solutions and / or application scenario descriptions that are not described in detail later can be referred to their previous relevant detailed descriptions.

[0086] In this application, the descriptions of the various embodiments have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0087] The technical features of the present application can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present application.

[0088] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for selecting vehicle model year, characterized in that, This method includes the following steps: Obtain vehicle attributes and calculate the vehicle attribute loss function based on the vehicle attributes; Acquire labeled model year data and calculate the characteristics of the labeled model year data; Based on the year-based branch features, pseudo-labels for the year-based designations are generated, and the loss function for each year-based branch is calculated; specifically, Obtain a first training view and a second training view, and use the first training view and the second training view to calculate and generate a first pseudo label and a second pseudo label through an iterative sinkhorn-knopp algorithm; Based on the first pseudo-label and the second pseudo-label, the loss function of the year branch is calculated; the loss function of the year branch includes a first view loss function and a second view loss function, wherein the first view loss function is related to the first pseudo-label and the second view loss function is related to the second pseudo-label; A vehicle model year recognition model based on a deep neural network is constructed based on the vehicle attribute loss function and the model year branch loss function; the loss function of the deep neural network is the sum of the vehicle attribute loss function and the model year branch loss function. Obtain an image of the vehicle to be identified, and then filter the vehicle image by year using the vehicle year recognition model.

2. The method for selecting vehicle model year as described in claim 1, characterized in that, The calculation of the characteristics of the marked year data specifically involves: Acquire training data, wherein the training data is pre-labeled data; The training data is processed through a first image enhancement module and a second image enhancement module to obtain a first training view and a second training view. The first training view and the second training view are then input into the backbone network to obtain a first year feature and a second year feature.

3. The method for selecting vehicle model year as described in claim 1, characterized in that, The vehicle attributes include vehicle orientation, model, and brand; the calculation of the vehicle attribute loss function based on the vehicle attributes specifically involves... Acquire training data, which is a dataset labeled with model year, vehicle orientation, model, and brand; The dataset, labeled with model year, vehicle orientation, model, and brand, is sequentially processed through the image enhancement module to obtain a view, which is then input into the backbone network to obtain features. The features are input into a fully connected layer, and the loss function is calculated for the labeled year, vehicle orientation, model, and brand.

4. The method for selecting vehicle model year as described in claim 2, characterized in that, First year features for: Where z1 is a feature, z1 = B(D) a1 ), D a1 This is the first training view.

5. The method for selecting vehicle model year as described in claim 4, characterized in that, The formula for calculating the first view loss function is as follows: in, Let k be the first pseudo-label, and k be the category of the feature. Let i be the i-th dimension feature in the first year's features; and Where t is a constant, This is the i-th dimension of the features in the second year's designation; The formula for calculating the second-view loss function is:

6. A device, characterized in that, It includes multiple processors, a memory, and a computer program stored in the memory and executable on the processors. When the processor executes the computer program, it implements a method for selecting vehicle model year as described in any one of claims 1 to 5.

7. A storage medium, characterized in that, It stores a computer program, which, when executed, implements a method for selecting vehicle model year as described in any one of claims 1 to 5.

Images