CN105844289A - Automobile charging interface identification method - Google Patents

Abstract

The vehicle charging interface identification method disclosed in the present invention mainly includes: acquiring a plurality of vehicle appearance images, respectively constructing feature vectors of the vehicle appearance images; performing regression classification on the feature vectors, summarizing the car models, Obtain a pre-classification model; select a number of car models and car appearance pictures corresponding to the car model, train the pre-classification model, and obtain a classification model; establish the corresponding relationship between the car model and the charging interface type in advance; obtain the car model to be judged According to the classification model, it is judged that the car shape image of the vehicle type to be determined corresponds to the model of the car; according to the corresponding relationship and the recognized car model, the charging interface type of the car is obtained. The automobile charging interface identification method provided by the present invention is helpful for efficiently and accurately judging the automobile charging interface, and is conducive to selecting an accurate charging gun.

Description

A method for identifying a car charging interface

technical field

The invention relates to the technical field of information identification, and more specifically, relates to an identification method for an automobile charging interface.

Background technique

Car charging is based on the charging interface of the car body to select the matching charging gun interface. The wrong choice of charging gun may damage the car's power storage system or charging interface, affecting the use of the car. However, at present, there are various charging interfaces for automobiles. When charging in a parking lot with charging services, it is necessary to manually select a matching vehicle charging interface from a large number of charging interfaces. Manual judgment and selection will consume a lot of manpower and take a long time, which will affect the charging efficiency.

It can be seen that how to efficiently and accurately judge the judgment of the charging interface of the car and select an accurate charging gun is an urgent problem to be solved by those skilled in the art.

Contents of the invention

The object of the present invention is to provide a method for identifying the charging port of a car, which is helpful for efficient and accurate judgment of the charging port of a car, and is conducive to the selection of an accurate charging gun.

In order to solve the above technical problems, the present invention provides the following technical solutions:

A method for identifying an automobile charging interface provided by the present invention, the method comprising:

Obtaining several car shape images, respectively constructing feature vectors of the car shape images;

Wherein, the constructing the feature vector mainly includes extracting the car shape image, performing the first convolution operation on the car shape image to obtain the first feature image,

performing the first pooling operation on the first feature image to obtain a second feature image,

performing a second convolution operation on the second feature image to obtain a third feature image,

performing a second pooling operation on the third feature image to obtain a fourth feature image,

performing a third convolution operation on the fourth feature image to obtain a fifth feature image,

performing a third pooling operation on the fifth feature image to obtain a sixth feature image,

Convert the sixth feature image into the feature vector through the feature vector conversion formula Y _l (n)=∑W _l (n, m)*Y _l-1 (m), wherein, l is the lth feature image , W _l is the weight, Y _l (n) represents the feature vector, Y _l-1 (m) is the feature image before vector transformation;

Carrying out regression classification on the feature vector, summarizing the car models to obtain a pre-classification model;

Select a number of car models and car outline pictures corresponding to the car models, and train the pre-classification model to obtain a classification model;

Pre-establish the corresponding relationship between the car model and the charging interface type;

Obtaining the car shape image of the vehicle type to be judged, and judging the model of the car corresponding to the car shape image of the car model to be judged by the classification model;

According to the corresponding relationship and the identified car model, the charging interface type of the car is obtained.

Preferably, in the above-mentioned method for identifying an automobile charging interface, in the first convolution operation, the size of the convolution kernel is 5*5, and the number of the first feature images is 5; the second convolution operation , the size of the convolution kernel is 5*5, and the number of the third feature images is 10; in the third convolution operation, the size of the convolution kernel is 4*4, and the number of the fifth feature images is 16.

Preferably, in the above-mentioned method for identifying an automobile charging interface, the first pooling operation specifically includes summing the 2*2 neighborhoods of the first feature image, multiplying them by random weights, adding a bias, and then Perform neuron activation, the neuron activation is logistic, and the logistic activation function is

Preferably, in the above identification method for the charging interface of a car, the parameters of each layer are adjusted, and the adjustment of the parameters of each layer specifically includes calculating a loss function, performing reverse derivation of the loss function, and adjusting the parameters of each layer to minimize the loss function .

Preferably, in the above-mentioned method for identifying a charging port of a car, the image of the car's appearance includes an image containing a charging port and an image not containing a charging port.

The automobile charging interface identification method provided by the present invention obtains several automobile appearance images, performs convolution, pooling, full connection and regression analysis on the automobile appearance images, establishes a pre-classification model of automobile models, and selects several automobile appearance pictures of known automobile models , to train the pre-classification model to obtain a classification model with higher accuracy. Utilize the obtained classification model to judge the car shape image of the car model to be judged, obtain the car model corresponding to the car shape image, and determine the charging port type through the corresponding relationship between the car model and the charging port type established in advance. judgment.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings that need to be used in the description of the embodiments will be briefly introduced below. Other drawings can also be obtained based on these drawings.

Fig. 1 is a flow chart of a method for identifying an automobile charging interface provided by an embodiment of the present invention;

Fig. 2 is a schematic diagram of feature vector conversion provided by an embodiment of the present invention;

Fig. 3 is a diagram of the corresponding relationship between the vehicle model and the charging interface type provided by the embodiment of the present invention.

detailed description

The automobile charging interface identification method provided by the embodiment of the present invention is helpful for efficient and accurate judgment of the automobile charging interface, and is conducive to selecting an accurate charging gun.

In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, and to make the above-mentioned purposes, features and advantages of the embodiments of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention are described below in conjunction with the accompanying drawings The program is described in further detail.

With reference to accompanying drawing 1, this figure shows the basic flow of the method for identifying the vehicle charging interface provided by the embodiment of the present invention, specifically including:

S101: Acquire several car appearance images, and respectively construct feature vectors of the car appearance images.

A number of car appearance pictures are selected, wherein the car appearance images include various models of cars and charging ports on the market as comprehensively as possible. The format and size of the car shape image are different. In order to ensure the convenience of calculation and processing, the car shape image will be obtained for processing. In this embodiment, the car shape image is preferably unified into a 64*64 grayscale image, but it is not limited to Therefore, taking into account the conditions of computing and hardware processing equipment, you can choose by yourself. In this embodiment, the car appearance image may include a car appearance image including a charging interface and a car appearance image not including a charging interface. In order to ensure the convenience of data processing, the image of the car shape including the charging interface is generally selected.

After the selection of the picture is completed, the picture is subjected to convolution operation, pooling operation, convolution operation, pooling operation, convolution operation, pooling operation, and full connection to obtain the feature vector of the shape image respectively. The convolution operation and the pooling operation are a downsampling method, in order to extract the features of the car shape image more conveniently. Specific operation: extract the car shape image, perform the first convolution operation on the car shape image to obtain the first feature image; perform the first pooling operation on the first feature image to obtain the second feature image ; performing a second convolution operation on the second feature image to obtain a third feature image; performing a second pooling operation on the third feature image to obtain a fourth feature image; Perform the third convolution operation to obtain the fifth feature image; perform the third pooling operation on the fifth feature image to obtain the sixth feature image; use the feature vector conversion formula Y _l (n)=∑W _l ( n, m)*Y _l-1 (m) converts the sixth feature image into the feature vector, wherein, l is the lth feature image, W _l is a weight, Y _l (n) represents a feature vector, and Y _l-1 (m) is the feature image before vector transformation. In the present invention, the sixth characteristic image is selected, but not only the sixth characteristic image can be selected according to actual needs. The number of convolution operations and pooling operations can also be adjusted according to actual needs.

The application also provides a specific embodiment, as shown in Figure 2, the specific process is as follows:

Taking a car shape image as an example, the first convolution operation is performed on the car shape image that has been converted into a 64*64 grayscale image. The convolution kernel size is 5*5, and five 60*60 size images are obtained. The first feature image.

Perform the first pooling operation on five first feature images of 60*60 size to obtain five second feature images of 30*30 size. The first pooling operation filters out most of the relevant information, so that the image The contained information is more independent, reduces the amount of image information, and greatly reduces the amount of calculation. The preferred first pooling operation in the embodiment of the present invention is to sum the 2*2 neighborhoods of the first feature image respectively, then multiply by random weights and add bias, such as g(x,y) ^l =wf (x, y) + b, where g(x, y) ^l is the total, w is the weight, b is the upper bias, f(x, y) is the sum of the neighborhood, and finally the neuron activation operation is commonly used The activation functions include logistic activation, tanh activation and ReLu activation. In the embodiment of the present invention, the logistic activation function is preferably used for activation, and the logistic activation function is:

Pooling and convolution operations often map multiple feature images to multiple feature images. In order to avoid the feature images of the latter layer being convolved by all the pooled feature images of the previous layer, the amount of calculation is greatly increased. According to a certain encoding method in the present invention, the feature image of the latter convolutional layer is convoluted by several images of the previous pooling layer, but not all of them, which not only preserves the information contained in the original image, but also reduces the amount of calculation.

Perform the second convolution operation on 5 second feature images of 30*30 size, the convolution kernel size is 5*5, and obtain 10 third feature images of 26*26 size, the third feature image is obtained by the second Several feature maps of the feature image are jointly convolved, as shown in the following table, two indicates the second feature image, three indicates the third feature image, and X indicates the image used for the convolution operation.

Perform the second pooling operation on ten third feature images of size 26*26 to obtain ten fourth feature images of size 13*13. The second pooling operation is similar to the first pooling operation.

Perform the third convolution operation on ten fourth feature images of size 13*13, the size of the convolution kernel is 4*4, and obtain 16 fifth feature images of size 10*10, the third convolution operation and the first The second convolution operation is similar.

Perform the third pooling operation on the 16 fifth feature images with a size of 10*10 to obtain 16 sixth feature images of 5*5. The third pooling operation is similar to the first pooling operation.

After the above convolution and pooling operations, the obvious features of each car shape image can be extracted more clearly. It is convenient for vector conversion without losing image information. Then perform a full connection operation to convert the 16 5*5 sixth feature images of the above operation into a vector, and the conversion process is often accompanied by the operation to be used. In this embodiment, a 120-dimensional vector is obtained first, and then a 120-dimensional feature vector is converted into a 60-dimensional feature vector.

S102: Perform regression classification on the feature vectors, collect the car models, and obtain a pre-classification model.

After the conversion of the feature vector, the regression classification is performed, and the softmax regression function is used for calculation to obtain the pre-classification model. Among them, h _θ (x ⁽ⁱ⁾ ) is the regression function, x ⁽ⁱ⁾ is the image feature, k is the number of categories, θ is the weight parameter, is the transpose of the weight parameter for the kth feature.

S103: Select several car models and car appearance pictures corresponding to the car models, train the pre-classification model, and obtain a classification model.

Select a certain amount of car models and car appearance pictures corresponding to the car models, and perform pre-classification model training to obtain optimal parameters and weights, and obtain an excellent classification model.

Training is a kind of learning with labels. In the process of training, it is necessary to adjust the parameters of each layer, including calculating the loss function, performing the reverse derivation of the loss function, adjusting the parameters of each layer, and making the loss The function is minimal. Among them, the loss function is in, m is the number of training samples, k is the number of categories, and λ is a constant, generally 0.01-0.02.

The loss function is derived layer by layer from back to front according to the stochastic gradient descent principle and the chain derivation rule, and then the weights are updated.

Such as,

Among them, y _l is the output of layer l, and w is the weight.

According to the result of the derivation, the average gradient method is applied to calculate the weight update, Among them, W(t+1) is the weight after update, W(t) is the weight before update, and λ is a constant, generally 0.01-0.02.

S104: Pre-establish a corresponding relationship between the vehicle model and the charging interface type.

Establish the corresponding relationship between the vehicle model and the charging interface type. In the implementation of the present invention, the mapping relationship between the vehicle model and the charging interface type is established, as shown in FIG. 3 .

S105: Obtain the car shape image of the vehicle type to be judged, and determine the model of the car corresponding to the car shape image of the car model to be judged by the classification model.

When it is necessary to judge the model of the car, obtain the car shape image of the car model to be judged, perform convolution and pooling operations on the image, judge the car model corresponding to the car shape image of the car model to be judged by the classification model, and obtain its corresponding The corresponding car model.

S106: Obtain the charging interface type of the car according to the corresponding relationship and the identified car model.

After obtaining the model of the vehicle, the charging interface type of the vehicle can be obtained according to the mapping relationship between the vehicle type and the charging interface type in the database.

The automobile charging interface identification method provided by the present invention obtains several automobile appearance images, performs convolution, pooling, full connection and regression analysis on the automobile appearance images, establishes a pre-classification model of automobile models, and selects several automobile appearance pictures of known automobile models , to train the pre-classification model to obtain a classification model with higher accuracy. Utilize the obtained classification model to judge the car shape image of the car model to be judged, obtain the car model corresponding to the car shape image, and determine the charging port type through the corresponding relationship between the car model and the charging port type established in advance. judgment.

Each embodiment in this specification is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments.

The embodiments of the present invention described above are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (5)

1. A method for identifying an automobile charging interface is characterized by comprising the following steps:

obtaining a plurality of automobile appearance images, and respectively constructing a feature vector of the automobile appearance images;

wherein the constructing the feature vector mainly comprises extracting the automobile outline image, performing a first convolution operation on the automobile outline image to obtain a first feature image,

performing a first pooling operation on the first characteristic image to obtain a second characteristic image,

performing a second convolution operation on the second characteristic image to obtain a third characteristic image,

performing second pooling operation on the third characteristic image to obtain a fourth characteristic image,

performing a third convolution operation on the fourth characteristic image to obtain a fifth characteristic image,

performing third pooling operation on the fifth characteristic image to obtain a sixth characteristic image,

transforming formula Y by eigenvectors_l(n)＝∑W_l(n,m)*Y_l-1(m) converting the sixth feature image into the feature vector, wherein l is the l-th feature image, W_lIs a weight, Y_l(n) denotes a feature vector, Y_l-1(m) is a feature image before vector conversion;

performing regression classification on the feature vectors, and summarizing the automobile models to obtain a pre-classification model;

selecting a plurality of automobile models and automobile appearance pictures corresponding to the automobile models, and training the pre-classification model to obtain a classification model;

pre-establishing a corresponding relation between the automobile model and the charging interface type;

obtaining an automobile appearance image of the automobile type to be judged, and judging the model of the automobile corresponding to the automobile appearance image of the automobile type to be judged through the classification model;

and obtaining the type of the charging interface of the automobile according to the corresponding relation and the identified automobile model.

2. The vehicle charging interface identification method according to claim 1, wherein in the first convolution operation, the size of a convolution kernel is 5 x 5, and the number of the first characteristic images is 5; the second convolution operation has a convolution kernel size of 5 × 5, and the number of the third feature images is 10; and the third convolution operation has a convolution kernel size of 4 x 4, and the number of the fifth characteristic images is 16.

3. The method according to claim 1, wherein the first pooling operation is performed by summing 2 x 2 neighborhoods of the first feature image, multiplying by a random weight, adding a bias, and performing neuron activation, wherein the neuron activation is logistic, and the logistic activation function is

4. The vehicle charging interface recognition method according to claim 1, further comprising,

and adjusting parameters of each layer, wherein the adjustment of the parameters of each layer specifically comprises solving a loss function, carrying out reverse derivation on the loss function, and adjusting the parameters of each layer to minimize the loss function.

5. The vehicle charging interface recognition method according to claim 4, further comprising,

and (4) conducting layer-by-layer derivation on the loss function from back to front according to a random gradient descent principle and a chain derivation rule, and then updating the weight.