Flight track identification method based on deep residual neural network

Abstract

The embodiment of the invention relates to the technical field of signal processing, in particular to a flight path identification method based on a deep residual neural network, which comprises the following steps: preprocessing acquired sample flight path data to remove abnormal value points, and after marking real labels, dividing according to a preset proportion to obtain a training set and a test set; constructing a track recognition model, wherein the track recognition model is composed of a feature extraction network and a classification network; wherein the feature extraction network is a deep residual neural network adopting a 1D-ResNet structure; performing supervised iterative training on the track identification model based on the training set to continuously realize network parameter optimization of the track identification model until the track identification model converges, and obtaining an optimal track identification model; and finally, inputting target track data to be identified into the optimal track identification model to obtain an identification result of the target track data output by the optimal track identification model. According to the method, the accuracy and efficiency of track identification are remarkably improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of signal processing technology, and in particular to a track recognition method based on a deep residual neural network. Background Art

[0002] With the rapid development of modern society and technology, target track recognition and classification has become an important research topic in various fields. Accurately identifying the target type is crucial for reaction speed and decision quality. Traditional target track recognition methods usually rely on extracting information about the target track, such as the time, longitude and latitude, speed, turn, altitude, etc. However, these traditional methods often fail to fully explore the hidden features in the target track data, such as the deep-level motion patterns and potential information in the target track data.

[0003] Deep neural networks have achieved remarkable results in the field of target track recognition in recent years. However, the deep learning methods currently proposed face problems such as insufficient target track feature extraction, complex and changeable track patterns of different types of targets, and gradient explosion of deep neural learning networks, which leads to certain limitations in track recognition accuracy. Summary of the invention

[0004] In view of this, an embodiment of the present application proposes a track recognition method based on a deep residual neural network, which uses a deep residual neural network to deeply mine the implicit features of the target track data, significantly improving the accuracy and efficiency of track recognition, and meeting the high requirements of track recognition accuracy in various application scenarios.

[0005] In the first aspect, an embodiment of the present application proposes a track recognition method based on a deep residual neural network, comprising the following steps: obtaining sample track data, preprocessing the obtained sample track data to remove outlier points, and after marking the real labels, dividing them into a training set and a test set according to a preset ratio; constructing a track recognition model, the track recognition model consisting of a feature extraction network and a classification network; wherein the feature extraction network is a deep residual neural network with a 1D-ResNet structure, which is used to extract the implicit features of the input track data, and the classification network is used to identify and classify based on the implicit features, and output the recognition result of the input track data; based on the training set, the track recognition model is supervised iteratively trained to continuously optimize the network parameters of the track recognition model until the track recognition model converges to obtain the optimal track recognition model; the target track data to be identified is input into the optimal track recognition model to obtain the recognition result of the target track data output by the optimal track recognition model.

[0006] Optionally, the acquired sample track data is preprocessed to remove outlier points, including: determining data points in the sample track data that suddenly jump to a preset range position away from the track within a preset short time as wild points, and eliminating all wild points; for data points in the sample track data that lack position information and / or speed information, using interpolation to fill in data based on the position information and / or speed information of data points at adjacent moments; determining each data point in the sample track data whose speed information is continuously 0 within a preset time as a stationary point, and eliminating all stationary points. Wild points, data points with missing data, and stationary points are all outlier points in the sample track data. Data points with missing data can be remedied by interpolation of data from adjacent data points. Wild points are major erroneous data and need to be directly eliminated to avoid adverse effects on subsequent model training. Stationary points may also have an adverse effect on subsequent track analysis, so they also need to be directly eliminated to ensure the quality of the sample track data.

[0007] Optionally, the feature extraction network consists of a number of sequentially connected 1D residual blocks, each of which consists of a sequentially connected 1D convolutional layer, a batch normalization layer and an activation function layer, and a jump connection, one end of the jump connection is connected to the input of the 1D convolutional layer, and the other end is connected to the output of the activation function layer.

[0008] Optionally, the activation function layer is implemented using a ReLU activation function, and the output of the 1D residual block is expressed by the formula:

[0009] y a =ReLU{BatchNorm[Conv1D(x a ,W a )]}+x a ;

[0010] Among them, x a represents the input of the a-th 1D residual block, W a represents the learnable network parameters, Conv1D(·) represents the 1D convolution layer, BatchNorm(·) represents the batch normalization layer, ReLU(·) represents the activation function layer, and y a It represents the output of the ath 1D residual block, and the output of the last 1D residual block is the implicit feature extracted by the feature extraction network.

[0011] Optionally, the classification network consists of a first fully connected layer, a first activation layer, a second fully connected layer, a second activation layer and a Softmax layer connected in sequence, the first fully connected layer is provided with N1 neurons, the second fully connected layer is provided with N2 neurons, and the third fully connected layer is provided with N3 neurons; the first fully connected layer is used to perform weighted processing on implicit features and further perform feature extraction, the second fully connected layer is used to further extract features from the features output by the first fully connected layer, the third fully connected layer is used to perform higher-order feature extraction on the features output by the second fully connected layer, the first activation layer and the second activation layer are used to introduce nonlinear mapping capabilities to enable the classification network to learn more complex patterns, and the Softmax layer is used to convert the features output by the third fully connected layer into probabilities belonging to each category to obtain recognition results.

[0012] Optionally, the first activation layer and the second activation layer are both implemented using a ReLU activation function, and the recognition result of the input track data output by the classification network is expressed by the formula:

[0013] y f =f classifer (R n )=f Softmax {f Linear3 [f ReLU2 (f Linear2 [f ReLU1 (f Linear1 [R n ])])]};

[0014] Among them, R n represents a recessive feature, f classifer (·) represents the classification network, y f represents the recognition result, f Linear1 (·) represents the first fully connected layer, f ReLU1 (·) represents the first activation layer, f Linear2 (·) represents the second fully connected layer, f ReLU2 (·) represents the second activation layer, f Linear3 (·) indicates the third fully connected layer.

[0015] Optionally, the supervised iterative training of the track recognition model based on the training set to continuously optimize the network parameters of the track recognition model until the track recognition model converges to obtain the optimal track recognition model includes:

[0016] The sample track data in the training set is input into the track recognition model for supervised training, and the recognition result of the sample track data output by the track recognition model is obtained;

[0017] Using the cross entropy loss function, the loss value between the true label of the sample track data and the recognition result of the sample track data output by the track recognition model is calculated;

[0018] Based on the loss value, the back propagation algorithm is used to continuously optimize and adjust the network parameters of the track recognition model until the track recognition model converges and the optimal track recognition model is obtained;

[0019] The cross entropy loss function is used to calculate the loss value between the true label of the sample track data and the recognition result of the sample track data output by the track recognition model, which is expressed by the formula:

[0020]

[0021] Where C represents the batch size, represents the true label of the cth sample track data, y c It represents the recognition result of the c-th sample track data output by the track recognition model, and LOSS is the calculated loss value.

[0022] The embodiment of the present application proposes a track recognition method based on a deep residual neural network, which constructs, trains and uses a track recognition model to perform aircraft recognition tasks. The constructed track recognition model consists of a feature extraction network and a classification network. The feature extraction network is a deep residual neural network with a 1D-ResNet structure. Such a feature extraction network can deeply explore the deep implicit features of the input track data and fully explore the kinematic laws of the input track data. The classification network performs recognition and classification based on the implicit features, and finally outputs the recognition result of the input track data. The sample track data used in training the model needs to be preprocessed to remove outliers. Such sample track data is scientific, reliable and pollution-free, and can train the track recognition model with high quality. In addition, the deep residual neural network with a 1D-ResNet structure can also overcome the gradient vanishing and gradient explosion problems existing in traditional deep neural networks, thereby significantly improving the accuracy, efficiency and stability of track recognition, and meeting the high requirements of track recognition accuracy in various application scenarios.

[0023] In the second aspect, an embodiment of the present application proposes a track identification system based on a deep residual neural network, the system comprising: a sample division module, used to obtain sample track data, pre-process the obtained sample track data to remove outlier points, and after marking the real labels, divide the data into a training set and a test set according to a preset ratio; a model construction module, used to construct a track identification model, the track identification model consists of a feature extraction network and a classification network, wherein the feature extraction network is a deep residual neural network with a 1D-ResNet structure, which is used to extract the implicit features of the input track data, and the classification network is used to identify and classify based on the implicit features, and output the identification result of the input track data; a model training module, used to perform supervised iterative training on the track identification model based on the training set to continuously optimize the network parameters of the track identification model until the track identification model converges to obtain the optimal track identification model; a model usage module, used to input the target track data to be identified into the optimal track identification model, and obtain the identification result of the target track data output by the optimal track identification model.

[0024] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related technologies, the drawings required for use in the embodiments of the present application or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 is a flow chart of a track recognition method based on a deep residual neural network provided in one embodiment of the present application;

[0027] Figure 2 is a structural schematic diagram of a track recognition model provided in an embodiment of the present application;

[0028] Figure 3 is a structural schematic diagram of a track recognition system based on a deep residual neural network provided in another embodiment of the present application;

[0029] Figure 4 It is a structural schematic diagram of an electronic device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. In the various embodiments of the present application, many technical details are proposed in order to make the reader better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is only for the convenience of description, and the specific implementation mode of the present application should not constitute any limitation. The various embodiments can be combined with each other and referenced to each other under the premise of no contradiction.

[0031] An embodiment of the present application proposes a track recognition method based on a deep residual neural network, which is applied to an electronic device, wherein the electronic device can be a terminal or a server. In this embodiment and the following embodiments, the electronic device is described using a server as an example. The following is a detailed description of the implementation details of the track recognition method based on a deep residual neural network proposed in this embodiment. The following content is only the relevant implementation details provided for ease of understanding and is not necessary for the implementation of this solution. The specific process of the track recognition method based on a deep residual neural network proposed in this embodiment can be as follows: Figure 1 As shown, including:

[0032] Step 101, obtain sample track data, pre-process the obtained sample track data to remove outlier points, and after marking the real labels, divide them into training set and test set according to a preset ratio.

[0033] In the specific implementation, the server first needs to obtain a large amount of sample track data, and then preprocess the obtained sample track data to remove outliers. After the preprocessing is completed, the real label is marked for each sample track data, and then divided according to the preset ratio to obtain the training set and test set for training the track recognition model.

[0034] In one example, the preset ratio may be set to 7:3 or 8:2.

[0035] In one example, there are three types of outlier points, namely, wild points, data points with missing data, and stationary points.

[0036] The server first needs to identify the data points in the sample track data that suddenly jump to a preset range position far away from the track within a preset short time as wild points, and directly remove all wild points. Wild points are major erroneous data and need to be directly removed to avoid adverse effects on subsequent model training. Among them, the preset short time and the preset range can be set by technicians in this field according to actual needs, and the embodiments of this application do not specifically limit this.

[0037] Data points with missing data are data points in the sample track data that lack position information and / or speed information. For data missing points, the server can use interpolation to fill in the data based on the position information and / or speed information of data points at adjacent times, thereby making up for the loss.

[0038] The server determines the data points in the sample track data whose speed information is continuously 0 within a preset time as stationary points. Long-term stillness may also have an adverse effect on subsequent track analysis, so they also need to be directly eliminated to ensure the quality of the sample track data.

[0039] Step 102, construct a track recognition model, which consists of a feature extraction network and a classification network, wherein the feature extraction network is a deep residual neural network using a 1D-ResNet structure, which is used to extract implicit features of input track data, and the classification network is used to identify and classify based on implicit features, and output the recognition result of the input track data.

[0040] In the specific implementation, after the server obtains the training set and the test machine, it can build the track recognition model. The track recognition model consists of a feature extraction network and a classification network. The feature extraction network is a deep residual neural network with a 1D-ResNet structure, which is used to extract the implicit features of the input track data. The classification network is used to identify and classify based on the implicit features, and finally outputs the recognition results of the input track data.

[0041] In one example, the specific structure of the track recognition model can be as follows: Figure 2 shown.

[0042] The classic residual network (ResNet) solves the problems of gradient vanishing and gradient exploding in deep neural networks by means of jump connections. In order to conform to the characteristics of track data and help the model better understand the motion laws and motion characteristics of the target, this embodiment converts the classic residual network into a 1D-ResNet that adapts to 1D time series data to extract implicit features of track data.

[0043] The basic building block of ResNet is the residual block, which consists of two parts, namely the ordinary convolution layer and the jump connection that adds the input and output. In the residual block, the jump connection enables the network to bypass the intermediate layer and pass the track features directly to the next layer, thereby helping to retain the important feature information of the track in the learning of each layer, avoiding the attenuation of information in the multi-layer network, so as to better extract and process the track features, and also solve the problems of gradient disappearance and gradient explosion in deep neural networks.

[0044] 1D-ResNet is an extension of ResNet, which is specially used for processing one-dimensional data. In order to adapt to the one-dimensional characteristics of track data, this embodiment uses 1D convolution in 1D-ResNet to process track data, deeply mine the temporal characteristics of track data, and identify the movement law of the target.

[0045] In one example, the feature extraction network consists of several sequentially connected 1D residual blocks, each of which consists of a sequentially connected 1D convolutional layer, a batch normalization layer, an activation function layer, and a jump connection, where one end of the jump connection is connected to the input of the 1D convolutional layer and the other end is connected to the output of the activation function layer.

[0046] In an example, the function layer in the 1D residual block activation is implemented using the ReLU activation function, and the output of the 1D residual block can be expressed by the formula:

[0047] y a =ReLU{BatchNorm[Conv1D(x a ,W a )]}+x a ;

[0048] Among them, x a represents the input of the a-th 1D residual block, W a represents the learnable network parameters, Conv1D(·) represents the 1D convolution layer, BatchNorm(·) represents the batch normalization layer, ReLU(·) represents the activation function layer, and y a It represents the output of the ath 1D residual block, and the output of the last 1D residual block is the implicit feature extracted by the feature extraction network.

[0049] It should be noted that after comprehensive consideration of model accuracy and network complexity, this embodiment uses the 1D-ResNet18 structure as the feature extraction network to extract track data features, which effectively solves the gradient problem of deep neural networks during long sequence learning and improves the model's ability to process complex track data.

[0050] In an example, the goal of the classification network is to map the latent features of the input to the target category. The main part of the classification network is three fully connected layers and two activation layers, specifically composed of a first fully connected layer, a first activation layer, a second fully connected layer, a second activation layer and a Softmax layer connected sequentially.

[0051] There are N1 neurons in the first fully connected layer, N2 neurons in the second fully connected layer, and N3 neurons in the third fully connected layer. The first fully connected layer is used to perform weighted processing on implicit features and further extract features. The second fully connected layer is used to further extract features from the features output by the first fully connected layer. The third fully connected layer is used to perform higher-order feature extraction on the features output by the second fully connected layer. The first activation layer and the second activation layer are used to introduce nonlinear mapping capabilities to enable the classification network to learn more complex patterns. The Softmax layer is used to convert the features output by the third fully connected layer into probabilities belonging to each category to obtain recognition results.

[0052] In an example, the first activation layer and the second activation layer are both implemented using the ReLU activation function, and the recognition result of the input track data output by the classification network can be expressed by the formula:

[0053] y f =f classifer (R n )=f Softmax {f Linear3 [f ReLU2 (f Linear2 [f ReLU1 (f Linear1 [R n ])])]};

[0054] Among them, R n represents a recessive feature, f classifer (·) represents the classification network, y f represents the recognition result, f Linear1 (·) represents the first fully connected layer, f ReLU1 (·) represents the first activation layer, f Linear2 (·) represents the second fully connected layer, f ReLU2 (·) represents the second activation layer, f Linear3 (·) indicates the third fully connected layer.

[0055] It is worth noting that for the first activation layer and the second activation layer, ReLU(x)=max(0,x).

[0056] Step 103, supervised iterative training is performed on the track recognition model based on the training set to continuously optimize the network parameters of the track recognition model until the track recognition model converges to obtain the optimal track recognition model.

[0057] In a specific implementation, after completing the construction of the track recognition model, the server can perform supervised iterative training on the track recognition model based on the training set to continuously optimize the network parameters of the track recognition model until the track recognition model converges and obtains the optimal track recognition model.

[0058] In one example, when the server performs model training, it needs to input the sample track data in the training set into the track recognition model for supervised training, and obtain the recognition result of the sample track data output by the track recognition model. Then, the cross entropy loss function is used to calculate the loss value between the true label of the sample track data and the recognition result of the sample track data output by the track recognition model. Finally, based on the loss value, the back propagation algorithm is used to continuously optimize and adjust the network parameters of the track recognition model until the track recognition model converges and the optimal track recognition model is obtained.

[0059] In one example, the server uses the cross entropy loss function to calculate the loss value between the true label of the sample track data and the recognition result of the sample track data output by the track recognition model, which can be expressed by the formula:

[0060]

[0061] Where C represents the batch size, represents the true label of the cth sample track data, y c It represents the recognition result of the c-th sample track data output by the track recognition model, and LOSS is the calculated loss value.

[0062] In one example, the test set is used to perform various performance tests on the optimal track recognition model. The server inputs the sample track data in the test set into the optimal track recognition model, and obtains various test indicators based on the recognition results of the sample track data in the test set output by the optimal track recognition model and the true labels of the sample track data in the test set. If one or several tests fail, the server will return the optimal track recognition model for retraining and network parameter tuning.

[0063] Step 104, inputting the target track data to be identified into the optimal track identification model, and obtaining the identification result of the target track data output by the optimal track identification model.

[0064] In one example, after obtaining the optimal track recognition model, the server can deploy the optimal track recognition model to the application scenario in need. When using the model, the target track data to be identified is input into the optimal track recognition model to obtain the recognition result of the target track data output by the optimal track recognition model.

[0065] It is worth noting that in order to further improve the accuracy and efficiency of track recognition, the server needs to perform the same preprocessing on the target track data to be identified as the sample track data before inputting it into the optimal track recognition model to eliminate outlier points.

[0066] The present embodiment proposes a track recognition method based on a deep residual neural network, which constructs, trains and uses a track recognition model to perform aircraft recognition tasks. The constructed track recognition model consists of a feature extraction network and a classification network. The feature extraction network is a deep residual neural network with a 1D-ResNet structure. Such a feature extraction network can deeply mine the deep hidden features of the input track data and fully mine the kinematic laws of the input track data. The classification network performs recognition and classification based on the hidden features, and finally outputs the recognition result of the input track data. The sample track data used in training the model needs to be preprocessed to remove outliers. Such sample track data is scientific, reliable and pollution-free, and can train the track recognition model with high quality. In addition, the deep residual neural network with a 1D-ResNet structure can also overcome the gradient vanishing and gradient explosion problems existing in traditional deep neural networks, thereby significantly improving the accuracy, efficiency and stability of track recognition, and meeting the high requirements of track recognition accuracy in various application scenarios.

[0067] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this application.

[0068] Another embodiment of the present application proposes a track recognition system based on a deep residual neural network. The implementation details of the track recognition system based on a deep residual neural network proposed in this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this example. Figure 3 It is a structural diagram of a track identification system based on a deep residual neural network proposed in this embodiment. The system includes: a sample division module 201, a model construction module 202, a model training module 203 and a model use module 204.

[0069] The sample division module 201 is used to obtain sample track data, pre-process the obtained sample track data to remove abnormal value points, and after marking the real labels, divide them into training sets and test sets according to a preset ratio.

[0070] The model building module 202 is used to build a track recognition model. The track recognition model consists of a feature extraction network and a classification network. The feature extraction network is a deep residual neural network with a 1D-ResNet structure, which is used to extract the implicit features of the input track data. The classification network is used to identify and classify based on the implicit features, and output the recognition results of the input track data.

[0071] The model training module 203 is used to perform supervised iterative training on the track recognition model based on the training set to continuously optimize the network parameters of the track recognition model until the track recognition model converges to obtain the optimal track recognition model.

[0072] The model using module 204 is used to input the target track data to be identified into the optimal track identification model to obtain the identification result of the target track data output by the optimal track identification model.

[0073] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0074] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiments.

[0075] Another embodiment of the present application deploys an electronic device, whose specific structure is as follows Figure 4 As shown, it includes: at least one processor 301; and a memory 302 that is communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions that can be executed by the at least one processor 301, and the instructions are executed by the at least one processor 301 so that the at least one processor 301 can execute a track recognition method based on a deep residual neural network described in the above-mentioned method embodiments.

[0076] Among them, the memory and the processor can be connected in a bus manner, and the bus can include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and will not be further described in this article. The bus interface is responsible for providing an interface between the bus and the transceiver. The transceiver can be one component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0077] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0078] Those skilled in the art can understand that all or part of the steps in the above-mentioned method embodiments can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions for a device (such as a single-chip microcomputer, chip, etc.) or a processor (processor) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.

[0079] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A track recognition method based on deep residual neural network, characterized in that: include: Obtain sample track data, preprocess the obtained sample track data to remove outlier points, and after marking the true labels, divide them into training sets and test sets according to a preset ratio; Construct a track recognition model, which consists of a feature extraction network and a classification network. The feature extraction network is a deep residual neural network with a 1D-ResNet structure, which is used to extract the implicit features of the input track data. The classification network is used to identify and classify based on the implicit features and output the recognition results of the input track data. Based on the training set, supervised iterative training is performed on the track recognition model to continuously optimize the network parameters of the track recognition model until the track recognition model converges and the optimal track recognition model is obtained; The target track data to be identified is input into the optimal track identification model to obtain the identification result of the target track data output by the optimal track identification model.

2. The track recognition method based on deep residual neural network according to claim 1 is characterized in that: The preprocessing of the acquired sample track data to remove outlier points includes: Determine data points in the sample track data that suddenly jump to a preset range of positions far away from the track within a preset short time as outliers, and remove all outliers; For data points with missing position information and / or speed information in the sample track data, interpolation is used to fill in the data based on the position information and / or speed information of data points at adjacent times; Each data point in the sample track data whose speed information is continuously 0 within a preset time is determined as a stationary point, and all stationary points are eliminated.

3. The track recognition method based on deep residual neural network according to claim 1 is characterized in that: The feature extraction network consists of a number of sequentially connected 1D residual blocks, each of which consists of a sequentially connected 1D convolutional layer, a batch normalization layer, an activation function layer, and a jump connection, one end of the jump connection is connected to the input of the 1D convolutional layer, and the other end is connected to the output of the activation function layer.

4. The track recognition method based on deep residual neural network according to claim 3 is characterized in that: The activation function layer is implemented using the ReLU activation function, and the output of the 1D residual block is expressed by the formula: y a =ReLU{BatchNorm[Conv1D(x a ,W a )]}+x a ; Among them, x a represents the input of the a-th 1D residual block, W a represents the learnable network parameters, Conv1D(·) represents the 1D convolution layer, BatchNorm(·) represents the batch normalization layer, ReLU(·) represents the activation function layer, and y a It represents the output of the ath 1D residual block, and the output of the last 1D residual block is the implicit feature extracted by the feature extraction network.

5. The track recognition method based on deep residual neural network according to claim 1 is characterized in that: The classification network consists of a first fully connected layer, a first activation layer, a second fully connected layer, a second activation layer and a Softmax layer connected in sequence, wherein the first fully connected layer is provided with N1 neurons, the second fully connected layer is provided with N2 neurons, and the third fully connected layer is provided with N3 neurons; The first fully connected layer is used to perform weighted processing on implicit features and further perform feature extraction. The second fully connected layer is used to further extract features from the features output by the first fully connected layer. The third fully connected layer is used to perform higher-order feature extraction on the features output by the second fully connected layer. The first activation layer and the second activation layer are used to introduce nonlinear mapping capabilities to enable the classification network to learn more complex patterns. The Softmax layer is used to convert the features output by the third fully connected layer into probabilities belonging to each category to obtain recognition results.

6. The track recognition method based on deep residual neural network according to claim 5 is characterized in that: The first activation layer and the second activation layer are both implemented using the ReLU activation function. The recognition result of the input track data output by the classification network is expressed by the formula: y f =f classifer (R n )=f Softmax {f Linear3 [f ReLU2 (f Linear2 [f ReLU1 (f Linear1 [R n ])])]}; Among them, R n represents a hidden feature, f classifer (·) represents the classification network, y f represents the recognition result, f Linear1 (·) represents the first fully connected layer, f reLU1 (·) represents the first activation layer, f Linear2 (·) represents the second fully connected layer, f ReLU2 (·) represents the second activation layer, f Linear3 (·) indicates the third fully connected layer.

7. The track recognition method based on deep residual neural network according to any one of claims 1 to 6, characterized in that: The supervised iterative training of the track recognition model based on the training set to continuously optimize the network parameters of the track recognition model until the track recognition model converges to obtain the optimal track recognition model includes: The sample track data in the training set is input into the track recognition model for supervised training, and the recognition result of the sample track data output by the track recognition model is obtained; Using the cross entropy loss function, the loss value between the true label of the sample track data and the recognition result of the sample track data output by the track recognition model is calculated; Based on the loss value, the back propagation algorithm is used to continuously optimize and adjust the network parameters of the track recognition model until the track recognition model converges and the optimal track recognition model is obtained; The cross entropy loss function is used to calculate the loss value between the true label of the sample track data and the recognition result of the sample track data output by the track recognition model, which is expressed by the formula: Where C represents the batch size, represents the true label of the cth sample track data, y c It represents the recognition result of the c-th sample track data output by the track recognition model, and LOSS is the calculated loss value.

8. A track recognition system based on deep residual neural network, characterized in that: include: The sample division module is used to obtain sample track data, pre-process the obtained sample track data to remove outlier points, and divide the data into training set and test set according to a preset ratio after marking the real labels; The model building module is used to build a track recognition model. The track recognition model consists of a feature extraction network and a classification network. The feature extraction network is a deep residual neural network with a 1D-ResNet structure, which is used to extract the implicit features of the input track data. The classification network is used to identify and classify based on the implicit features and output the recognition results of the input track data. The model training module is used to perform supervised iterative training on the track recognition model based on the training set to continuously optimize the network parameters of the track recognition model until the track recognition model converges and obtains the optimal track recognition model; The model using module is used to input the target track data to be identified into the optimal track identification model to obtain the identification result of the target track data output by the optimal track identification model.

Images