Radar target recognition method and model training method based on convolutional neural network

Abstract

The present disclosure provides a radar target recognition method and model training method based on a convolutional neural network, which relates to radar signal processing technology, including: obtaining a training data set; wherein the training data set includes multiple sets of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object; inputting the feature map of the target object into a preset model to obtain a predicted category of the target object; optimizing the parameters in the preset model according to the labeled category and predicted category of the target object to obtain a radar target recognition model. This solution, which performs target recognition on radar echo data based on a convolutional neural network, does not require a large amount of manual work, and also has a good target recognition effect under a complex background with a lot of noise and clutter contained in the radar echo data.

Description

Technical Field

[0001] The present disclosure relates to radar signal processing technology, and in particular to a radar target recognition method and a model training method based on convolutional neural network. Background Art

[0002] Radar target recognition is an important issue in the field of radar signal processing. Radar target recognition is a technology that uses radar and computers to identify distant targets. It estimates the size, shape, weight and physical characteristic parameters of the target surface layer by analyzing the target characteristic information such as amplitude, phase, spectrum and polarization in the radar echo, and finally makes recognition decisions in the classifier based on the discriminant function determined by a large number of training samples.

[0003] In the prior art, traditional target recognition methods rely on artificially extracted target features, mainly including: target recognition based on Radar Cross Section (RCS), target recognition based on one-dimensional range profile, target recognition based on micro-Doppler effect, etc. By artificially extracting features such as RCS size, fluctuation characteristics, radial size, number of scattering points, center distance, micro-Doppler, etc., support vector machine (SVM), classification tree, k-nearest neighbor (KNN) and other classification methods are used to classify and recognize targets.

[0004] However, the above method requires a lot of manual work; and in the complex background of radar echoes containing a lot of noise and clutter, the target recognition effect is not satisfactory. Summary of the invention

[0005] The present invention provides a radar target recognition method and a model training method based on a convolutional neural network to solve the problems that a large amount of manual work is required in the prior art; and the target recognition effect is unsatisfactory under the complex background of a large amount of noise and clutter contained in the radar echo.

[0006] According to a first aspect of the present disclosure, a model training method is provided, comprising:

[0007] Acquire a training data set; wherein the training data set includes multiple sets of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object;

[0008] Inputting the feature map of the target object into a preset model to obtain a predicted category of the target object;

[0009] According to the labeled category and predicted category of the target object, the parameters in the preset model are optimized to obtain a radar target recognition model.

[0010] In one achievable manner, obtaining a training data set includes:

[0011] Acquire radar echo data, and preprocess the radar echo data to obtain an original feature map;

[0012] Determine the target object included in the radar echo data according to the original feature map;

[0013] Determining the location information of the target object according to the radar echo data, and determining the labeling category of the target object according to the location information of the target object;

[0014] A feature map of the target object is determined in the original feature map.

[0015] In an implementable manner, the original characteristic graph includes a range-pulse characteristic graph, a range-azimuth characteristic graph, and a range-Doppler characteristic graph;

[0016] The preprocessing of the radar echo data to obtain the original feature map includes:

[0017] Performing pulse compression on the radar echo data to obtain the range-pulse characteristic diagram;

[0018] Performing moving target display on the radar echo data to obtain the range-azimuth characteristic diagram;

[0019] Moving target detection is performed on the radar echo data to obtain the range-Doppler characteristic diagram.

[0020] In an implementable manner, the original characteristic graph includes a range-pulse characteristic graph, a range-azimuth characteristic graph, and a range-Doppler characteristic graph;

[0021] Determining the feature map of the target object in the original feature map includes:

[0022] According to the position information of the target object in the radar echo data, a sub-region corresponding to the target object is determined in the range-pulse characteristic graph, the range-azimuth characteristic graph, and the range-Doppler characteristic graph, respectively; the sub-region includes the target object;

[0023] The sub-regions of the target object are fused to obtain a feature map of the target object.

[0024] In an implementable manner, the sub-regions of the target object have the same size;

[0025] The sub-regions of the target object are fused to obtain a feature map of the target object, including:

[0026] The feature map data at the same position in each of the sub-areas are superimposed to obtain the feature map of the target object.

[0027] In one achievable manner, the radar target recognition model does not include a pooling layer.

[0028] In an implementable manner, the radar echo data includes echo data of sea observation and ground observation using coherent pulse radar.

[0029] According to a second aspect of the present disclosure, a radar target recognition method is provided, comprising:

[0030] Acquire radar echo data, and preprocess the radar echo data to obtain an original feature map;

[0031] Determining a target object included in the radar echo data according to the original feature map, and determining a feature map of the target object according to the original feature map;

[0032] Inputting the characteristic map of the target object into a preset radar target recognition model to obtain a category corresponding to the target object;

[0033] The radar target recognition model is trained using a training data set, which includes multiple groups of training data. The training data includes radar echo sample data used to train the model, a feature map of the target object included in the radar echo sample data, and a labeled category of the target object.

[0034] According to a third aspect of the present disclosure, a model training device is provided, comprising:

[0035] An acquisition unit, configured to acquire a training data set; wherein the training data set includes multiple sets of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object;

[0036] A training unit, used for inputting the feature map of the target object into a preset model to obtain a predicted category of the target object;

[0037] The training unit is also used to optimize the parameters in the preset model according to the labeled category and predicted category of the target object to obtain a radar target recognition model.

[0038] According to a fourth aspect of the present disclosure, a radar target recognition device is provided, comprising:

[0039] An acquisition unit, used for acquiring radar echo data, and preprocessing the radar echo data to obtain an original feature map;

[0040] a processing unit, configured to determine a target object included in the radar echo data according to the original feature map, and determine a feature map of the target object according to the original feature map;

[0041] An identification unit, used for inputting the characteristic map of the target object into a preset radar target identification model to obtain a category corresponding to the target object;

[0042] The radar target recognition model is trained using a training data set, which includes multiple groups of training data. The training data includes radar echo sample data used to train the model, a feature map of the target object included in the radar echo sample data, and a labeled category of the target object.

[0043] According to a fifth aspect of the present disclosure, an electronic device is provided, including a memory and a processor; wherein:

[0044] The memory is used to store computer programs;

[0045] The processor is used to read the computer program stored in the memory and execute the method as described in the first aspect and the second aspect according to the computer program in the memory.

[0046] According to a sixth aspect of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method described in the first aspect and the second aspect is implemented.

[0047] According to a seventh aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method described in the first aspect and the second aspect is implemented.

[0048] The radar target recognition method and model training method based on convolutional neural network provided by the present disclosure include: obtaining a training data set; wherein the training data set includes multiple groups of training data, and the training data includes radar echo data, a feature map of the target object included in the radar echo data, and a labeled category of the target object; inputting the feature map of the target object into a preset model to obtain a predicted category of the target object; optimizing the parameters in the preset model according to the labeled category and the predicted category of the target object to obtain a radar target recognition model. In the radar target recognition method and model training method based on convolutional neural network provided by the present solution, target recognition of radar echo data based on convolutional neural network does not require a lot of manual work, and has a good target recognition effect in a complex background with a lot of noise and clutter contained in the radar echo data. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0050] Figure 1 A flowchart of a model training method according to an exemplary embodiment of the present disclosure is shown;

[0051] Figure 2 A flowchart of a model training method according to another exemplary embodiment of the present disclosure is shown;

[0052] Figure 3 A schematic diagram of a distance-pulse characteristic diagram of a target object shown as an exemplary embodiment of the present disclosure;

[0053] Figure 4 A schematic diagram of a distance-azimuth characteristic diagram of a target object shown in an exemplary embodiment of the present disclosure;

[0054] Figure 5 A schematic diagram of a range-Doppler characteristic diagram of a target object shown in an exemplary embodiment of the present disclosure;

[0055] Figure 6 A schematic diagram of a feature map of a target object shown in an exemplary embodiment of the present disclosure;

[0056] Figure 7 A schematic diagram of a flow chart of a radar target recognition method according to an exemplary embodiment of the present disclosure;

[0057] Figure 8 A structural diagram of a model training device shown as an exemplary embodiment of the present disclosure;

[0058] Fig. 9 This is a structural diagram of a radar target recognition device according to an exemplary embodiment of the present disclosure;

[0059] Fig.10The present invention is a structural diagram of an electronic device according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0060] Radar target recognition is an important issue in the field of radar signal processing. Radar target recognition is a technology that uses radar and computers to identify distant targets. By analyzing the target feature information such as amplitude, phase, spectrum and polarization in the radar echo, various mathematical multidimensional space transformations are used to estimate the size, shape, weight and physical characteristic parameters of the surface layer of the target. Finally, the identification judgment is made in the classifier based on the discriminant function determined by a large number of training samples. Modern radar is not only a tool for detecting and locating distant targets, but also can measure parameters related to the target shape and surface physical characteristics, and then classify and identify the target. In the prior art, traditional target recognition methods rely on artificial extraction of target features, mainly including: target recognition based on radar cross section (RCS), target recognition based on one-dimensional range image, target recognition based on micro-Doppler effect, etc. Among them, RCS is a measure of the target's ability to reflect radar signals in the direction of radar reception. The RCS of a target is equal to the ratio of the power reflected by the target per unit solid angle in the direction of the radar receiving antenna (per single solid angle) to the power density incident on the target (per square meter). By manually extracting features such as RCS size, fluctuation characteristics, radial size, number of scattering points, center distance, micro-Doppler, etc., classification methods such as Support Vector Machine (SVM), classification tree, and K-Nearest Neighbor (KNN) are used to classify and identify targets.

[0061] However, the above method requires a lot of manual work; and in the complex background of radar echoes containing a lot of noise and clutter, the target recognition effect is not satisfactory.

[0062] In order to solve the above technical problems, the solution provided in the present invention performs target recognition on radar echo data based on a convolutional neural network, which does not require a lot of manual work and has a good target recognition effect even in a complex background with a lot of noise and clutter contained in the radar echo data.

[0063] Figure 1 A flowchart of a model training method shown as an exemplary embodiment of the present disclosure.

[0064] like Figure 1 As shown, the model training method provided in this embodiment includes:

[0065] Step 101, obtaining a training data set; wherein the training data set includes multiple groups of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object.

[0066] The method provided in the present disclosure may be executed by an electronic device with computing capability, such as a computer, etc. The electronic device is capable of acquiring a training data set.

[0067] The training data set may include radar echo data collected by the acquisition device and sent to the electronic device, a feature map of the target object obtained by the electronic device based on the received radar echo data, and a labeled category of the target object.

[0068] The acquisition device refers to the radar echo data acquisition device, such as a coherent pulse radar. Pulse radar is relative to continuous wave radar, and refers to a radar that detects by sending short pulses. Coherence is also called phase coherence, which means that there is a certain phase relationship between the pulses emitted by the radar, and Doppler processing can be performed. Coherent pulse radar is a radar that can emit phase-determined pulses for detection. Currently, most early warning radars are coherent pulse radars.

[0069] Specifically, the object identified from the radar echo data may be referred to as a target object.

[0070] The original radar echo data may be echo data of coherent pulse radar for sea observation and ground observation, and the targets may be ships, vehicles, and tugboats, respectively.

[0071] Specifically, the characteristic map of the target object may be a characteristic map obtained by processing the radar echo data based on the position range of the target object, wherein the position information of the target object may be obtained based on the position information included in the radar echo data.

[0072] The target object's labeled categories may include vehicles, ships, and towballs. Specifically, the geographic information within the radar detection range may be queried, and the points within the land range may be marked as category 1: vehicles; the points on the sea surface may be marked as category 2: ships; and the cooperative experiment target may be marked as category 3: towballs. Here, the geographic tags of the cooperative experiment targets are obtained by the positioning device.

[0073] Specifically, a tow ball can be towed on a ship, and a positioning device can be placed on the ship or the tow ball, and the positioning device can locate the position of the tow ball. If the position information of the target object matches the position information located by the positioning device, the target object is determined to be a tow ball.

[0074] The positioning device may be, for example, a Global Positioning System (GPS).

[0075] The target detected in the radar echo signal is displayed on the radar display according to its azimuth and distance coordinates, which is called a point trace. Step 102, input the feature map of the target object into a preset model to obtain the predicted category of the target object.

[0076] The predicted categories may include vehicles, ships, and tugballs. Specifically, the categories included in the predicted categories are the same as the categories included in the marked categories.

[0077] The preset model is a pre-set network model, which is set based on a convolutional neural network.

[0078] Specifically, the noise and clutter contained in the radar echo can be reduced after being processed by a network model based on a convolutional neural network.

[0079] Step 103, according to the labeled category and predicted category of the target object, optimize the parameters in the preset model to obtain a radar target recognition model.

[0080] Specifically, the labeled category of the target object can be used as the label of the predicted category of the target object, and the preset model can be trained to optimize the parameters in the preset model. This makes the predicted category closer and closer to the labeled category. When the preset conditions are met, such as the degree of identity between the predicted category and the labeled category reaches 90%, the training is stopped to obtain the radar target recognition model.

[0081] Specifically, the radar target recognition model may output the category of the target object included in the collected radar echo data.

[0082] The radar target recognition method and model training method based on convolutional neural network provided by the present disclosure include: obtaining a training data set; wherein the training data set includes multiple groups of training data, and the training data includes radar echo data, a feature map of the target object included in the radar echo data, and a labeled category of the target object; inputting the feature map of the target object into a preset model to obtain a predicted category of the target object; optimizing the parameters in the preset model according to the labeled category and the predicted category of the target object to obtain a radar target recognition model. In the radar target recognition method and model training method based on convolutional neural network provided by the present solution, target recognition of radar echo data based on convolutional neural network does not require a lot of manual work, and has a good target recognition effect in a complex background with a lot of noise and clutter contained in the radar echo data.

[0083] Figure 2A flowchart of a model training method shown in another exemplary embodiment of the present disclosure.

[0084] like Figure 2 As shown, the model training method provided in this embodiment includes:

[0085] Step 201: Acquire radar echo data, and preprocess the radar echo data to obtain an original feature map.

[0086] Specifically, the electronic device may acquire the radar echo data acquired by the acquisition device, and preprocess the acquired radar echo data to obtain an original feature map.

[0087] Optionally, the radar echo data includes echo data of sea observation and ground observation using coherent pulse radar.

[0088] The acquisition device may be a coherent pulse radar. The coherent pulse radar can observe the sea and the ground, obtain radar echo data, and send the collected radar echo data to the electronic device. The targets of the coherent pulse radar may be vehicles, ships, and tugboats, respectively. The electronic device may obtain the radar echo data sent by the coherent pulse radar.

[0089] Optionally, the original feature graph includes a range-pulse feature graph, a range-azimuth feature graph, and a range-Doppler feature graph;

[0090] Specifically, the original feature graph may include multiple feature graphs, such as a range-pulse feature graph, a range-azimuth feature graph, and a range-Doppler feature graph.

[0091] Perform pulse compression on radar echo data to obtain a distance-pulse characteristic diagram; perform moving target display on radar echo data to obtain a distance-azimuth characteristic diagram;

[0092] Perform moving target detection on radar echo data to obtain the range-Doppler characteristic diagram.

[0093] Step 202: determining the target object included in the radar echo data according to the original feature map.

[0094] Specifically, a constant false alarm detection method may be used to determine the target object included in the radar echo data in the original feature map.

[0095] Specifically, for the echo signal of coherent pulse radar, we generally take the four steps of pulse compression, moving target display, moving target detection, and constant false alarm detection to process it. Among them, the three feature maps obtained by pulse compression, moving target display, and moving target detection are the range-pulse feature map, the range-azimuth feature map, and the range-Doppler feature map. These three steps are essentially a kind of filter to suppress clutter. Constant false alarm detection obtains point trace information.

[0096] Step 203: determine the location information of the target object according to the radar echo data, and determine the labeling category of the target object according to the location information of the target object.

[0097] Specifically, the position information of the target object can be determined according to the azimuth and distance information included in the radar echo data, wherein the position information of the target object refers to the latitude and longitude information.

[0098] The target object's labeled categories may include vehicles, ships, and towballs. Specifically, the geographic information within the radar detection range may be queried, and the points within the land range may be marked as category 1: vehicles; the points on the sea surface may be marked as category 2: ships; and the cooperative experiment target may be marked as category 3: towballs. Here, the geographic tags of the cooperative experiment targets are obtained by the positioning device.

[0099] Specifically, a tow ball can be towed on a ship, and a positioning device can be placed on the ship or the tow ball, and the positioning device can locate the position of the tow ball. If the position information of the target object matches the position information located by the positioning device, the target object is determined to be a tow ball.

[0100] Step 204: determine a feature map of the target object in the original feature map.

[0101] Specifically, the feature map including the target object may be intercepted from the original feature map.

[0102] Optionally, the original feature graph includes a range-pulse feature graph, a range-azimuth feature graph, and a range-Doppler feature graph;

[0103] According to the position information of the target object in the radar echo data, a sub-region corresponding to the target object is determined in the range-pulse characteristic diagram, the range-azimuth characteristic diagram, and the range-Doppler characteristic diagram respectively; the sub-region includes the target object;

[0104] Specifically, the scanning range of the radar can be divided into a grid of M*N, where M is the number of range units and N is the number of wave positions. Each range unit can be set to, for example, 1.5 meters; each wave position can be set to, for example, 0.5 degrees.

[0105] Specifically, according to the range spread and pulse number of the target object in the radar echo data, the 10 range units before and after the target location and the 3 wave positions on the left and right can be divided into the target range, which is 21*6 in size.

[0106] Among them, sub-areas corresponding to the target object are respectively determined in the distance-pulse characteristic map, the distance-azimuth characteristic map, and the distance-Doppler characteristic map, and each sub-area can be respectively called the distance-pulse characteristic map of the target object, the distance-azimuth characteristic map of the target object, and the distance-Doppler characteristic map of the target object.

[0107] like Figure 3 As shown, it is a distance-pulse characteristic diagram of the target object, wherein coordinates 0-1 represent grayscale values; coordinates 0-20 represent distance units; and coordinates 1-6 represent wave positions.

[0108] like Figure 4 As shown, it is a distance-azimuth characteristic diagram of the target object, where the coordinates 0-1 represent the grayscale value; the coordinates 0-20 represent the distance unit; and the coordinates 1-6 represent the wave position.

[0109] like Figure 5 As shown, it is a distance-Doppler characteristic diagram of the target object, wherein coordinates 0-1 represent grayscale values; coordinates 0-20 represent distance units; and coordinates 1-6 represent wave positions.

[0110] The sub-regions of the target object are fused to obtain the feature map of the target object.

[0111] Specifically, multiple feature maps of the target object may be fused to obtain a feature map of the target object.

[0112] Optionally, the sub-regions of the target object have the same size; the feature map data at the same position in each sub-region are superimposed to obtain the feature map of the target object.

[0113] According to the range spread and pulse number of the target object in the radar echo data, the 10 range units before and after the target location and the 3 wave positions on the left and right can be divided into a target range, the size of which is 21*6. Specifically, the size of each sub-area of ​​the target object can be 21*6.

[0114] like Figure 6 As shown, it is through Figure 3 , Figure 4 , Figure 5 The fused feature map of the target object. Coordinates 0-20 represent distance units, and coordinates 1-6 represent wave positions.

[0115] Step 205: input the feature map of the target object into a preset model to obtain a predicted category of the target object.

[0116] Specifically, the principle and implementation of step 205 are similar to those of step 102 and will not be described in detail.

[0117] Step 206, optimizing the parameters in the preset model according to the labeled category and predicted category of the target object to obtain a radar target recognition model.

[0118] Specifically, the labeled category of the target object can be used as the label of the predicted category of the target object, and the preset model can be trained to optimize the parameters in the preset model. This makes the predicted category closer and closer to the labeled category. When the preset conditions are met, such as the degree of identity between the predicted category and the labeled category reaches 90%, the training is stopped to obtain the radar target recognition model.

[0119] Specifically, the radar target recognition model may output the category of the target object included in the collected radar echo data.

[0120] Optionally, no pooling layer is included in the radar target recognition model.

[0121] Exemplarily, the input layer size of the preset model can be 21*6*3 (21 distance units, 6 wave positions, 3 prediction categories), using two convolutional layers and two activation layers alternately connected, where the first convolution layer size is 3*3*3, the step size is 1, and the number of convolution kernels is 16, the second convolution layer size is 3*3*3, the step size is 1, and the number of convolution kernels is 8. The activation function uses the Relu function, and its expression is f(x) = {x, x≥0; 0, x<0}, where x represents the input value of the Relu function. Since the input feature map is small, this network does not set a pooling layer. Finally, two fully connected layers are connected, with sizes of 100 and 30 respectively, and the number of neurons in the output layer is 3, corresponding to the number of categories. The activation function of the fully connected layer uses the sigmoid function, and its expression is Among them, x represents the input value of the sigmoid function. The output layer uses the softmax function, which is expressed as Among them, j represents all categories; i represents the current predicted category, and z represents the input value of the softmax function. The network loss function adopts the cross entropy loss function, which is expressed as Where N is the number of samples, M is the number of categories; y ic is 0 or 1, the true category of sample i is equal to c and takes 1, otherwise it takes 0; p ic is the predicted probability that the observed sample i belongs to category c.

[0122] For example, the model training process can use random batch gradient descent method, the learning rate is set to be adjusted between 0.01-0.1, and the number of epochs is set to be adjusted between 20 and 100. For the problem of imbalanced prediction of three categories, a random oversampling method is used to ensure that the number of samples of the three categories is not much different during the training process.

[0123] Figure 7 The figure is a flowchart of a radar target recognition method according to an exemplary embodiment of the present disclosure.

[0124] like Figure 7 As shown, the radar target recognition method provided in this embodiment includes:

[0125] Step 701: Acquire radar echo data, and preprocess the radar echo data to obtain an original feature map.

[0126] Specifically, the principle and implementation of step 701 are similar to those of step 201 and will not be described in detail.

[0127] Step 702: determine the target object included in the radar echo data according to the original feature map, and determine the feature map of the target object according to the original feature map.

[0128] Specifically, a constant false alarm detection method may be used to determine the target object included in the radar echo data in the original feature map.

[0129] Specifically, a feature map including the target object may be intercepted from the original feature map, thereby determining the feature map of the target object.

[0130] Optionally, the original feature graph includes a range-pulse feature graph, a range-azimuth feature graph, and a range-Doppler feature graph;

[0131] According to the position information of the target object in the radar echo data, a sub-region corresponding to the target object is determined in the range-pulse characteristic diagram, the range-azimuth characteristic diagram, and the range-Doppler characteristic diagram respectively; the sub-region includes the target object;

[0132] Specifically, the scanning range of the radar can be divided into a grid of M*N, where M is the number of range units and N is the number of wave positions. Each range unit can be set to, for example, 1.5 meters; each wave position can be set to, for example, 0.5 degrees.

[0133] Specifically, there may be multiple target objects in the radar echo data of one wave position.

[0134] Specifically, according to the range spread and pulse number of the target object in the radar echo data, the 10 range units before and after the target location and the 3 wave positions on the left and right can be divided into the target range, which is 21*6 in size.

[0135] Among them, sub-areas corresponding to the target object are respectively determined in the distance-pulse characteristic map, the distance-azimuth characteristic map, and the distance-Doppler characteristic map, and each sub-area can be respectively called the distance-pulse characteristic map of the target object, the distance-azimuth characteristic map of the target object, and the distance-Doppler characteristic map of the target object.

[0136] The sub-regions of the target object are fused to obtain the feature map of the target object.

[0137] Specifically, multiple feature maps of the target object may be fused to obtain a feature map of the target object.

[0138] Optionally, the sub-regions of the target object have the same size; the feature map data at the same position in each sub-region are superimposed to obtain the feature map of the target object.

[0139] According to the range spread and pulse number of the target object in the radar echo data, the 10 range units before and after the target location and the 3 wave positions on the left and right can be divided into a target range, the size of which is 21*6. Specifically, the size of each sub-area of ​​the target object can be 21*6.

[0140] Step 703, inputting the feature map of the target object into a preset radar target recognition model to obtain a category corresponding to the target object;

[0141] Among them, the radar target recognition model is trained using a training data set, which includes multiple groups of training data. The training data includes radar echo sample data used to train the model, feature maps of target objects included in the radar echo sample data, and labeled categories of the target objects.

[0142] The preset radar target recognition model is obtained by pre-training using a training data set, wherein the training data set includes multiple sets of training data, each set of training data includes radar echo sample data for training the model, and a feature map of the target object included in the radar echo sample data and a labeled category of the target object.

[0143] The radar target recognition model obtained by training the labeled categories can recognize the same categories as those included in the labeled categories. Specifically, the labeled categories may include multiple categories, such as vehicles, ships, and tugs, and the preset radar target recognition model can recognize vehicles, ships, and tugs.

[0144] Figure 8 The figure is a structural diagram of a model training device according to an exemplary embodiment of the present disclosure.

[0145] like Figure 8 As shown, the model training device 800 provided by the present disclosure includes:

[0146] An acquisition unit 810 is used to acquire a training data set; wherein the training data set includes multiple sets of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object;

[0147] A training unit 820 is used to input the feature map of the target object into a preset model to obtain a predicted category of the target object;

[0148] The training unit 830 is also used to optimize the parameters in the preset model according to the labeled category and predicted category of the target object to obtain the radar target recognition model.

[0149] The acquisition unit 810 is specifically used to acquire radar echo data and preprocess the radar echo data to obtain an original feature map;

[0150] Determine the target object included in the radar echo data according to the original feature map;

[0151] Determine the location information of the target object according to the radar echo data, and determine the labeling category of the target object according to the location information of the target object;

[0152] Determine the feature map of the target object in the original feature map.

[0153] The original characteristic graph includes a range-pulse characteristic graph, a range-azimuth characteristic graph and a range-Doppler characteristic graph; the acquisition unit 810 is specifically used to perform pulse compression on the radar echo data to obtain a range-pulse characteristic graph; perform moving target display on the radar echo data to obtain a range-azimuth characteristic graph; perform moving target detection on the radar echo data to obtain a range-Doppler characteristic graph.

[0154] The original feature map includes a range-pulse feature map, a range-azimuth feature map and a range-Doppler feature map; the acquisition unit 810 is specifically used to determine the sub-areas corresponding to the target object in the range-pulse feature map, the range-azimuth feature map and the range-Doppler feature map according to the position information of the target object in the radar echo data; the sub-areas include the target object; the sub-areas of the target object are fused to obtain the feature map of the target object.

[0155] The sub-regions of the target object have the same size. The acquisition unit 810 is specifically used to superimpose the feature map data at the same position in each sub-region to obtain the feature map of the target object.

[0156] Optionally, no pooling layer is included in the radar target recognition model.

[0157] Optionally, the radar echo data includes echo data of sea observation and ground observation using coherent pulse radar.

[0158] Fig. 9 The present invention is a structural diagram of a radar target recognition device according to an exemplary embodiment of the present invention.

[0159] like Fig. 9 As shown, the radar target identification device 900 provided by the present disclosure includes:

[0160] An acquisition unit 910 is used to acquire radar echo data and preprocess the radar echo data to obtain an original feature map;

[0161] A processing unit 920, configured to determine a target object included in the radar echo data according to the original feature map, and determine a feature map of the target object according to the original feature map;

[0162] The recognition unit 930 is used to input the characteristic map of the target object into a preset radar target recognition model to obtain a category corresponding to the target object;

[0163] Among them, the radar target recognition model is trained using a training data set, which includes multiple groups of training data. The training data includes radar echo sample data used to train the model, feature maps of target objects included in the radar echo sample data, and labeled categories of the target objects.

[0164] Fig.10 The present invention is a structural diagram of an electronic device according to an exemplary embodiment of the present invention.

[0165] like Fig.10 As shown, the electronic device provided in this embodiment includes:

[0166] Memory 1001;

[0167] Processor 1002; and

[0168] Computer programs;

[0169] The computer program is stored in the memory 1001 and is configured to be executed by the processor 1002 to implement any of the above methods.

[0170] This embodiment also provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by a processor to implement any of the above methods.

[0171] This embodiment also provides a computer program product, including a computer program, and when the computer program is executed by a processor, any of the above methods is implemented.

[0172] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A model training method, characterized in that: The method comprises: Acquire a training data set; wherein the training data set includes multiple sets of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object; Inputting the feature map of the target object into a preset model to obtain a predicted category of the target object; According to the labeled category and predicted category of the target object, the parameters in the preset model are optimized to obtain a radar target recognition model; The step of obtaining a training data set includes: Acquire radar echo data, and pre-process the radar echo data to obtain an original feature map; the radar echo data is echo data of coherent pulse radar for sea observation and ground observation; Using a constant false alarm detection method, determining the target object included in the radar echo data according to the original feature map; wherein the constant false alarm detection obtains point trace information; Determine the location information of the target object according to the radar echo data, and determine the labeling category of the target object according to the location information of the target object; wherein the labeling category of the target object is: mark the points within the land range as category 1: vehicle; mark the points on the sea surface as category 2: ship; mark the cooperative experiment target as category 3: tugball; The original characteristic graphs include a range-pulse characteristic graph, a range-azimuth characteristic graph, and a range-Doppler characteristic graph; According to the position information of the target object in the radar echo data, a sub-region corresponding to the target object is determined in the range-pulse characteristic graph, the range-azimuth characteristic graph, and the range-Doppler characteristic graph, respectively; the sub-region includes the target object; The feature map data at the same position in each of the sub-areas are superimposed to obtain the feature map of the target object.

2. The method according to claim 1, characterized in that The preprocessing of the radar echo data to obtain the original feature map includes: Performing pulse compression on the radar echo data to obtain the range-pulse characteristic diagram; Performing moving target display on the radar echo data to obtain the range-azimuth characteristic diagram; Moving target detection is performed on the radar echo data to obtain the range-Doppler characteristic diagram.

3. A radar target recognition method, characterized in that: The method comprises: Acquire radar echo data, and preprocess the radar echo data to obtain an original feature map; A constant false alarm detection method is used to determine the target object included in the radar echo data according to the original feature map, wherein the constant false alarm detection obtains point trace information; the original feature map includes a range-pulse feature map, a range-azimuth feature map and a range-Doppler feature map; according to the position information of the target object in the radar echo data, a sub-region corresponding to the target object is determined in the range-pulse feature map, the range-azimuth feature map and the range-Doppler feature map respectively; the sub-region includes the target object; and the sub-regions of the target object are fused to obtain a feature map of the target object; Inputting the characteristic map of the target object into a preset radar target recognition model to obtain a category corresponding to the target object; The radar target recognition model is trained using a training data set, wherein the training data set includes multiple sets of training data, wherein the training data includes radar echo sample data for training the model, a feature map of a target object included in the radar echo sample data, and a labeling category of the target object; wherein the labeling category of the target object is: marking points within the land range as category 1: vehicle; marking points on the sea surface as category 2: ship; marking cooperative experimental targets as category 3: tugball; The training data set is obtained in the following manner: obtaining the radar echo data, and preprocessing the radar echo data to obtain an original feature map; the radar echo data is echo data of a coherent pulse radar for sea observation and ground observation; determining a target object included in the radar echo data according to the original feature map; determining position information of the target object according to the radar echo data, and determining a labeling category of the target object according to the position information of the target object, and determining a feature map of the target object in the original feature map.

4. A model training device, characterized in that: The device comprises: An acquisition unit, configured to acquire a training data set; wherein the training data set includes multiple sets of training data, and the training data includes radar echo data, a feature map of a target object included in the radar echo data, and a labeled category of the target object; A training unit, used for inputting the feature map of the target object into a preset model to obtain a predicted category of the target object; The training unit is further used to optimize the parameters in the preset model according to the labeled category and the predicted category of the target object to obtain a radar target recognition model; The acquisition unit is specifically used to acquire radar echo data, and pre-process the radar echo data to obtain an original feature map; use a constant false alarm detection method to determine the target object included in the radar echo data according to the original feature map; wherein the constant false alarm detection obtains point information; determines the position information of the target object according to the radar echo data, and determines the labeling category of the target object according to the position information of the target object; wherein the labeling category of the target object is: marking the point traces within the land range as category 1: vehicle; marking the point traces on the sea surface as category 2: ship; marking the cooperative experiment target as category 3: tow ball; the original feature map includes a range-pulse feature map, a range-azimuth feature map and a range-Doppler feature map; according to the position information of the target object in the radar echo data, the sub-areas corresponding to the target object are respectively determined in the range-pulse feature map, the range-azimuth feature map and the range-Doppler feature map; the sub-areas include the target object; and the feature map data of the same position in each sub-area are superimposed to obtain the feature map of the target object.

5. A radar target recognition device, characterized in that: The device comprises: An acquisition unit, used for acquiring radar echo data, and preprocessing the radar echo data to obtain an original feature map; A processing unit, configured to determine a target object included in the radar echo data according to the original feature map by using a constant false alarm detection method, wherein the constant false alarm detection obtains point trace information; and determine a feature map of the target object according to the original feature map; An identification unit, used for inputting the characteristic map of the target object into a preset radar target identification model to obtain a category corresponding to the target object; The radar target recognition model is trained using a training data set, wherein the training data set includes multiple sets of training data, wherein the training data includes radar echo sample data for training the model, a feature map of a target object included in the radar echo sample data, and a labeling category of the target object; wherein the labeling category of the target object is: marking points within the land range as category 1: vehicle; marking points on the sea surface as category 2: ship; marking cooperative experimental targets as category 3: tugball; The processing unit is specifically used for the original feature map including a range-pulse feature map, a range-azimuth feature map and a range-Doppler feature map; according to the position information of the target object in the radar echo data, the sub-regions corresponding to the target object are respectively determined in the range-pulse feature map, the range-azimuth feature map and the range-Doppler feature map; the sub-regions include the target object; and the sub-regions of the target object are fused to obtain the feature map of the target object; The acquisition unit is specifically used to acquire the radar echo data, and preprocess the radar echo data to obtain an original feature map; the radar echo data is echo data of a coherent pulse radar for sea observation and ground observation; determine a target object included in the radar echo data according to the original feature map; determine position information of the target object according to the radar echo data, and determine a labeling category of the target object according to the position information of the target object, and determine a feature map of the target object in the original feature map.

6. An electronic device, characterized in that: comprising a memory and a processor; wherein, The memory is used to store computer programs; The processor is used to read the computer program stored in the memory and execute the method according to any one of claims 1 to 2 or 3 above according to the computer program in the memory.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the processor executes the computer-executable instructions, the method described in any one of claims 1 to 2 or 3 is implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method described in any one of claims 1 to 2 or 3 is implemented.

Images