A dual-domain combined narrowband radar moving target classification and identification method and related product

Abstract

The application discloses a kind of dual-domain combined narrowband radar moving target classification identification method and related product;The target echo signal received by radar is processed, and the range doppler spectrum is obtained;Based on the target echo signal received by radar, the micro-doppler time-frequency spectrum is obtained;The range doppler spectrum and micro-doppler time-frequency spectrum are cut respectively, and the range doppler domain feature data and micro-doppler domain feature data are obtained;The range doppler domain feature data and micro-doppler domain feature data are input into the RFICNet constructed in advance, and the target classification result is obtained.The application combines the range doppler domain and micro-doppler domain dual-domain data characteristics, does not depend on complex echo data preprocessing operation and too much priori physical hypothesis, and the dual-domain combined feature information can make up for the deficiency of feature basis during classification, to improve the classification identification precision of narrowband radar moving target.

Description

Technical Field

[0001] This invention belongs to the field of narrowband radar target recognition technology, and relates to a dual-domain joint narrowband radar moving target classification and recognition method and related products. Background Technology

[0002] Narrowband radar, a detection device widely used in military reconnaissance, civilian surveillance, and air traffic management, has always faced key limitations in performance due to its inherent bandwidth and low resolution. The narrow bandwidth of narrowband radar results in a limited amount of target information that can be acquired during detection, significantly increasing the difficulty of subsequent target classification and identification.

[0003] In existing narrowband radar moving target classification and recognition methods, most rely on manual feature extraction and parameter estimation based on physical models. These methods typically require complex preprocessing of radar echo signals to extract feature parameters that reflect target characteristics, such as velocity, acceleration, and shape information. However, due to the resolution limitations of narrowband radar, these feature parameters are often not accurate or comprehensive enough, thus affecting the accuracy of classification and recognition.

[0004] Furthermore, existing classification and recognition methods often struggle to fully utilize the target's feature information in radar echoes when processing narrowband radar data. On one hand, traditional signal processing methods may overlook certain details that significantly impact classification and recognition when extracting features; on the other hand, existing machine learning algorithms may face problems such as high computational complexity and poor model generalization ability when processing high-dimensional data, thus limiting their effectiveness in practical applications. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problem that existing classification and recognition methods are unable to fully utilize the target feature information of narrowband radar, resulting in inaccurate identification of moving targets. This invention provides a dual-domain joint narrowband radar moving target classification and recognition method and related products.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] The first aspect of this invention provides a dual-domain joint narrowband radar moving target classification and identification method, comprising the following steps:

[0008] The target echo signal received by the radar is processed to obtain the range Doppler spectrum;

[0009] The micro-Doppler time spectrum is obtained based on the target echo signal received by the radar.

[0010] The range-Doppler spectrum and the micro-Doppler time spectrum are cropped separately to obtain the range-Doppler domain feature data and the micro-Doppler domain feature data;

[0011] Input the distance Doppler domain feature data and the micro Doppler domain feature data into the pre-constructed RFICNet to obtain the classification and recognition results;

[0012] The pre-built RFICNet includes an input module, two feature extraction modules, two spatial attention modules, and a classification module. The input module is connected to the feature extraction module, and each feature extraction module is followed by a spatial attention module. The spatial attention module is connected to the classification module. The input module is used to normalize the input data. The feature extraction module uses max pooling. The spatial attention module includes max pooling and average pooling. The classification module uses three fully connected layers.

[0013] Furthermore, the process of processing the target echo signal received by the radar to obtain the range Doppler spectrum is as follows:

[0014] The target echo signal received by the radar is mixed to obtain the baseband signal;

[0015] The baseband signal is processed to obtain the range Doppler spectrum.

[0016] Furthermore, the mixing process performed on the target echo signal received by the radar to obtain the baseband signal specifically involves:

[0017] For a linear frequency modulated pulse radar, the target echo signal received by the radar is digitally down-converted to obtain the baseband signal.

[0018] For linear frequency modulated continuous wave radar, the target echo signal received by the radar is deskewed to obtain the baseband signal.

[0019] Furthermore, the micro-Doppler time spectrum obtained based on the target echo signal received by the radar is specifically as follows:

[0020] The target echo signal received by the radar is mixed to obtain the baseband signal;

[0021] The baseband signal is processed to obtain the range pulse spectrum;

[0022] The micro-Doppler time spectrum is obtained based on the distance pulse spectrum.

[0023] Furthermore, the micro-Doppler time spectrum obtained based on the range pulse spectrum is specifically as follows:

[0024] Slow-time dimension data corresponding to the range unit of the target in the range pulse spectrum is selected, and time-frequency transformation is performed on the selected slow-time dimension data to obtain the micro-Doppler time spectrum; the range unit of the target is the range unit of the target obtained after constant false alarm rate detection and point trace aggregation processing of the range Doppler spectrum.

[0025] Furthermore, the step of cropping the range-Doppler spectrum and the micro-Doppler time spectrum to obtain range-Doppler domain feature data and micro-Doppler domain feature data respectively is as follows:

[0026] After constant false alarm rate detection of the range Doppler spectrum, point clustering is performed to obtain the range cell and Doppler cell of the target.

[0027] By using the range cell and Doppler cell corresponding to the condensed target in the range-Doppler spectrum as the center, the range-Doppler spectrum is cropped to obtain the range-Doppler domain feature data.

[0028] The micro-Doppler time spectrum is cropped by taking the frequency data with the largest amplitude in the micro-Doppler time spectrum as the center.

[0029] The cropped micro-Doppler time spectrum is stretched and downsampled to obtain micro-Doppler domain feature data; the micro-Doppler domain feature data has the same dimension as the distance Doppler domain feature data.

[0030] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described dual-domain joint narrowband radar moving target classification and recognition method.

[0031] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described dual-domain joint narrowband radar moving target classification and recognition method.

[0032] A fourth aspect of the present invention provides a computer program product comprising computer instructions that instruct a computer to execute the above-described dual-domain joint narrowband radar moving target classification and identification method.

[0033] A fifth aspect of the present invention provides a dual-domain joint narrowband radar moving target classification and recognition system, comprising:

[0034] The moving target detection module is used to process the target echo signal received by the radar to obtain the range Doppler spectrum;

[0035] The time-frequency analysis module is used to obtain the micro-Doppler time spectrum based on the target echo signal received by the radar;

[0036] The dual-domain data preprocessing module is used to cut the range-Doppler spectrum and the micro-Doppler time spectrum separately to obtain range-Doppler domain feature data and micro-Doppler domain feature data;

[0037] The target classification and recognition module is used to input the distance Doppler domain feature data and the micro Doppler domain feature data into the pre-constructed RFICNet to obtain the target classification results;

[0038] The pre-built RFICNet includes an input module, two feature extraction modules, two spatial attention modules, and a classification module. The input module is connected to the feature extraction module, and each feature extraction module is followed by a spatial attention module. The spatial attention module is connected to the classification module. The input module is used to normalize the input data. The feature extraction module uses max pooling. The spatial attention module includes max pooling and average pooling. The classification module uses three fully connected layers.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] This invention discloses a dual-domain joint narrowband radar moving target classification and recognition method. By combining range-Doppler and micro-Doppler domain features, the method integrates these two domains. Range-Doppler features provide information such as the target's distance, velocity, amplitude, range-dimensional extension characteristics, Doppler-dimensional extension characteristics, and background clutter characteristics along the radar's line-of-sight. Micro-Doppler time-spectrum features reveal the Doppler frequency shift differences caused by relative motion in different parts of the target. The combination of these two methods more accurately describes the target's fused feature information, comprehensively capturing the target's motion characteristics and structural information, thus achieving more precise classification of moving targets. Furthermore, by pruning the range-Doppler and micro-Doppler time-spectrum data, key feature data is extracted, effectively reducing the interference of redundant information on the classification and recognition process. This not only improves data processing efficiency but also enhances the representativeness of the feature data, making the classification and recognition results more reliable.

[0041] Furthermore, this invention selects data by taking the target centroid in the range-Doppler spectrum as the center and the frequency data with the largest amplitude in the micro-Doppler spectrum as the center. This can ignore the interference caused by useless position information, reduce the amount of computation, and avoid the accuracy reduction problem caused by the sensitivity of the convolutional neural network in RFICNet to the position information of two-dimensional input feature data.

[0042] Furthermore, this invention selects a certain size of effective region as data sample for both range Doppler and micro Doppler domain feature data, avoiding interference caused by too much irrelevant data, greatly reducing the amount of input data for RFICNet, improving the efficiency of data sample learning, training, and classification, and also making the radar target classification method more versatile. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A schematic diagram of a dual-domain joint narrowband radar moving target intelligent classification and recognition method provided in an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the radar fusion information classification and recognition network in an embodiment of the present invention;

[0046] Figure 3 The confusion matrix is ​​used to verify the classification results of the experiments in this embodiment of the invention;

[0047] Figure 4 This is a block diagram of the dual-domain joint narrowband radar moving target intelligent classification and recognition method of the present invention;

[0048] Figure 5 This is a block diagram of the dual-domain joint narrowband radar moving target intelligent classification and recognition system of the present invention.

[0049] Among them: 501-Moving target detection module; 502-Time-frequency analysis module; 503-Dual-domain data pre-processing module; 504-Target classification and recognition module. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and marked in the accompanying drawings can generally be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0052] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0053] The present invention will now be described in further detail with reference to the accompanying drawings:

[0054] See Figure 4 One embodiment of the present invention discloses a dual-domain joint narrowband radar moving target classification and identification method, comprising the following steps:

[0055] S1, Process the target echo signal received by the radar to obtain the range Doppler spectrum;

[0056] S2, based on the target echo signal received by the radar, the micro-Doppler time spectrum is obtained;

[0057] S3, the range-Doppler spectrum and the micro-Doppler time spectrum are cropped respectively to obtain the range-Doppler domain feature data and the micro-Doppler domain feature data;

[0058] S4. Input the distance Doppler domain feature data and micro Doppler domain feature data into the pre-constructed RFICNet to obtain the classification and recognition results;

[0059] The pre-built RFICNet includes an input module, two feature extraction modules, two spatial attention modules, and a classification module. The input module is connected to the feature extraction module, and each feature extraction module is followed by a spatial attention module. The spatial attention module is connected to the classification module. The input module is used to normalize the input data. The feature extraction module uses max pooling. The spatial attention module includes max pooling and average pooling. The classification module uses three fully connected layers.

[0060] See Figure 1 One embodiment of the present invention discloses a dual-domain joint narrowband radar moving target classification and identification method, comprising the following steps:

[0061] S1 processes the target echo signal received by the radar to obtain the range Doppler spectrum; based on the range Doppler spectrum, the range cell and Doppler cell information of the target are obtained.

[0062] S101, the target echo signal received by the radar is mixed to obtain the baseband signal. In this embodiment, echo data collected by an FMCW (Frequency Modulated Continuous Wave) radar is used for analysis. For an FMCW radar, it is assumed that the target echo signal received by the radar is... The radar system transmits signals as The baseband signal after mixing the target echo signal received by the radar is:

[0063]

[0064] in, LPF This indicates a low-pass filter.

[0065] S102, 2D-FFT (Two-Dimensional Fast Fourier Transform) is used to implement MTD (Moving Target Detection) on the baseband signal to obtain the range-Doppler spectrum; the range-Doppler spectrum is represented as 4096. A 512 matrix, where rows are distance units, columns are Doppler frequencies, and values ​​represent amplitudes.

[0066] S103, the range-Doppler spectrum is processed using CFAR (Constant False Alarm Rate) detection and point clustering to obtain the target's range cell and Doppler cell information. For single-target scenes, CA-CFAR (Cell Averaging CFAR) is used for CFAR detection, while SO-CFAR (Smallest Of CFAR) is used for multi-target scenes. Point clustering is performed using a point clustering method based on the center of mass approach.

[0067] S2, based on the target echo signal received by the radar, obtains the micro-Doppler time spectrum of the target echo.

[0068] S201, perform a 1D-FFT (One Dimensional Fast Fourier Transform) on the baseband signal in step S101 to obtain the range pulse spectrum.

[0069] S202, based on the target range cell obtained in step S103, select the slow time dimension data corresponding to the target range cell in the range pulse spectrum as the target signal. For target signal In this embodiment, a short-time Fourier transform is used to perform a time-frequency transformation on the target signal to obtain the micro-Doppler time spectrum of the target echo. The time-frequency transformation method is described as follows:

[0070]

[0071] in, Angular frequency, For analysis window functions.

[0072] The micro-Doppler time spectrum is described as a micro-Doppler domain time spectrum matrix, where the matrix's rows represent time, the columns represent Doppler frequencies, and the values ​​represent amplitudes.

[0073] S3, the range-Doppler spectrum and the micro-Doppler time spectrum are cropped respectively to obtain the range-Doppler domain feature data and the micro-Doppler domain feature data;

[0074] S301, for the range-Doppler spectrum in step S1, based on the target's range and Doppler cell information obtained in S103, a fixed-size region is selected and cropped using the range and Doppler cells corresponding to the condensed target in the range-Doppler spectrum as the center, generating range-Doppler domain feature data. This range-Doppler domain feature data is described as a matrix, with rows representing range cells and columns representing Doppler cells. This matrix serves as the data sample for the range-Doppler domain in the dual-domain joint identification. This invention obtains range-Doppler domain feature data by selecting a fixed-size effective region centered on the range and Doppler cells corresponding to the condensed target. This removes meaningless range and velocity information and excessive background clutter information in target type determination, while also reducing the size of the two-dimensional data. The range-Doppler domain feature data includes target information such as the target's average energy characteristics, range extension characteristics, Doppler extension characteristics, and background clutter characteristics.

[0075] S302, for the micro-Doppler time spectrum generated in step S202, the Doppler frequency data with the largest amplitude is used as the sample center, and a fixed-size region is selected to crop the micro-Doppler time spectrum to obtain micro-Doppler domain feature data; the micro-Doppler domain feature data is stretched and appropriately downsampled to make the micro-Doppler domain feature data to be input into RFICNet (Radar Fusion Information Classification Network) consistent with the range Doppler domain feature data in dimension; this stretched and appropriately downsampled micro-Doppler domain feature data is used as the micro-Doppler domain sample in the dual-domain joint training.

[0076] The stretching and downsampling processes described in this embodiment result in the rows of the micro-Doppler domain time-spectrum matrix (i.e., micro-Doppler domain feature data) having no specific meaning, but still reflecting time dimension information; the columns of the micro-Doppler domain time-spectrum matrix have no specific meaning, but still reflect frequency dimension information; the stretching operation makes the frequency extension characteristics reflected by the micro-Doppler domain time-spectrum matrix as obvious as possible, and the downsampling operation reduces the sample data size, thereby improving the efficiency of narrowband radar target classification and recognition.

[0077] S4. Construct a Radar Fusion Information Classification Network (RFICNet) for radar target classification and recognition.

[0078] See Figure 2 The RFICNet constructed in this invention consists of an input module, a feature extraction module, a spatial attention module, and a classification and recognition module.

[0079] The input module is used to normalize the input two-dimensional data with two domains so that the data range is [0,1].

[0080] Two feature extraction modules are used to extract features at different scales. In the first feature extraction module, the convolutional layer has a kernel size of 3×3 and a number of kernels of 10. Max pooling is used with a kernel size of 2×2 and a pooling operation stride of 2. The ReLU activation function is used. In the second feature extraction module, the convolutional layer has a kernel size of 3×3 and a number of kernels of 2. Max pooling is used with a kernel size of 2×2 and a pooling operation stride of 2. The ReLU activation function is used.

[0081] Two spatial attention modules are placed after the feature extraction module. They perform global max pooling and global average pooling on the input feature map, respectively. After pooling, the downsampled output is concatted to obtain the context information feature map. After convolution with a 3×3 kernel, the dimensionality is reduced to generate spatial attention weights, which are then weighted with the feature map of the input spatial attention module.

[0082] Finally, the classification module uses three fully connected layers to classify the input feature map. The first fully connected layer has a size of 256 and uses the ReLU activation function, the second fully connected layer has a size of 256 and uses the ReLU activation function, and the third fully connected layer has a size of 64 and uses the ReLU activation function. Finally, a Softmax layer is used to output the probability distribution of the predicted categories.

[0083] In RFICNet, the input module performs data normalization, mapping the two-dimensional data uniformly to the [0, 1] interval, allowing features of different dimensions to be numerically comparable, which can improve the convergence speed and classification accuracy of the network. The feature extraction module increases the receptive field of neurons through convolution operations, realizing the extraction of semantic information from relatively detailed to abstract. Pooling extracts more representative features, reduces the dimensionality of features, and reduces the size of the feature map without changing the number of channels. The ReLU activation function increases the nonlinear relationship of the network and reduces the interdependence of parameters. The spatial attention module enables the network to adaptively focus on important two-dimensional spatial local regions, improving the effectiveness of feature extraction. The classification and recognition module recombines all local features extracted by the feature extraction module and the spatial attention module through a weight matrix, nonlinearly mapping the feature map obtained in the previous process into a one-dimensional feature vector. This feature vector contains all the extracted feature information. The input of the Softmax layer is the feature activation value of the input two-dimensional image in each category, and the output is the probability corresponding to each category.

[0084] S5 inputs the dual-domain joint training data (i.e., range-Doppler domain feature data and micro-Doppler domain feature data) into RFICNet (Radar Fusion Information Classification Network) to obtain the trained RFICNet. When classifying and recognizing unknown types of targets, the echo signal of the moving target is processed by S1-S3 and then input into the trained RFICNet to obtain the target classification result.

[0085] To demonstrate the superiority of this invention, this embodiment verifies the dual-domain joint narrowband radar moving target classification and identification method described in this invention using echo data from low-altitude drones, pedestrians, and vehicles.

[0086] The narrowband radar system collects target echo data from low-altitude UAVs, pedestrians, and vehicles. After processing through steps S1-S3, training samples of range Doppler domain feature data and micro Doppler domain feature data are generated, i.e., dual-domain joint training sample data. A corresponding category label is set for the sample data corresponding to each type of target. At the same time, all the obtained dual-domain joint training samples are divided into training set, validation set, and test set in a ratio of 7:2:1 for training and testing the radar fusion information classification and recognition network.

[0087] Drones fly at low altitudes, making it difficult to distinguish them using elevation and azimuth information estimated from radar target parameters. At the same time, drones and vehicles travel at relatively slow speeds, similar to the radial speed of pedestrians relative to radar, making it difficult to identify them using simple target speed information.

[0088] In the range-Doppler domain training samples, due to the high RCS (Radar Cross Section), the vehicle's echo power is usually large. According to the radar equation, the relationship between the received echo power and the RCS can be expressed as:

[0089]

[0090] in, The radial distance between the radar and the target. For system losses, For transmission power, For radar transmitter gain, For radar receiver gain, For the emission wavelength, This refers to the gain of the signal processing system.

[0091] It's understandable that vehicle echoes exhibit significant extension in the distance dimension due to the vehicle's large target size, occupying multiple distance units. However, they occupy very few units in the Doppler dimension because the components that generate the micro-Doppler effect, namely the wheels, constitute a small proportion of the vehicle's overall size. The motion of a moving vehicle can be divided into two parts: the translation of the vehicle body and the rotation of the wheels. The translation of the vehicle body generates the Doppler effect, while the rotation of the wheels generates the micro-Doppler effect. In actual moving vehicles, due to the low backscattering of tire rubber and the fact that the wheels represent a much smaller proportion of the vehicle's size, the micro-Doppler components are difficult to observe.

[0092] Meanwhile, the Doppler range samples of pedestrians exhibit more irregularities. This is because the RCS (Radar Cross Section) of the human body varies greatly at different angles, leading to amplitude variability. Furthermore, the movement of body parts during walking causes the echo to extend to varying degrees in the Doppler dimension. Also, because pedestrians walk relatively slowly, their echoes often appear near static clutter, which means that the area near the pedestrian target echo cell is often accompanied by clutter covering the entire range dimension.

[0093] Compared to pedestrians and vehicles, drones have a smaller RCS (Radar Cross Section), which translates to a lower amplitude in the range Doppler domain.

[0094] In the micro-Doppler domain time-spectrum training samples, the micro-Doppler domain feature data of the vehicle target did not show obvious micro-Doppler phenomena. The only approximation was the obvious Doppler component caused by the vehicle translation. The samples reflected a straight spectral line in the center. At the same time, due to the large echo power of the vehicle, the time-spectrum line also showed a high amplitude, and the energy spread significantly across the frequency range.

[0095] The motion state of a UAV target is divided into spatial movement and rotor rotation. The rotor rotation generates the micro-Doppler effect of the UAV target. UAV rotors generally have a high reflection coefficient. For traditional narrowband radar, the electromagnetic scattering can be analyzed using a point scattering model. Each rotor on the UAV is regarded as a scattering center, and each scattering center is independent of the others. Therefore, the total scattering component of the UAV rotor can be regarded as the linear sum of the scattering echoes from each blade.

[0096] Assuming the initial distance from the rotor center to the radar is The height is The azimuth angle relative to the radar line of sight is The pitch angle is The angular velocity of the rotor rotation speed is The distance from the blade to the rotor center is The radar echo signal generated by a scattering point can be expressed as:

[0097]

[0098] in, For wavelength, For the initial phase, The Doppler frequency is generated by the radial velocity of the target relative to the radar.

[0099] The echo from the blade can be considered as the superposition of scattering points. After removing the carrier frequency from the radar echo signal, the distance along the blade length... Integrating the above equation yields the radar echo signal of a single blade:

[0100]

[0101] If each rotor is composed of Composed of 10 blades, the radar echo signal of a single rotor is:

[0102]

[0103] Therefore, the echo signal of a single rotor of a UAV exhibits the same characteristics in the time domain as... The function exhibits consistent periodic flickering, with a periodicity in the time domain. Depending on the rotor speed and the number of blades, it can be expressed as:

[0104]

[0105] in , It is a constant whose value depends on the number of blades in a single rotor. ,when When it is an odd number , When it is even .

[0106] The radar echo signal of a single rotor is converted to the time-frequency domain by performing a short-time Fourier transform, and the time spectrum of the single rotor echo is obtained. The time-frequency domain modulation characteristics of a single rotor are characterized by periodic spectral line fluctuations centered on the Doppler frequency shift. The micro-Doppler characteristics of the UAV are reflected as the superposition of the micro-Doppler characteristics of multiple rotors.

[0107] In this embodiment, the UAV is a quadcopter. The Doppler component caused by the spatial motion of the main body occupies the main part of the time spectrum and has high energy. Since the UAV rotor has a high rotation speed during flight, the time spectrum exhibits continuous periodic micro-Doppler characteristics. Its graphic characteristics are reflected as periodic fluctuations similar to a sine function. At the same time, because the UAV is small in size, the backscattering is weak, and some micro-Doppler characteristics are submerged in the noise background.

[0108] For radar echo signals of pedestrian targets, the baseline of the spine is defined as the height of the human body. At the 58th position, a human body coordinate system is established with this as the origin to describe the human walking model. In the case of far-field plane waves of narrowband low-resolution radar, the electromagnetic scattering of the human body can also be described by the point scattering model.

[0109] The pedestrian's motion state, based on the Boulic global walking model, can be roughly modeled as 12 scattering points, namely the head (…). ),trunk( ), left and right upper arms ( , ), left and right thighs ( , ), left and right thighs ( , ), left and right calves ( , ), left and right feet ( , ).

[0110] In a simplified form, the motion of the human body is represented by the trajectory of the scattering points, and the electromagnetic scattering intensity of each part is represented by its respective RCS. The motion of a pedestrian can be divided into the overall translation of the body and the relative motion between each part and the body's center of mass. The motion of each part of the body can be considered to have obvious periodicity, which can be simply described as the left-right swaying of the torso and the swaying of the limbs. Approximately, the following sine function can be used to describe the motion of the pedestrian's... Each part Relative motion on the axis relative to the origin of the established human body coordinate system:

[0111]

[0112] in, For the first Micro-movement parts swing amplitude For the first Micro-movement parts The oscillation frequency, For the first Micro-movement parts The initial phase of motion is the initial angle between the corresponding micro-moving part and the torso.

[0113] Based on the Boulic empirical model of human walking, it can be assumed that the swaying frequency of the head and torso during walking is twice the frequency of human motion, while the other micro-movement parts have the same motion frequency as walking, that is:

[0114]

[0115] Assuming the radar is the origin of the radar coordinate system, and the initial distance from the pedestrian's centroid is... azimuth angle is The pitch angle is The pedestrian's radial velocity relative to the radar is For each of the aforementioned micro-motion parts, their positions in the created radar coordinate system can be considered as follows:

[0116]

[0117] Furthermore, the human body's first The radar echo signal delay of a micro-movement part can be expressed as:

[0118]

[0119] Assuming that the mutual occlusion between different parts of the human body can be ignored, the radar echo signal caused by the micro-motion effect of the human body can be approximated as follows:

[0120]

[0121] in, This indicates the number of micro-movement parts, i.e., scattering points, and represents the th... RCS of each scattering point The wavelength of the radar carrier wave.

[0122] Perform a short-time Fourier transform on the radar echo signal to obtain the micro-Doppler domain matrix of the pedestrian target echo. It can be understood that in the micro-Doppler domain matrix of this embodiment, the time-frequency characteristics of the pedestrian target are mainly caused by the swaying of the torso, which is the main basis for the classification of the micro-Doppler domain matrix. The micro-Doppler matrix caused by the swaying of the limbs is basically submerged by background noise due to the small reflection coefficient and easy obstruction.

[0123] The training samples from low-altitude drones, pedestrians, and vehicles are input into RFICNet for training. During the training process of RFICNet, the learning rate is... It has a significant impact on the training speed and stability of the model. The learning rate is dynamically adjusted during training to control its decline, allowing the model to more precisely approximate the optimal solution in the later stages of training; batch size. The number of samples used in each iteration of the model is determined; the optimizer determines the algorithm used to update the model weights. In this embodiment, the hyperparameters are adjusted during training using the validation set, and the initial learning rate is set to... ;use Learning rate adjuster and will adjust , Setting the batch size to 32 and using mini-batch gradient descent can improve training efficiency and generalization ability.

[0124] After training, validation is performed using a validation set, based on the classification accuracy. To evaluate the classification accuracy of the method of this invention, the validation set of this embodiment... With a success rate of 95.3%, the dual-domain joint narrowband radar moving target classification and identification method of the present invention can be considered to have good performance.

[0125] Validate RFICNet's generalization ability using the test set and based on classification accuracy. Evaluate the performance of the detection model. The expression is as follows:

[0126]

[0127] test set The accuracy rate can reach 94.4%. Test results show that the present invention can fit the training data well, has good generalization ability, and demonstrates good detection performance. The confusion matrix of the test set classification recognition is as follows: Figure 3As shown, each column of the confusion matrix represents the predicted category, and the total number in each column represents the number of data points predicted as belonging to that category; each row of the confusion matrix represents the true category to which the data belongs, and the total number of data points in each row represents the number of data instances belonging to that category; the value in each column represents the number of actual data points predicted as belonging to that category.

[0128] In summary, this invention provides a dual-domain joint narrowband radar moving target classification and recognition method. By jointly training feature data from the range Doppler and micro-Doppler domains of radar echo signals in both domains, a radar fusion information classification and recognition network based on a two-dimensional convolutional neural network is constructed. This network achieves good narrowband radar target classification performance. Compared to traditional narrowband radar target classification methods, this invention's dual-domain joint narrowband radar moving target classification and recognition method eliminates the need for complex manual feature extraction based on physical assumptions and extensive mathematical statistics. It intelligently extracts joint features from both domains, avoiding problems such as insufficient feature extraction or excessive redundant features for classification, resulting in higher classification accuracy and efficiency.

[0129] See Figure 5 One embodiment of the present invention proposes a dual-domain joint narrowband radar moving target classification and recognition system, comprising:

[0130] The moving target detection module 501 is used to process the target echo signal received by the radar to obtain the range Doppler spectrum;

[0131] The time-frequency analysis module 502 is used to obtain the micro-Doppler time spectrum based on the target echo signal received by the radar;

[0132] The dual-domain data preprocessing module 503 is used to cut the range-Doppler spectrum and the micro-Doppler time spectrum respectively to obtain range-Doppler domain feature data and micro-Doppler domain feature data;

[0133] The target classification and recognition module 504 is used to input the distance Doppler domain feature data and the micro Doppler domain feature data into the pre-constructed RFICNet to obtain the target classification result;

[0134] The pre-built RFICNet includes an input module, two feature extraction modules, two spatial attention modules, and a classification module. The input module is connected to the feature extraction module, and each feature extraction module is followed by a spatial attention module. The spatial attention module is connected to the classification module. The input module is used to normalize the input data. The feature extraction module uses max pooling. The spatial attention module includes max pooling and average pooling. The classification module uses three fully connected layers.

[0135] In one embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that more specific examples (a non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0136] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0137] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0138] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the dual-domain joint narrowband radar moving target classification and identification method in the above embodiments.

[0139] One embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned dual-domain joint narrowband radar moving target classification and recognition method. The memory can be any type of storage device, such as random access memory (RAM), read-only memory (ROM), hard disk drive (HDD), solid-state drive (SSD), or cloud storage. In this embodiment, the memory stores a computer program for implementing a photovoltaic power generation prediction method. The processor is responsible for executing the computer program stored in the memory. The processor can be any type of computing device, such as a central processing unit (CPU), graphics processing unit (GPU), field-programmable gate array (FPGA), or application-specific integrated circuit (ASIC). In this embodiment, the processor executes the computer program to implement the dual-domain joint narrowband radar moving target classification and recognition method.

[0140] One embodiment of the present invention provides a computer program product comprising a series of computer instructions. These instructions are designed to guide a computer to perform a series of operations to achieve dual-domain joint narrowband radar moving target classification and recognition. These instructions can be loaded into any type of computer system, including but not limited to personal computers, servers, cloud computing platforms, etc., to execute the described dual-domain joint narrowband radar moving target classification and recognition method. The main function of the computer instructions in the computer program product of this embodiment is to guide the computer to perform the following steps to achieve the dual-domain joint narrowband radar moving target classification and recognition method.

[0141] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dual-domain joint narrowband radar moving target classification and identification method, characterized in that, Includes the following steps: The target echo signal received by the radar is processed to obtain the range Doppler spectrum; The micro-Doppler time spectrum is obtained based on the target echo signal received by the radar. The range-Doppler spectrum and the micro-Doppler time spectrum are cropped separately to obtain the range-Doppler domain feature data and the micro-Doppler domain feature data; By inputting the distance Doppler domain feature data and the micro Doppler domain feature data into the pre-constructed RFICNet, the classification and recognition results are obtained. The pre-built RFICNet includes an input module, two feature extraction modules, two spatial attention modules, and a classification and recognition module. The input module is connected to the feature extraction module, and each feature extraction module is followed by a spatial attention module. The spatial attention module is connected to the classification and recognition module. The input module is used to normalize the input data. The feature extraction module uses max pooling. The spatial attention module includes both max pooling and average pooling. The classification and recognition module uses three fully connected layers. The micro-Doppler time spectrum obtained based on the target echo signal received by radar is as follows: The target echo signal received by the radar is mixed to obtain the baseband signal; The baseband signal is processed to obtain the range pulse spectrum; The micro-Doppler time spectrum is obtained based on the range pulse spectrum; specifically, the micro-Doppler time spectrum is obtained based on the range pulse spectrum. Slow-time dimension data corresponding to the range unit of the target in the range pulse spectrum is selected, and time-frequency transformation is performed on the selected slow-time dimension data to obtain the micro-Doppler time spectrum; the range unit of the target is the range unit of the target obtained after constant false alarm rate detection and point trace aggregation processing of the range Doppler spectrum; The process of cropping the range-Doppler spectrum and the micro-Doppler time spectrum to obtain range-Doppler domain feature data and micro-Doppler domain feature data respectively is as follows: After constant false alarm rate detection of the range Doppler spectrum, point clustering is performed to obtain the range cell and Doppler cell of the target. By using the range cell and Doppler cell corresponding to the condensed target in the range-Doppler spectrum as the center, the range-Doppler spectrum is cropped to obtain the range-Doppler domain feature data. The micro-Doppler time spectrum is cropped by taking the frequency data with the largest amplitude in the micro-Doppler time spectrum as the center. The cropped micro-Doppler time spectrum is stretched and downsampled to obtain micro-Doppler domain feature data; the micro-Doppler domain feature data has the same dimension as the distance Doppler domain feature data.

2. The dual-domain joint narrowband radar moving target classification and identification method according to claim 1, characterized in that, The process of processing the target echo signal received by the radar to obtain the range Doppler spectrum is as follows: The target echo signal received by the radar is mixed to obtain the baseband signal; The baseband signal is processed to obtain the range Doppler spectrum.

3. The dual-domain joint narrowband radar moving target classification and identification method according to claim 2, characterized in that, The process of mixing the target echo signal received by the radar to obtain the baseband signal is specifically as follows: For a linear frequency modulated pulse radar, the target echo signal received by the radar is digitally down-converted to obtain the baseband signal. For linear frequency modulated continuous wave radar, the target echo signal received by the radar is deskewed to obtain the baseband signal.

4. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the dual-domain joint narrowband radar moving target classification and identification method according to any one of claims 1-3.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the dual-domain joint narrowband radar moving target classification and identification method according to any one of claims 1-3.

6. A computer program product, said computer program product comprising computer instructions, characterized in that, The computer instructions instruct the computer to execute the dual-domain joint narrowband radar moving target classification and identification method as described in any one of claims 1-3.

7. A dual-domain joint narrowband radar moving target classification and recognition system, based on the dual-domain joint narrowband radar moving target classification and recognition method of claim 1, characterized in that, include: The moving target detection module is used to process the target echo signal received by the radar to obtain the range Doppler spectrum; The time-frequency analysis module is used to obtain the micro-Doppler time spectrum based on the target echo signal received by the radar; The dual-domain data preprocessing module is used to cut the range-Doppler spectrum and the micro-Doppler time spectrum separately to obtain range-Doppler domain feature data and micro-Doppler domain feature data; The target classification and recognition module is used to input the distance Doppler domain feature data and the micro Doppler domain feature data into the pre-constructed RFICNet to obtain the target classification results; The pre-built RFICNet includes an input module, two feature extraction modules, two spatial attention modules, and a classification module. The input module is connected to the feature extraction module, and each feature extraction module is followed by a spatial attention module. The spatial attention module is connected to the classification module. The input module is used to normalize the input data. The feature extraction module uses max pooling. The spatial attention module includes max pooling and average pooling. The classification module uses three fully connected layers.

Images