Multi-band echo coherent registration method and system based on deep learning
Through a deep learning-based method, the phase deviation of radar echoes is predicted using deep learning network models, and the problem of large amount of fixed phase term estimation and susceptible to noise in the prior art is solved, and high-precision multi-band echo phase-registration registration is achieved.
Patent Information
- Application Number
- CN202510497333.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-07-22
AI Technical Summary
In the existing multi-band phase-registration registration method, the estimation amount of fixed phase terms is large, the estimation results are susceptible to noise, and the accuracy is poor.
Using a deep learning method, by training a deep learning network model, the real and imaginary data of the first radar echo and the second radar echo are used to predict the phase deviation and phase compensation is performed on the second radar echo to achieve phase reference registration of multi-band echo.
High-precision phase compensation is achieved in a low signal-to-noise ratio environment, reducing the calculation amount, improving the accuracy and adaptability of phase registration, and reducing the sensitivity to noise.
Smart Images
Figure CN120352846A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radar signal processing, and in particular, to a multi-band echo coherent registration method and system based on deep learning. Background Art
[0002] Improving the radar target resolution is of great significance in the fields of target detection, imaging, and recognition. Increasing the signal bandwidth can effectively improve the resolution. The implementation methods include updating the hardware to expand the bandwidth and synthesizing ultra-wideband signals through signal processing. In view of the hardware cost and implementation difficulties, the radar signal processing method of multi-band sub-band fusion is actually mostly used to expand the bandwidth. However, due to the differences in the spatial position, frequency band, and initial time of multiple radars, the received echoes are non-coherent. Therefore, the key to realizing frequency band fusion lies in the coherent registration of multiple frequency bands.
[0003] The phase deviation between multi-band data includes a linear phase term and a fixed phase term. The former is due to different radar observation positions, and the latter is caused by the difference in the initial transmission phase. Coherent registration is to estimate and compensate for these phase deviations. Currently, there are many estimation methods for linear phase differences and the accuracy is high, while the estimation accuracy of the fixed phase term is poor. The specific estimation methods are mainly of two types: one is to use the echo of a certain radar as a reference, construct a coherence function, and search for non-coherent parameters through an optimization algorithm; the other is the pole coherence of the all-pole model, combined with relevant parameter estimation methods to inversely deduce the phase term value.
[0004] However, both of the above two estimation methods have obvious deficiencies. The algorithm of constructing a coherence function and searching for solutions has high parameter estimation accuracy, but it is only applicable to high signal-to-noise ratio conditions, and has problems such as large computational complexity and being easily trapped in local minima. The algorithm based on the pole coherence of the all-pole model has the application premise of accurately estimating the poles of the all-pole model. In practice, due to the interference of factors such as noise, it is difficult to accurately estimate the model order, and in the face of complex targets with numerous scattering points, which may exceed the number of poles that can be estimated, this algorithm is also difficult to apply. Summary of the Invention
[0005] The problem to be solved by the present invention is that in the existing coherent registration method, the estimation amount of the fixed phase term is large, the estimation result is easily affected by noise, and the accuracy is poor.
[0006] To solve the above problems, in the first aspect, the present invention provides a multi-band echo coherent registration method based on deep learning, including:
[0007] Obtain the radar echo data after velocity compensation, where the radar echo data includes the first radar echo and the second radar echo;
[0008] Take the real and imaginary parts of the first radar echo and the real and imaginary parts of the second radar echo as input data and send them into the trained deep learning network model to obtain a predicted phase deviation, where the phase deviation is the phase deviation of the second radar echo relative to the first radar echo;
[0009] Perform phase compensation on the second radar echo according to the predicted phase deviation to obtain a corrected second radar echo, where there is coherence between the corrected second radar echo and the first radar echo.
[0010] Optionally, the training process of the deep learning network model includes:
[0011] In the constructed simulation scenario, randomly generate observed targets, use the observed targets as the monitoring objects, and generate a training data set, where the training data set includes multiple groups of data, and each group of data includes a first radar echo, a second radar echo, and an actual phase deviation, and the actual phase deviation is the phase deviation between the second radar echo and the first radar echo;
[0012] Input the real and imaginary parts of the first radar echo and the real and imaginary parts of the second radar echo into the initial deep learning network model at the same time to obtain a predicted phase deviation;
[0013] Determine the loss value according to the predicted phase deviation and the actual phase deviation;
[0014] Adjust the parameters of the initial deep learning network model according to the loss value;
[0015] Use multiple groups of data in the training data set to perform cyclic training and adjustment on the initial deep learning network model until the loss value converges within the expected range or the number of cyclic training times reaches the preset number of times, then stop training to obtain the trained deep learning network model.
[0016] Optionally, the step of randomly generating observed targets in the constructed simulation scenario and using the observed targets as the monitoring objects to generate a training data set includes:
[0017] Place two radars adjacent to each other with the same observation angle. The observed target consists of multiple strong scattering points. Set the observed target to move towards the radars. Assume that the starting frequency of the first radar is f c , the frequency hopping interval is Δf, and the number of frequency steps is N; the starting frequency of the second radar is f c +N c Δf, the frequency hopping interval is Δf, and the number of frequency steps is N;
[0018] Monitor the observed target with the two radars, perform velocity compensation on the echoes received by the two radars to obtain the first radar echo and the second radar echo;
[0019] Based on the first radar echo, the expressions of the first radar echo and the second radar echo are deformed to obtain the actual phase deviation. Among them, each first radar echo, the corresponding second radar echo, and the corresponding actual phase deviation are used as the standard data group in the training data set;
[0020] Randomly select the standard data group, add noise to the standard data group to generate an extended data group. Among them, the standard data group and the extended data group constitute the training data set.
[0021] Optionally, the deformed first radar echo is:
[0022]
[0023] The deformed second radar echo is:
[0024]
[0025] The actual phase deviation is:
[0026] e j(kα+β)
[0027]
[0028] Among them, p represents the number of strong scattering points contained in the observed target, A i represents the scattering intensity of the i-th scattering point, R1 represents the distance between the observed target and the first radar, R2 represents the distance between the observed target and the second radar, f c represents the starting frequency of the first radar, Δf represents the frequency hopping interval, N represents the number of frequency steps, f c +N c Δf represents the starting frequency of the second radar, c represents the speed of light, represents the initial phase of the first radar echo, represents the initial phase of the second radar echo, j represents the imaginary unit, k represents the number of frequency hopping intervals, α represents the linear phase, and β represents the fixed phase.
[0029] Optionally, the loss value is:
[0030]
[0031] Among them, y i represents the i-th predicted phase deviation, y ′ i represents the i-th actual phase deviation, and N represents the total number of training data used.
[0032] Optionally, the deep learning network model is a 1D-Unet network model.
[0033] In a second aspect, the present invention further provides a multi-band echo coherent registration system based on deep learning, including:
[0034] A data acquisition module for acquiring radar echo data after velocity compensation, where the radar echo data includes a first radar echo and a second radar echo;
[0035] A deviation prediction module for using the real part and the imaginary part of the first radar echo and the real part and the imaginary part of the second radar echo as input data and sending them into a trained deep learning network model to obtain a predicted phase deviation, where the phase deviation is the phase deviation of the second radar echo relative to the first radar echo;
[0036] An echo correction module for performing phase compensation on the second radar echo according to the predicted phase deviation to obtain a second radar corrected echo, where there is coherence between the second radar corrected echo and the first radar echo.
[0037] Optionally, the multi-band echo coherent registration system based on deep learning further includes a model training module, and the model training module is used for:
[0038] In a constructed simulation scenario, randomly generate an observation target, use the observation target as the monitoring object, and generate a training data set, where the training data set includes multiple groups of data, and each group of data includes a first radar echo, a second radar echo, and an actual phase deviation, and the actual phase deviation is the phase deviation between the second radar echo and the first radar echo;
[0039] Simultaneously input the real part and the imaginary part of the first radar echo and the real part and the imaginary part of the second radar echo into an initial deep learning network model to obtain a predicted phase deviation;
[0040] Determine a loss value according to the predicted phase deviation and the actual phase deviation;
[0041] Adjust the parameters of the initial deep learning network model according to the loss value;
[0042] Use multiple groups of data in the training data set to perform cyclic training and adjustment on the initial deep learning network model until the loss value converges within an expected range or the number of cyclic training times reaches a preset number, then stop training to obtain a trained deep learning network model.
[0043] In a third aspect, the present invention provides an electronic device, including a memory and a processor;
[0044] The memory is used for storing a computer program;
[0045] The processor is configured to implement the multi-band echo coherent registration method based on deep learning as described in the first aspect when executing the computer program.
[0046] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the multi-band echo coherent registration method based on deep learning as described in the first aspect is implemented.
[0047] The present invention provides a multi-band echo coherent registration method and system based on deep learning. Compared with the prior art, the following beneficial effects are achieved:
[0048] By constructing a deep learning network model, the phase deviation relationship between the echoes of two radars is learned, and noise can be added during the training process to improve the adaptability of the deep learning network model. By inputting the real and imaginary parts of the echoes of the two radars into the trained deep learning network model, the predicted phase deviation can be obtained. Taking the echo of one radar as the reference echo, the echo data of the other radar is phase-compensated using the predicted phase deviation, thereby realizing the coherent registration of multi-band echoes. After training, the deep learning network model can obtain good prediction results in an environment with low signal-to-noise ratio and has a small computational load. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0050] Figure 1 It is a schematic flowchart of a multi-band echo coherent registration method based on deep learning provided by an embodiment of the present invention.
[0051] Figure 2 It is a schematic diagram of the multi-band echoes of two radars before phase compensation in the first test;
[0052] Figure 3 It is a schematic diagram of the multi-band echoes of two radars after phase compensation in the first test;
[0053] Figure 4 It is a schematic diagram of the multi-band echoes of two radars before phase compensation in the second test;
[0054] Figure 5 It is a schematic diagram of the multi-band echoes of two radars after phase compensation in the second test;
[0055] Figure 6 Schematic diagram of the structure of a multi - band echo coherent registration system based on deep learning provided by an embodiment of the present invention. Specific implementation manners
[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application are described clearly and completely. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.
[0057] To better understand the above - mentioned technical solutions, the above - mentioned technical solutions will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners.
[0058] As Figure 1 shown, a multi - band echo coherent registration method based on deep learning provided by an embodiment of the present application includes:
[0059] S1: Obtain radar echo data after velocity compensation, where the radar echo data includes a first radar echo and a second radar echo.
[0060] Specifically, perform velocity compensation on the first radar echo and the second radar echo respectively to eliminate the phase deviation caused by velocity, and obtain the compensated first radar echo A and the compensated second radar echo B.
[0061] S2: Use the real part and imaginary part of the first radar echo and the real part and imaginary part of the second radar echo as input data and send them into a trained deep - learning network model to obtain a predicted phase deviation, where the phase deviation is the phase deviation of the second radar echo relative to the first radar echo.
[0062] Specifically, the first radar echo A and the compensated second radar echo B are complex numbers. Four - channel data is formed according to [real part of A, imaginary part of A, real part of B, imaginary part of B] and sent into a trained 1D - Unet model for prediction, and the prediction result is the phase deviation.
[0063] S3: Perform phase compensation on the second radar echo according to the predicted phase deviation to obtain a second radar corrected echo, where the second radar corrected echo has coherence with the first radar echo.
[0064] Specifically, the deep learning network model can be a 1D-Unet network model. Under the same detection and classification accuracy, the fully convolutional network Unet requires less data volume and has a relatively stable and simple structure, making it suitable for processing radar signals. By reducing all types of convolutional layers, pooling layers, and other functional layers of Unet to one dimension, 1D-Unet is obtained. 1D-Unet includes an input, an output, an encoder, a decoder, and skip connections. The encoder, which is a feature extraction network, performs four downsamplings on the input two-radar echo data after velocity compensation through convolution-pooling to obtain four levels of features. The decoder, also known as a feature fusion network, performs feature fusion on the features at each level and the features obtained through transposed convolution in a skip connection manner.
[0065] In this optional embodiment, through the constructed deep learning network model, such as the 1D-Unet network, the deep learning network model learns the phase deviation relationship between the two radar echoes through training, and noise can be added during the training process to improve the adaptability of the deep learning network model; by inputting the real and imaginary parts of the two radar echoes into the trained deep learning network model, the predicted phase deviation can be obtained. Taking one of the radar echoes as the reference echo, the predicted phase deviation is used to perform phase compensation on the other radar echo data, thereby realizing the coherent registration of multi-band echoes. Considering the noise factor during training, the deep learning network model can obtain better prediction results in an environment with low signal-to-noise ratio. Compared with the existing coherent registration methods, this method does not require estimating the number of poles and determining the model order, has a small computational amount, and can achieve better coherent registration effects at low signal-to-noise ratios.
[0066] The training process of the deep learning network model is described in detail below.
[0067] The coherent registration process of multi-band echoes based on deep learning is divided into two parts: a training stage and a prediction stage. The task of the training stage is to send the echo data after velocity compensation of Radar 1 and Radar 2 into the deep learning network 1D-Unet for training to obtain a 1D-Unet model that can be used for prediction. The prediction stage is to send the echo data of Radar 1 and Radar 2 after velocity compensation into the trained 1D-Unet model to obtain the phase deviation, and then perform phase compensation on the Radar 2 echo data to achieve the coherent registration of multi-band echoes.
[0068] The training process of the 1D-Unet network is as follows: First, the required training data set and training label set are generated through computer simulation, and then the data obtained from the simulation is used to train the network model. The specific steps are as follows:
[0069] S10: In the constructed simulation scenario, randomly generate an observation target. Using the observation target as the monitoring object, generate a training data set. The training data set includes multiple groups of data, and each group of data includes a first radar echo, a second radar echo, and an actual phase deviation. The actual phase deviation is the phase deviation between the second radar echo and the first radar echo. The specific content of this step is as follows.
[0070] S101: Place two radars adjacent to each other with the same observation angle. The observation target consists of multiple strong scatter points. Set the observation target to move towards the radars. Specifically, set the starting frequency of the first radar to f c , the frequency hopping interval to Δf, and the number of frequency steps to N; the starting frequency of the second radar is f c +N c Δf, the frequency hopping interval is Δf, and the number of frequency steps is N.
[0071] Specifically, construct the following simulation scenario: Two frequency stepped radars are placed adjacent to each other with the same observation angle; the observation target consists of multiple strong scatter points and moves towards the radars at a certain speed; the distance between the observation target and the radars is much greater than the distance between the two radars. Let the starting frequency of radar 1 be f c , the frequency hopping interval be Δf, and the number of frequency steps be N. The starting frequency of radar 2 is f c +N c Δf, the frequency hopping interval is Δf, and the number of frequency steps is N.
[0072] S102: The two radars monitor the observation target, and perform velocity compensation on the echoes received by the two radars to obtain the first radar echo and the second radar echo. According to the ideal point scattering model of the radar target, after performing velocity compensation on the two radar echoes received, the first radar echo and the second radar echo are respectively:
[0073]
[0074] where p represents the number of strong scatter points included in the observation target, A i represents the scattering intensity of the i-th scatter point, R1 represents the distance between the observation target and the first radar, R2 represents the distance between the observation target and the second radar, f c represents the starting frequency of the first radar, Δf represents the frequency hopping interval, N represents the number of frequency steps, f c +N c Δf represents the starting frequency of the second radar, c represents the speed of light, φ1 represents the initial phase of the first radar echo, φ2 represents the initial phase of the second radar echo, j represents the imaginary unit, and k represents the number of frequency hopping intervals.
[0075] S103: Based on the first radar echo, deform the expressions of the first radar echo and the second radar echo to obtain the actual phase deviation. Among them, each first radar echo, the corresponding second radar echo, and the corresponding actual phase deviation are used as the standard data group in the training data set.
[0076] Taking Radar 1 as the reference, the deformed first radar echo is:
[0077]
[0078] The deformed second radar echo is:
[0079]
[0080] The actual phase deviation is:
[0081] e j(kα+β)
[0082]
[0083] Among them, p represents the number of strong scattering points included in the observed target, A i represents the scattering intensity of the i-th scattering point, R1 represents the distance between the observed target and the first radar, R2 represents the distance between the observed target and the second radar, f c represents the starting frequency of the first radar, Δf represents the frequency hopping interval, N represents the number of frequency steps, f c +N c Δf represents the starting frequency of the second radar, c represents the speed of light, φ1 represents the initial phase of the first radar echo, φ2 represents the initial phase of the second radar echo, j represents the imaginary unit, k represents the number of frequency hopping intervals, α represents the linear phase, and β represents the fixed phase. e j(kα+β) is the factor causing the incoherence of the radar echo data. To achieve coherent registration between different sub-bands, it is necessary to accurately estimate the linear phase and the fixed phase, and then perform phase compensation on the echo.
[0084] S104: Randomly select the standard data group, add noise to the standard data group to generate the extended data group. Among them, the standard data group and the extended data group constitute the training data set.
[0085] Specifically, construct the data set. First, generate observed targets whose number, position, and scattering coefficient of scattering points all follow Gaussian random distributions. Then, for each observed target, generate radar echoes (S1(k), S2(k)) under different (α, β) according to the formula in S103, and the corresponding label is e j(kα+β) . Considering the low signal-to-noise ratio of the actual radar echo, add noise to the radar echo (S1(k), S2(k)) to simulate the actual situation.
[0086] S20: Input the real part and imaginary part of the first radar echo and the real part and imaginary part of the second radar echo into the initial deep learning network model simultaneously to obtain the predicted phase deviation.
[0087] S30: Determine the loss value according to the predicted phase deviation and the actual phase deviation.
[0088] S40: Adjust the parameters of the initial deep learning network model according to the loss value.
[0089] S50: Use multiple groups of data in the training dataset to perform cyclic training and adjustment on the initial deep learning network model until the loss value converges within the expected range or the number of cyclic training reaches the preset number, then stop training to obtain the trained deep learning network model.
[0090] Specifically, after setting training parameters such as the total number of samples, number of training epochs, and initial learning rate in the training dataset, input the echo data of Radar 1 and Radar 2 in the training dataset into the 1D-Unet network. Compare the output (predicted phase deviation) of the network with the label (actual phase deviation), calculate the loss value, and continuously adjust the network parameters through the loss value to train the network until convergence to obtain the trained network model. The loss function is as follows.
[0091]
[0092] where y i represents the i-th predicted phase deviation, y ′ i represents the i-th actual phase deviation, and N represents the total number of used training data.
[0093] Figures 2 - 5 is a schematic diagram of the results before and after coherent registration of multi-band echoes based on deep learning. The observed targets in the figure all have three scattering points. In each diagram, one line represents the echo of Radar 1 and the other line represents the echo of Radar 2; comparing Figure 2 and Figure 3 , or comparing Figure 4 and Figure 5 , it can be seen from the comparison of the display diagrams before and after phase compensation that after compensating the echo of Radar 2, the two lines basically coincide and the registration effect is good.
[0094] As Figure 6 shown, a multi-band echo coherent registration system based on deep learning provided by an embodiment of the present application includes:
[0095] A data acquisition module 100, configured to acquire radar echo data after velocity compensation, where the radar echo data includes a first radar echo and a second radar echo.
[0096] The deviation prediction module 200 is configured to use the real part and the imaginary part of the first radar echo and the real part and the imaginary part of the second radar echo as input data and send them into the trained deep learning network model to obtain a predicted phase deviation, where the phase deviation is the phase deviation of the second radar echo relative to the first radar echo.
[0097] The echo correction module 300 is configured to perform phase compensation on the second radar echo according to the predicted phase deviation to obtain a corrected second radar echo, where there is coherence between the corrected second radar echo and the first radar echo.
[0098] In an optional embodiment of the present application, the multi-band echo coherence registration system based on deep learning further includes a model training module 400, and the model training module 400 is configured to:
[0099] In the constructed simulation scenario, an observation target is randomly generated, and the observation target is used as the monitoring object to generate a training data set, where the training data set includes multiple groups of data, and each group of data includes a first radar echo, a second radar echo, and an actual phase deviation, and the actual phase deviation is the phase deviation between the second radar echo and the first radar echo.
[0100] The real part and the imaginary part of the first radar echo and the real part and the imaginary part of the second radar echo are simultaneously input into the initial deep learning network model to obtain a predicted phase deviation.
[0101] According to the predicted phase deviation and the actual phase deviation, a loss value is determined.
[0102] According to the loss value, the parameters of the initial deep learning network model are adjusted.
[0103] The initial deep learning network model is cyclically trained and adjusted using multiple groups of data in the training data set until the loss value converges within the expected range or the number of cyclic training reaches the preset number of times, and then the training is stopped to obtain a trained deep learning network model.
[0104] In this embodiment, the beneficial effects of the multi-band echo coherence registration system based on deep learning are similar to those of the above-mentioned multi-band echo coherence registration method based on deep learning, and will not be elaborated here.
[0105] An electronic device provided in an embodiment of the present application includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the above-mentioned multi-band echo coherence registration method based on deep learning when executing the computer program.
[0106] A computer-readable storage medium provided by an embodiment of the present application, on which a computer program is stored. When the computer program is executed by a processor, the multi-band echo coherent registration method based on deep learning as described above is implemented.
[0107] In this embodiment, the beneficial effects of the electronic device and the computer-readable storage medium are similar to those of the multi-band echo coherent registration method based on deep learning as described above, and will not be elaborated here.
[0108] Now, an electronic device that can be used as a server or a client of the present application will be described. It is an example of a hardware device that can be applied to various aspects of the present application. The electronic device is intended to represent various forms of digital electronic computer devices, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.
[0109] The electronic device includes a computing unit that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The computing unit, the ROM, and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.
[0110] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc. In this application, the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. One can select some or all of the units according to actual needs to achieve the purpose of the solution of the embodiments of this application. In addition, in each embodiment of this application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.
[0111] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.
[0112] The above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A multi-band echo coherent registration method based on deep learning, characterized in that Including: Obtain the radar echo data after speed compensation, where the radar echo data includes a first radar echo and a second radar echo; Use the real part and imaginary part of the first radar echo and the real part and imaginary part of the second radar echo as input data and send them into the trained deep learning network model to obtain a predicted phase deviation, where the phase deviation is the phase deviation of the second radar echo relative to the first radar echo; According to the predicted phase deviation, perform phase compensation on the second radar echo to obtain a second radar corrected echo, where there is coherence between the second radar corrected echo and the first radar echo.
2. The multi-band echo coherent registration method based on deep learning according to claim 1, wherein The training process of the deep learning network model includes: In the constructed simulation scenario, randomly generate observed targets, use the observed targets as the monitoring objects, and generate a training data set, where the training data set includes multiple groups of data, and each group of data includes a first radar echo, a second radar echo, and an actual phase deviation, and the actual phase deviation is the phase deviation between the second radar echo and the first radar echo; Simultaneously input the real part and imaginary part of the first radar echo and the real part and imaginary part of the second radar echo into the initial deep learning network model to obtain a predicted phase deviation; Determine the loss value according to the predicted phase deviation and the actual phase deviation; Adjust the parameters of the initial deep learning network model according to the loss value; Use multiple groups of data in the training data set to perform cyclic training and adjustment on the initial deep learning network model until the loss value converges within the expected range or the number of cyclic training times reaches the preset number of times, then stop training to obtain the trained deep learning network model.
3. The multi-band echo coherent registration method based on deep learning according to claim 2, wherein In the constructed simulation scenario, randomly generating observed targets and using the observed targets as the monitoring objects to generate a training data set includes: Place two radars adjacent to each other with the same observation angle. The observed target consists of multiple strong scattering points. Set the observed target to move towards the radars. Among them, set the starting frequency of the first radar as f c , the frequency hopping interval is Δf, and the number of frequency steps is N; the starting frequency of the second radar is f c +N c Δf, the frequency hopping interval is Δf, and the number of frequency steps is N; Two radars monitor the observed targets, perform speed compensation on the echoes received by the two radars to obtain a first radar echo and a second radar echo; Taking the first radar echo as a reference, deform the expressions of the first radar echo and the second radar echo to obtain the actual phase deviation, where each first radar echo, the corresponding second radar echo, and the corresponding actual phase deviation are used as the standard data groups in the training data set; Randomly select standard data groups, add noise to the standard data groups to generate extended data groups, where the standard data groups and the extended data groups constitute the training data set.
4. The multi-band echo coherent registration method based on deep learning according to claim 3, wherein The deformed first radar echo is: The deformed second radar echo is: The actual phase deviation is: e j(kα+β) Among them, p represents the number of strong scattering points contained in the observed target, A i represents the scattering intensity of the i-th scattering point, R1 represents the distance between the observed target and the first radar, R2 represents the distance between the observed target and the second radar, f c represents the starting frequency of the first radar, Δf represents the frequency hopping interval, N represents the number of frequency steps, f c +N c Δf represents the starting frequency of the second radar, c represents the speed of light, represents the initial phase of the echo of the first radar, represents the initial phase of the echo of the second radar, j represents the imaginary unit, k represents the number of frequency hopping intervals, α represents the linear phase, and β represents the fixed phase.
5. The multi-band echo coherent registration method based on deep learning according to claim 2, wherein The loss value is: where y i represents the i-th predicted phase deviation, and y ′ i represents the i-th actual phase deviation, and N represents the total number of training data used.
6. The multi-band echo coherent registration method based on deep learning according to claim 1, wherein The deep learning network model is a 1D-Unet network model.
7. A multi-band echo coherent registration system based on deep learning, characterized in that, Including: A data acquisition module for obtaining the radar echo data after speed compensation, where the radar echo data includes a first radar echo and a second radar echo; A deviation prediction module for using the real part and imaginary part of the first radar echo and the real part and imaginary part of the second radar echo as input data and sending them into the trained deep learning network model to obtain a predicted phase deviation, where the phase deviation is the phase deviation of the second radar echo relative to the first radar echo; An echo correction module is configured to perform phase compensation on the second radar echo according to the predicted phase deviation to obtain a corrected second radar echo, where there is coherence between the corrected second radar echo and the first radar echo.
8. The multi-band echo coherent registration system based on deep learning according to claim 7, characterized in that, It further includes a model training module, and the model training module is configured to: Randomly generate observed targets in a constructed simulation scenario, use the observed targets as monitoring objects, and generate a training data set, where the training data set includes multiple groups of data, and each group of data includes a first radar echo, a second radar echo, and an actual phase deviation, and the actual phase deviation is the phase deviation between the second radar echo and the first radar echo; Simultaneously input the real part and the imaginary part of the first radar echo and the real part and the imaginary part of the second radar echo into an initial deep learning network model to obtain a predicted phase deviation; Determine a loss value according to the predicted phase deviation and the actual phase deviation; Adjust the parameters of the initial deep learning network model according to the loss value; Perform cyclic training and adjustment on the initial deep learning network model using multiple groups of data in the training data set until the loss value converges within an expected range or the number of cyclic training reaches a preset number, then stop training to obtain a trained deep learning network model.
9. An electronic device, characterized in that, It includes a memory and a processor; The memory is configured to store a computer program; The processor is configured to, when executing the computer program, implement the multi-band echo coherence registration method based on deep learning according to any one of claims 1 to 6.
10. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by the processor, the multi-band echo coherence registration method based on deep learning according to any one of claims 1 to 6 is implemented.