Radar echo time delay prediction method and device, equipment, storage medium and product

Abstract

The invention belongs to the technical field of radars, and discloses a radar echo time delay prediction method and device, equipment, a storage medium and a product. The method comprises the steps of collecting a current echo signal of a ground radar at a current moment, then performing short-time Fourier transform on the current echo signal to obtain a current time-frequency graph, and then inputting the current time-frequency graph into a preset time delay prediction model to obtain a time delay value corresponding to the echo signal. The preset time delay prediction model comprises a preset denoising layer used for outputting a denoised time frequency graph, a preset reversible short-time Fourier transform layer used for outputting a time domain signal, and a preset full-connection regression layer used for outputting a signal time delay. The time-frequency domain can be entered by performing short-time Fourier transform on the echo signal received by the ground radar, and then the current time-frequency graph is input into the preset time delay prediction model, so that the time delay of the echo signal received by the ground radar can be accurately predicted through the preset time delay prediction model under the condition of urban scene shielding and noise superposition.

Description

Technical Field

[0001] This application relates to the field of radar technology, and in particular to a radar echo delay prediction method, apparatus, device, storage medium, and product. Background Technology

[0002] Traditional ground-based radar (GPR) links typically follow this pattern: echo acquisition → echo preprocessing (DC removal, normalization, windowing) → time-domain denoising / filtering → matched filtering / peak detection → range estimation. The core of existing links remains at the time-domain waveform level; they rarely delve into the TFR (Time-Frequency Reception) domain for structured characterization and guidance. This results in inaccurate time delay prediction of received echo signals under conditions of obstruction and noise superposition in urban environments. Summary of the Invention

[0003] The main objective of this application is to provide a radar echo delay prediction method, apparatus, device, storage medium, and product, aiming to solve the technical problem of accurately predicting the delay of echo signals received by ground radar under conditions of obstruction and noise superposition in urban scenarios.

[0004] To achieve the above objectives, this application provides a radar echo delay prediction method, which includes the following steps: Collect the current echo signal from the ground radar at the current moment; Perform a short-time Fourier transform on the current echo signal to obtain the current time-frequency diagram; The current time-frequency graph is input into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency graph, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal delay.

[0005] Optionally, before inputting the current time-frequency graph into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal, the method further includes: Collect historical echo signals from ground-based radar within a historical time period; Perform a short-time Fourier transform on the historical echo signal to obtain a historical time-frequency diagram; Based on the historical time-frequency graph, the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model are trained to obtain the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer. A preset time delay prediction model is constructed based on the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer.

[0006] Optionally, performing a short-time Fourier transform on the historical echo signal to obtain a historical time-frequency diagram includes: Perform a short-time Fourier transform on the historical echo signal to obtain the historical complex time-frequency signal; The historical complex time-frequency data is split into a preset number of real-value channels and combined into a historical time-frequency graph.

[0007] Optionally, the step of training the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model based on the historical time-frequency map to obtain the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer includes: The historical time-frequency graph is input into the initial time delay prediction model to obtain the first loss value corresponding to the initial denoising layer, the second loss value corresponding to the initial invertible short-time Fourier transform layer, and the third loss value corresponding to the initial fully connected regression layer. Calculate the target loss value based on the first loss value, the second loss value, and the third loss value; The initial denoising layer, the initial invertible short-time Fourier transform layer, and the initial fully connected regression layer are trained based on the target loss value to obtain the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer.

[0008] Optionally, the step of inputting the historical time-frequency graph into the initial time delay prediction model to obtain the first loss value corresponding to the initial denoising layer, the second loss value corresponding to the initial invertible short-time Fourier transform layer, and the third loss value corresponding to the initial fully connected regression layer includes: The historical time-frequency map is input into the initial denoising layer of the initial time delay prediction model to obtain the denoised historical time-frequency map. The denoised historical time-frequency graph is input into the initial reversible short-time Fourier transform layer to obtain the historical time-domain signal; The historical time-domain signal is input into the initial fully connected regression layer to obtain the historical signal delay; The ground radar is used to collect tag time-frequency maps, tag time-domain signals, and tag signal delays under ideal conditions. The first loss value corresponding to the initial denoising layer is determined based on the historical time-frequency graph and the tag time-frequency graph. The second loss value corresponding to the initial invertible short-time Fourier transform layer is determined based on the historical time-domain signal and the tag time-domain signal. The third loss value corresponding to the initial fully connected regression layer is determined based on the historical signal delay and the tag signal delay.

[0009] Optionally, the initial denoising layer includes: an encoding layer, a bottleneck layer, and a decoding layer; The step of inputting the historical time-frequency map into the initial denoising layer of the initial time delay prediction model to obtain the denoised historical time-frequency map includes: The historical time-frequency map is input into the coding layer to extract multi-scale features; The multi-scale features are input into the bottleneck layer carrying a spatiotemporal attention mechanism to obtain enhanced features; The enhanced features are input into the decoding layer to obtain the denoised historical time-frequency map.

[0010] Furthermore, to achieve the above objectives, this application also provides a radar echo delay prediction device, the radar echo delay prediction device comprising: The signal acquisition module is used to acquire the current echo signal of the ground radar at the current moment; The signal processing module is used to perform a short-time Fourier transform on the current echo signal to obtain the current time-frequency diagram; The time delay prediction module is used to input the current time-frequency graph into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency graph, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal delay.

[0011] In addition, to achieve the above objectives, this application also proposes a radar echo delay prediction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the radar echo delay prediction method as described above.

[0012] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the radar echo delay prediction method described above.

[0013] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the radar echo delay prediction method described above.

[0014] This application acquires the current echo signal from a ground radar at the current moment, performs a short-time Fourier transform on the current echo signal to obtain a current time-frequency map, and then inputs the current time-frequency map into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency map, a preset invertible short-time Fourier transform layer for outputting the time-domain signal, and a preset fully connected regression layer for outputting the signal delay. By performing a short-time Fourier transform on the echo signal received by the ground radar, this application can enter the time-frequency domain, convert the current echo signal into a current time-frequency map, and then input the current time-frequency map into the preset time delay prediction model. Thus, the preset time delay prediction model can accurately predict the time delay of the echo signal received by the ground radar under the conditions of urban scene occlusion and noise superposition. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the radar echo delay prediction method of this application; Figure 2 This is a flowchart illustrating the second embodiment of the radar echo delay prediction method of this application; Figure 3 This is a schematic diagram of the overall structure of an embodiment of the radar echo delay prediction method of this application; Figure 4 This is a schematic diagram of the overall process of an embodiment of the radar echo delay prediction method of this application; Figure 5 This is a structural block diagram of the first embodiment of the radar echo delay prediction device of this application; Figure 6 This is a schematic diagram of the structure of a radar echo delay prediction device in the hardware operating environment involved in the embodiments of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0020] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] The main solution of this application embodiment is: to acquire the current echo signal of the ground radar at the current moment; to perform a short-time Fourier transform on the current echo signal to obtain the current time-frequency map; to input the current time-frequency map into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency map, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal time delay.

[0022] Traditional ground-based radar (GPR) links typically follow this pattern: echo acquisition → echo preprocessing (DC removal, normalization, windowing) → time-domain denoising / filtering → matched filtering / peak detection → range estimation. The core of existing links remains at the time-domain waveform level; they rarely delve into the TFR (Time-Frequency Reception) domain for structured characterization and guidance. This results in inaccurate time delay prediction of received echo signals under conditions of obstruction and noise superposition in urban environments.

[0023] This application acquires the current echo signal from a ground radar at the current moment, performs a short-time Fourier transform on the current echo signal to obtain a current time-frequency map, and then inputs the current time-frequency map into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency map, a preset invertible short-time Fourier transform layer for outputting the time-domain signal, and a preset fully connected regression layer for outputting the signal delay. By performing a short-time Fourier transform on the echo signal received by the ground radar, this application can enter the time-frequency domain, convert the current echo signal into a current time-frequency map, and then input the current time-frequency map into the preset time delay prediction model. Thus, the preset time delay prediction model can accurately predict the time delay of the echo signal received by the ground radar under the conditions of urban scene occlusion and noise superposition.

[0024] It should be noted that the executing entity of this application can be a computing service device with data processing, network communication and program execution functions, such as a computer.

[0025] Based on this, embodiments of this application provide a radar echo delay prediction method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the radar echo delay prediction method of this application.

[0026] In this embodiment, the radar echo delay prediction method includes the following steps: Step S10: Collect the current echo signal of the ground radar at the current moment.

[0027] Understandably, this application addresses ground-based radar ranging under "urban obstruction + noise" conditions, proposing an end-to-end closed-loop framework (DTFIRNet) of "time-frequency guidance – learnable inverse transform – time-domain reconstruction – time delay regression". The system consists of a radar front-end and a processing unit (software algorithm): the radar front-end performs transmission / reception and analog-to-digital conversion (TX / RX+ADC) in a linear frequency modulation (LFM) mode to obtain the baseband echo sequence; the processing unit may include signal processing based on a preset time delay prediction model.

[0028] In practical implementation, the current echo signal from the ground radar at the current moment can be collected. Specifically, the simulation / measured parameters can be initialized according to the characterization range (sampling rate f). s = 256MHz, single pulse length, effective echo length, carrier frequency / bandwidth, target distance and speed range, etc.), and complete basic preprocessing such as DC removal and amplitude normalization to form the original time domain input, i.e. the current echo signal.

[0029] Step S20: Perform a short-time Fourier transform on the current echo signal to obtain the current time-frequency diagram.

[0030] It should be understood that a short-time Fourier transform (STFT) can be performed on the current echo signal to obtain the current time-frequency plot, which may include four channels.

[0031] Step S30: Input the current time-frequency graph into the preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency graph, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal delay.

[0032] Understandably, the current time-frequency graph can be input into a preset time delay prediction model. The preset time delay prediction model refers to a model used for signal time delay prediction. The preset time delay prediction model may include: a preset denoising layer for outputting the denoised time-frequency graph, which can achieve time-frequency domain denoising; a preset invertible short-time Fourier transform layer for outputting the time-domain signal, which can achieve inverse mapping from the time-frequency domain to the time domain; and a preset fully connected regression layer for outputting the signal time delay, which can achieve signal time delay prediction.

[0033] This embodiment acquires the current echo signal from a ground radar at the current moment, performs a short-time Fourier transform on the current echo signal to obtain a current time-frequency map, and then inputs the current time-frequency map into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency map, a preset invertible short-time Fourier transform layer for outputting the time-domain signal, and a preset fully connected regression layer for outputting the signal delay. This embodiment, by performing a short-time Fourier transform on the echo signal received by the ground radar, can enter the time-frequency domain, convert the current echo signal into a current time-frequency map, and then input the current time-frequency map into the preset time delay prediction model. Thus, the preset time delay prediction model accurately predicts the time delay of the echo signal received by the ground radar under urban scene occlusion and noise superposition conditions.

[0034] refer to Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the radar echo delay prediction method of this application.

[0035] Based on the first embodiment described above, in this embodiment, before step S30, the method further includes: Step S01: Collect historical echo signals from ground radar within a historical time period.

[0036] Understandably, in order to train a preset time delay prediction model, this embodiment can collect historical echo signals X from ground radar within a historical time period. in Historical echo signal Xin It can be used for subsequent model training.

[0037] Step S02: Perform a short-time Fourier transform on the historical echo signal to obtain a historical time-frequency diagram.

[0038] Further, in this embodiment, step S202 includes: performing a short-time Fourier transform on the historical echo signal to obtain historical complex-valued time-frequency data; splitting the historical complex-valued time-frequency data into a preset number of real-valued channels and combining them into a historical time-frequency diagram.

[0039] Understandably, referring to Figure 3 , Figure 3 This is a schematic diagram of the overall structure of an embodiment of the radar echo delay prediction method of this application. This embodiment can, based on urban obstruction scenarios, use historical echo signals... Xin Based on the physical mechanism, it is divided into three segments: the time delay silent segment [0,τ], the obstruction and damage segment (τ,τ2], and the normal reception segment (τ2, T], where τ represents the time delay corresponding to the historical echo signal, such as... Figure 3 The Corrupted Waveform in the text refers to the damaged waveform, which is divided into three segments. The Echo signal represents the historical echo signal.

[0040] It should be understood that a short-time Fourier transform (STFT) can be performed on the historical echo signal to obtain the historical complex-valued time-frequency representation X(t,ω). The window type, window length, and step size are selected according to the experimental configuration so that the final time-frequency amplitude dimension matches the network input requirements (e.g., constructing a grid of H×W = 512×512).

[0041] In the specific implementation, the historical complex-valued time-frequency X(t, ω) is divided into a preset number of real-valued channels according to the polarity of its real / imaginary parts. The preset number can be 4, that is, it is divided into 4 real-valued channels X. 1~4 These are combined into a 4×H×W tensor, explicitly encoding the energy and phase structure, which serves as the historical time-frequency graph (TFR), i.e., as the model input, such as... Figure 3 The Corrupted TFR in the diagram represents a damaged time-frequency plot with 4 channels and H×W = 512×512. It can explicitly encode phase polarity and energy distribution, enhancing the model's ability to identify energy scattering in shading regions.

[0042] Step S03: Train the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model based on the historical time-frequency map to obtain the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer.

[0043] Understandably, the initial delay prediction model refers to the untrained delay prediction model, which may include initial denoising layers (Denoised Modules), an initial invertible short-time Fourier transform layer (Learnable iSTFT), and an initial fully connected regression layer. The initial denoising layers, initial invertible short-time Fourier transform layers, and initial fully connected regression layers in the initial delay prediction model can be trained based on historical time-frequency maps to obtain the trained target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer.

[0044] Furthermore, in order to effectively train the initial denoising layer, the initial invertible short-time Fourier transform layer, and the initial fully connected regression layer, in this embodiment, step S203 includes: inputting the historical time-frequency map into the initial time delay prediction model to obtain a first loss value corresponding to the initial denoising layer, a second loss value corresponding to the initial invertible short-time Fourier transform layer, and a third loss value corresponding to the initial fully connected regression layer; calculating a target loss value based on the first loss value, the second loss value, and the third loss value; and training the initial denoising layer, the initial invertible short-time Fourier transform layer, and the initial fully connected regression layer based on the target loss value to obtain a target denoising layer, a target invertible short-time Fourier transform layer, and a target fully connected regression layer.

[0045] Understandably, inputting the historical time-frequency graph into the initial time delay prediction model yields the first loss value L corresponding to the initial denoising layer. TFR The second loss value L corresponding to the initial invertible short-time Fourier transform layer wave And the third loss value L corresponding to the initial fully connected regression layer τ Then, the first loss value, the second loss value, and the third loss value are weighted and summed to obtain the target loss value. The calculation formula is: L total =λ1L TFR +λ2L wave +λ3L τ For example, (λ1,λ2,λ3) = (0.5, 2, 2).

[0046] In the specific implementation, the training / test set is divided in an 8:2 ratio, and an L is constructed synchronously for each sample. TFR L wave L τ Batch size set to 8, weight decay 3×10 -5 Then, based on the target loss value, the initial denoising layer, the initial invertible short-time Fourier transform layer, and the initial fully connected regression layer are trained until the target loss value converges, and the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer are obtained at this point.

[0047] Further, in this embodiment, the step of inputting the historical time-frequency map into the initial delay prediction model to obtain the first loss value corresponding to the initial denoising layer, the second loss value corresponding to the initial invertible short-time Fourier transform layer, and the third loss value corresponding to the initial fully connected regression layer includes: inputting the historical time-frequency map into the initial denoising layer in the initial delay prediction model to obtain the denoised historical time-frequency map; inputting the denoised historical time-frequency map into the initial invertible short-time Fourier transform layer to obtain the historical time-domain signal; inputting the historical time-domain signal into the initial fully connected regression layer to obtain the historical signal delay; collecting the tag time-frequency map, tag time-domain signal, and tag signal delay of the ground radar under ideal conditions; determining the first loss value corresponding to the initial denoising layer based on the historical time-frequency map and the tag time-frequency map; determining the second loss value corresponding to the initial invertible short-time Fourier transform layer based on the historical time-domain signal and the tag time-domain signal; and determining the third loss value corresponding to the initial fully connected regression layer based on the historical signal delay and the tag signal delay.

[0048] It should be understood that inputting the historical time-frequency graph into the initial denoising layer of the initial time-delay prediction model yields the denoised historical time-frequency graph, such as... Figure 3 The Corrupted TFR is input into Denoised Modules, resulting in O TFR .

[0049] Understandably, the denoised historical time-frequency graph is input into the initial invertible short-time Fourier transform layer to obtain the historical time-domain signal, such as... Figure 3 O in TFR Input into Learnable iSTFT, get O wave A one-dimensional learnable convolution kernel Wc',c(ω) along the frequency axis can be used to achieve the inverse mapping from the TFR to the time domain, resulting in the reconstructed waveform O. wave This layer replaces the exponent kernel of the traditional analytical iSTFT, adaptively emphasizing information bands, suppressing noise bands, and compensating for phase distortion; when W converges to a Fourier basis, it degenerates into an analytical iSTFT. For O... wave With L wave Calculate region-weighted waveform loss: assign higher weights (e.g., weight 4, threshold 10) to temporal samples with near-zero amplitude (silent / occluded boundaries). -3 (In conjunction with SmoothL1 constraints, it significantly improves jump edge alignment and detail fidelity.)

[0050] In this embodiment, the historical time-domain signal is input into the initial fully connected regression layer to obtain the historical signal delay, such as... Figure 3 O in wave Input into N_delay, get O delay O wave Or its higher-order features are input into a fully connected regression layer, and the output is a discrete delay. (or consecutive) );according to = τ / f s With R = c / 2 yields the final distance estimate This branch road connects with L. delay Calculate the regression loss.

[0051] In practical implementation, the system collects the tag time-frequency map (Label TFR), tag time-domain signal (Label Waveform), and tag signal delay from ground radar under ideal conditions. The historical time-frequency map (O) can then be used... TFR The label time-frequency plot (Label TFR) is input into the MSE+SmoothL1 loss function to calculate the first loss value L. TFR , will historical time domain signal O wave The label time-domain signal (Label Waveform) is input into the region-weighted +Smooth L1 loss function to calculate the second loss value L. wave Delaying historical signals by O delayThe label delay is input into the MSE loss function to calculate the third loss value L. τ .

[0052] Further, in this embodiment, the step of inputting the historical time-frequency map into the initial denoising layer in the initial delay prediction model to obtain the denoised historical time-frequency map includes: inputting the historical time-frequency map into the coding layer to extract multi-scale features; inputting the multi-scale features into the bottleneck layer carrying a spatiotemporal attention mechanism to obtain enhanced features; and inputting the enhanced features into the decoding layer to obtain the denoised historical time-frequency map.

[0053] Understandably, the initial denoising layer may include an encoding layer, a bottleneck layer, and a decoding layer. Historical time-frequency maps are input into the encoding layer, where multi-level downsampling and residual convolution extract multi-scale features. These multi-scale features are then input into the bottleneck layer, which carries a spatiotemporal attention mechanism (STA), to capture long-range time-frequency dependencies and occlusion artifacts such as stripes / energy fragmentation, thereby obtaining enhanced features. This significantly suppresses stripes / energy fragmentation caused by occlusion and improves ridge continuity and concentration.

[0054] It should be understood that the enhanced features are input into the decoding layer, and the details are recovered step by step through upsampling and skip connections to obtain the denoised historical time-frequency map. TFR .

[0055] Step S04: Construct a preset time delay prediction model based on the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer.

[0056] Understandably, a preset time delay prediction model can be constructed based on the trained target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer to predict the echo signal delay.

[0057] It should be understood that the system primarily focuses on the joint optimization of three levels of objectives: time-frequency domain, time domain, and time delay. The input is a preprocessed echo waveform, which is first processed by STFT to obtain a complex-valued TFR. The complex matrix is ​​then split into four real-valued channels (Re±, Im±) as network input. The backbone is an enhanced UNet denoising module (including residual and spatiotemporal attention), and the output is a clean TFR (O...). TFR ); then, a learnable inverse transform from TFR to time-domain waveform is performed through a learnable iSTFT subnet to obtain the recovered waveform (O ); wave ); End-to-end regression latency / distance (O) via fully connected circuitry τ ), three outputs and corresponding supervision (L) TFR ,L wave ,L τ (Use composite loss for joint training)

[0058] During training, data and annotation: A dataset (2000 records in total, training / test ratio 8:2) was constructed based on randomization of physical parameters, and a clean waveform L was generated for each sample. wave Clean TFRL TFR With delay label L τ The input is decomposed by STFT and four quadrants to obtain a 4×512×512 TFR.

[0059] Training configuration: Batch size 8, weight decay 3×10 -5 Region weight 4, loss weight [0.5, 2.0, 2.0]; experimental environment: 10×Xeon Platinum 8352V + RTX 4090, PyTorch 2.4.0 / Python 3.12, MATLAB2024b.

[0060] Effects and Convergence: Visualization shows that the ridges of OTFR gradually become more coherent and the energy more concentrated during training; O wave The jump edge / high frequency band gradually recovers, and by the 10th round, it is highly consistent with the true value.

[0061] Ablation and Complexity: Removing TFR / waveform loss significantly degrades reconstruction and ranging; when using traditional iSTFT instead of learnable iSTFT, waveform distortion is obvious and distance error increases. Representative methods FLOPs: CtFreeNet 60.199 GF, DGDNet 187.477 GF, DTFTCNet 120.649 GF, this application 495.611 GF (high FLOPs are due to learnable iSTFT and deeper convolution, but are most robust under low SNR / occlusion).

[0062] In the specific implementation, refer to Figure 4 , Figure 4 This is a schematic diagram of the overall process of an embodiment of the radar echo delay prediction method of this application, as shown below. Figure 4 As shown, S301: Acquisition and input normalization, obtaining the ground radar LFM echo sequence (TX / RX+ADC), and initializing the simulation / measured parameters according to the characterization range (sampling rate f). s = 256MHz, single pulse length, effective echo length, carrier frequency / bandwidth, target range and velocity range, etc.), and perform basic preprocessing such as DC removal and amplitude normalization to form the original time-domain input X. in S302: Occlusion Mechanism Modeling and Label Definition. Based on urban occlusion scenarios, the echo is divided into three segments according to physical mechanisms: a time-delay silent segment [0,τ], an occlusion-damaged segment (τ, τ2], and a normal reception segment (τ2, T). This generates three types of labels required for training / evaluation: Clean TFR L. TFRClean waveform L wave Delay / Distance Label L delay (τ→ R → cτ / 2), and fix its one-to-one correspondence with the input. S303: Time-frequency representation construction (STFT), for X in STFT is performed to obtain the complex-valued time-frequency representation X(t, ω). The window type, window length, and step size are selected according to the experimental configuration to ensure that the final time-frequency amplitude dimension matches the network input requirements (e.g., constructing a grid of H×W = 512×512). S304: Quadrant channelization (TFR preprocessing) splits the complex-valued X(t, ω) into four real-valued channels X based on the positive / negative polarity of the real / imaginary parts. 1~4 These are combined into a 4×H×W tensor, explicitly encoding the energy and phase structure, which serves as the network input. S305: TFR denoising and enhancement (encoder stage): The four-channel TFR is input, and multi-scale features are extracted through multi-layer downsampling and residual convolution of the enhanced UNet; a spatiotemporal attention STA is embedded in the bottleneck layer to capture long-range time-frequency dependencies and occlusion artifacts such as stripes / energy fragmentation. S306: TFR denoising and enhancement (decoder stage): Details are recovered step-by-step through upsampling and skip connections, outputting a denoised / enhanced time-frequency map O. TFR This output is used for (i) and L TFR Calculate the TFR loss; (ii) send it to the learnable inverse transform layer. S307: Learnable iSTFT (deep inverse transform), using a one-dimensional learnable convolution kernel Wc',c (ω) along the frequency axis to achieve the inverse mapping of TFR to the time domain, and obtain the reconstructed waveform O. wave This layer replaces the exponential kernel of the traditional analytical iSTFT, adaptively emphasizing information bands, suppressing noise bands, and compensating for phase distortion; when W converges to a Fourier basis, it degenerates into an analytical iSTFT. S308: Waveform consistency constraint (region weighting), calculating region-weighted waveform loss for Owave and Lwave: assigning higher weights (e.g., weight 4, threshold 10) to time-domain samples close to zero amplitude (silent / occluded boundaries). -3 Combined with SmoothL1 constraints, it significantly improves takeoff edge alignment and detail fidelity. S309: Delay regression (rangefinder), which will reduce O wave Alternatively, it can take higher-order features as input to a fully connected regression head and output discrete delay. (or consecutive) );according to = τ / f s With R = c / 2 yields the final distance estimate This branch road connects with L. delay Calculate the regression loss. S310: Multi-task joint optimization (Hybrid Loss), with L...total =λ1L TFR +λ2L wave +λ3L τ End-to-end training is performed, and in this example, (λ1, λ2, λ3) = (0.5, 2, 2); L TFR For MSE+SmoothL1, L wave For regional weighted +SmoothL1, L τ For MSE. S311: Data organization and batch processing, dividing the training / test set into an 8:2 ratio (e.g., 2000 samples), and simultaneously constructing L for each sample. TFR , L wave , L τ Batch size is set to 8, weight decay is 3×10. -5 S312: Training environment and convergence monitoring, trained on a PyTorch 2.4.0 / Python 3.12 environment and an RTX 4090 device; monitoring the convergence curves and visualizing the evolution of the three outputs, observing O TFR Ridge line coherence and O wave The jump edge is gradually aligned (e.g., in round 101, it is highly consistent with the truth value). S313: Inference flow, the deployment phase only executes the forward pass: X in →STFT→O TFR →O wave →O delay → No external iSTFT or independent detector is required, reducing mismatches and error accumulation. S314: Complexity and deployment pruning: Record FLOPs and memory usage; if computational constraints exist, the computational load can be reduced by compressing the channel width / number of layers or by mixing / distilling the learnable iSTFT and the analytical iSTFT, achieving an engineering trade-off between accuracy and real-time performance. S315: Evaluation and performance analysis: A unified evaluation of TFR quality (ζ / SDR / SSIM / MSE / Rényi entropy), waveform fidelity (MAE / PCC / cosine similarity), and distance accuracy (MAE / scatter point to angle) is performed, and the loss weights and training epochs are adjusted based on scatter point concentration and MAE statistics. Three types of results are output: denoised TFR O TFR , Reconstructing waveform O wave With the final .

[0063] This embodiment acquires historical echo signals from ground radar over a historical time period, performs a short-time Fourier transform on the historical echo signals to obtain a historical time-frequency map, and then trains the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model based on the historical time-frequency map to obtain the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer. Finally, a preset time delay prediction model is constructed based on the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer. This embodiment effectively trains the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model based on the historical time-frequency map, constructing an accurate and effective preset time delay prediction model.

[0064] Reference Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the radar echo delay prediction device of this application.

[0065] like Figure 5 As shown, the radar echo delay prediction device proposed in this application includes: Signal acquisition module 10 is used to acquire the current echo signal of the ground radar at the current moment; Signal processing module 20 is used to perform short-time Fourier transform on the current echo signal to obtain the current time-frequency diagram; The time delay prediction module 30 is used to input the current time-frequency graph into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency graph, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal delay.

[0066] This embodiment acquires the current echo signal from a ground radar at the current moment, performs a short-time Fourier transform on the current echo signal to obtain a current time-frequency map, and then inputs the current time-frequency map into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency map, a preset invertible short-time Fourier transform layer for outputting the time-domain signal, and a preset fully connected regression layer for outputting the signal delay. This embodiment, by performing a short-time Fourier transform on the echo signal received by the ground radar, can enter the time-frequency domain, convert the current echo signal into a current time-frequency map, and then input the current time-frequency map into the preset time delay prediction model. Thus, the preset time delay prediction model accurately predicts the time delay of the echo signal received by the ground radar under urban scene occlusion and noise superposition conditions.

[0067] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In practical applications, those skilled in the art can select some or all of it to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0068] In addition, for technical details not described in detail in this embodiment, please refer to the radar echo delay prediction method provided in any embodiment of this application, which will not be repeated here.

[0069] Based on the first embodiment of the radar echo delay prediction device described in this application, a second embodiment of the radar echo delay prediction device of this application is proposed.

[0070] In this embodiment, the radar echo delay prediction device further includes a model training module, used to collect historical echo signals of ground radar within a historical time period; perform short-time Fourier transform on the historical echo signals to obtain a historical time-frequency map; train the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial delay prediction model based on the historical time-frequency map to obtain a target denoising layer, a target invertible short-time Fourier transform layer, and a target fully connected regression layer; and construct a preset delay prediction model based on the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer.

[0071] Furthermore, the model training module is also used to perform a short-time Fourier transform on the historical echo signal to obtain historical complex-valued time-frequency data; the historical complex-valued time-frequency data is split into a preset number of real-valued channels and combined into a historical time-frequency graph.

[0072] Furthermore, the model training module is also used to input the historical time-frequency map into the initial time delay prediction model to obtain a first loss value corresponding to the initial denoising layer, a second loss value corresponding to the initial invertible short-time Fourier transform layer, and a third loss value corresponding to the initial fully connected regression layer; calculate a target loss value based on the first loss value, the second loss value, and the third loss value; and train the initial denoising layer, the initial invertible short-time Fourier transform layer, and the initial fully connected regression layer based on the target loss value to obtain a target denoising layer, a target invertible short-time Fourier transform layer, and a target fully connected regression layer.

[0073] Furthermore, the model training module is also used to input the historical time-frequency map into the initial denoising layer in the initial delay prediction model to obtain the denoised historical time-frequency map; input the denoised historical time-frequency map into the initial invertible short-time Fourier transform layer to obtain the historical time-domain signal; input the historical time-domain signal into the initial fully connected regression layer to obtain the historical signal delay; collect the tag time-frequency map, tag time-domain signal, and tag signal delay of the ground radar under ideal conditions; determine the first loss value corresponding to the initial denoising layer based on the historical time-frequency map and the tag time-frequency map; determine the second loss value corresponding to the initial invertible short-time Fourier transform layer based on the historical time-domain signal and the tag time-domain signal; and determine the third loss value corresponding to the initial fully connected regression layer based on the historical signal delay and the tag signal delay.

[0074] Furthermore, the initial denoising layer includes an encoding layer, a bottleneck layer, and a decoding layer; the model training module is also used to input the historical time-frequency map into the encoding layer to extract multi-scale features; input the multi-scale features into the bottleneck layer carrying a spatiotemporal attention mechanism to obtain enhanced features; and input the enhanced features into the decoding layer to obtain the denoised historical time-frequency map.

[0075] Other embodiments or specific implementations of the radar echo delay prediction device of this application can be found in the above-described method embodiments, and will not be repeated here.

[0076] This application provides a radar echo delay prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the radar echo delay prediction method in the first embodiment described above.

[0077] The following is for reference. Figure 6 This document illustrates a schematic diagram of a radar echo delay prediction device suitable for implementing embodiments of this application. The radar echo delay prediction device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6The radar echo delay prediction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0078] like Figure 6 As shown, the radar echo delay prediction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the radar echo delay prediction device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the radar echo delay prediction device to communicate wirelessly or wiredly with other devices to exchange data. Although radar echo delay prediction devices with various systems are shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0079] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0080] The radar echo delay prediction device provided in this application, employing the radar echo delay prediction method described in the above embodiments, can solve the technical problem of accurately predicting the delay of echo signals received by ground radar under conditions of obstruction and noise superposition in urban scenes. Compared with the prior art, the beneficial effects of the radar echo delay prediction device provided in this application are the same as those of the radar echo delay prediction method provided in the above embodiments, and other technical features of this radar echo delay prediction device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0081] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0082] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0083] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the radar echo delay prediction method in the above embodiments.

[0084] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, 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 thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0085] The aforementioned computer-readable storage medium may be included in the radar echo delay prediction device; or it may exist independently and not be assembled into the radar echo delay prediction device.

[0086] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the radar echo delay prediction device, the radar echo delay prediction device: acquires the current echo signal of the ground radar at the current moment; performs a short-time Fourier transform on the current echo signal to obtain a current time-frequency map; inputs the current time-frequency map into a preset delay prediction model to obtain the delay value corresponding to the echo signal. The preset delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency map, a preset invertible short-time Fourier transform layer for outputting the time-domain signal, and a preset fully connected regression layer for outputting the signal delay.

[0087] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Python, Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0088] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0089] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0090] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described radar echo delay prediction method. This solves the technical problem of accurately predicting the delay of echo signals received by ground radar under conditions of occlusion and noise superposition in urban scenes. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the radar echo delay prediction method provided in the above embodiments, and will not be repeated here.

[0091] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the radar echo delay prediction method described above.

[0092] The computer program product provided in this application can solve the technical problem of accurately predicting the time delay of echo signals received by ground radar under conditions of occlusion and noise superposition in urban scenes. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the radar echo time delay prediction method provided in the above embodiments, and will not be repeated here.

[0093] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A radar echo delay prediction method, characterized in that, The radar echo delay prediction method includes the following steps: Collect the current echo signal from the ground radar at the current moment; Perform a short-time Fourier transform on the current echo signal to obtain the current time-frequency diagram; The current time-frequency graph is input into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency graph, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal delay.

2. The radar echo delay prediction method as described in claim 1, characterized in that, Before inputting the current time-frequency graph into the preset time delay prediction model to obtain the time delay value corresponding to the echo signal, the method further includes: Collect historical echo signals from ground-based radar within a historical time period; Perform a short-time Fourier transform on the historical echo signal to obtain a historical time-frequency diagram; Based on the historical time-frequency graph, the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model are trained to obtain the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer. A preset time delay prediction model is constructed based on the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer.

3. The radar echo delay prediction method as described in claim 2, characterized in that, The step of performing a short-time Fourier transform on the historical echo signal to obtain a historical time-frequency diagram includes: Perform a short-time Fourier transform on the historical echo signal to obtain the historical complex time-frequency signal; The historical complex time-frequency data is split into a preset number of real-value channels and combined into a historical time-frequency graph.

4. The radar echo delay prediction method as described in claim 2, characterized in that, The step of training the initial denoising layer, initial invertible short-time Fourier transform layer, and initial fully connected regression layer in the initial time delay prediction model based on the historical time-frequency map to obtain the target denoising layer, target invertible short-time Fourier transform layer, and target fully connected regression layer includes: The historical time-frequency graph is input into the initial time delay prediction model to obtain the first loss value corresponding to the initial denoising layer, the second loss value corresponding to the initial invertible short-time Fourier transform layer, and the third loss value corresponding to the initial fully connected regression layer. Calculate the target loss value based on the first loss value, the second loss value, and the third loss value; The initial denoising layer, the initial invertible short-time Fourier transform layer, and the initial fully connected regression layer are trained based on the target loss value to obtain the target denoising layer, the target invertible short-time Fourier transform layer, and the target fully connected regression layer.

5. The radar echo delay prediction method as described in claim 4, characterized in that, The step of inputting the historical time-frequency graph into the initial time delay prediction model to obtain the first loss value corresponding to the initial denoising layer, the second loss value corresponding to the initial invertible short-time Fourier transform layer, and the third loss value corresponding to the initial fully connected regression layer includes: The historical time-frequency map is input into the initial denoising layer of the initial time delay prediction model to obtain the denoised historical time-frequency map. The denoised historical time-frequency graph is input into the initial reversible short-time Fourier transform layer to obtain the historical time-domain signal; The historical time-domain signal is input into the initial fully connected regression layer to obtain the historical signal delay; The ground radar is used to collect tag time-frequency maps, tag time-domain signals, and tag signal delays under ideal conditions. The first loss value corresponding to the initial denoising layer is determined based on the historical time-frequency graph and the tag time-frequency graph. The second loss value corresponding to the initial invertible short-time Fourier transform layer is determined based on the historical time-domain signal and the tag time-domain signal. The third loss value corresponding to the initial fully connected regression layer is determined based on the historical signal delay and the tag signal delay.

6. The radar echo delay prediction method as described in claim 5, characterized in that, The initial denoising layer includes: an encoding layer, a bottleneck layer, and a decoding layer; The step of inputting the historical time-frequency map into the initial denoising layer of the initial time delay prediction model to obtain the denoised historical time-frequency map includes: The historical time-frequency map is input into the coding layer to extract multi-scale features; The multi-scale features are input into the bottleneck layer carrying a spatiotemporal attention mechanism to obtain enhanced features; The enhanced features are input into the decoding layer to obtain the denoised historical time-frequency map.

7. A radar echo delay prediction device, characterized in that, The radar echo delay prediction device includes: The signal acquisition module is used to acquire the current echo signal of the ground radar at the current moment; The signal processing module is used to perform a short-time Fourier transform on the current echo signal to obtain the current time-frequency diagram; The time delay prediction module is used to input the current time-frequency graph into a preset time delay prediction model to obtain the time delay value corresponding to the echo signal. The preset time delay prediction model includes: a preset denoising layer for outputting the denoised time-frequency graph, a preset invertible short-time Fourier transform layer for outputting the time domain signal, and a preset fully connected regression layer for outputting the signal delay.

8. A radar echo delay prediction device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the radar echo delay prediction method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the radar echo delay prediction method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the radar echo delay prediction method as described in any one of claims 1 to 6.

Images