Millimeter wave radar range domain target detection method based on complex binary transform

Abstract

The application discloses a millimeter wave radar distance domain target detection method based on complex binary transformation, and realizes target distance detection by constructing a double-path symmetric complex binary network, and the implementation scheme comprises the following steps: 1. constructing samples containing different target distances and noise intensities as a training data set; 2. constructing a complex binary transformation network (CVBT) model; 3. constructing a loss function by adopting a GrandNorm strategy, solving the gradient disappearance problem in the back propagation of the binary neural network by using a straight-through estimator, and training the CVBT network parameters by using the training sample data set to obtain the CVBT network parameters; 4. performing complex binary transformation on radar echo by using the trained CVBT network, obtaining a distance spectrum, and performing constant false alarm rate detection on the distance spectrum to obtain the specific position of the target. The method realizes the purpose of replacing FFT for frequency domain analysis by adopting the binary neural network parameter training mode for the first time.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave radar target detection, and specifically to a millimeter-wave radar range domain target detection method based on complex binary transformation. Background Technology

[0002] With the rapid development of the automotive industry, medical imaging systems, intelligent transportation, and security screening, increasingly higher demands are being placed on efficient and reliable target detection technologies. Among these, millimeter-wave radar, due to its high operating frequency, can effectively penetrate most non-metallic materials and has significant advantages such as no radiation hazards and safety for human health, thus gaining widespread application and attention.

[0003] Among various millimeter-wave radars, frequency-modulated continuous wave (FMCW) based millimeter-wave radar systems are widely used due to their high measurement accuracy and relatively low cost. These radars typically perform Discrete Fourier Transform (DFT) spectral analysis on the received echo signals to accurately extract key information such as target range, velocity, and angle, thereby achieving efficient target detection and localization. However, although Fast Fourier Transform (FFT) has been widely adopted to improve the computational efficiency of frequency analysis, its operation still involves a large number of complex multiplications. This high computational load significantly increases the hardware requirements and energy consumption of the radar system, becoming a major bottleneck limiting further miniaturization and low-power consumption of millimeter-wave radar systems.

[0004] Meanwhile, with the rapid development of artificial intelligence and deep learning technologies, deep neural networks and related technologies have demonstrated enormous potential in the field of remote sensing and have been successfully applied to various tasks such as image classification, target recognition, and detection. Among them, binary neural networks (BNNs), as an emerging network model, allow computation to be performed using bitwise operations because their network weights and activation functions are only binary, significantly reducing computational complexity and hardware resource requirements. This makes them particularly suitable for applications that are extremely sensitive to computational load and power consumption.

[0005] Therefore, how to learn from and fully utilize the advantages of binary neural networks in terms of low computational complexity and low hardware resource consumption, and design a signal frequency domain analysis method with the same function as Fourier transform but with lower computational complexity, so as to achieve efficient and fast target information extraction, has become an important technical problem that urgently needs to be solved in the field of millimeter-wave radar signal processing. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a range domain target detection method for millimeter-wave radar based on complex binary transformation. This method fully utilizes the characteristics and advantages of binary weights in binary neural networks (BNNs) to enable computation in an addition manner that is more suitable for hardware implementation. This fundamentally transforms the complex complex multiplication operations in traditional frequency domain analysis into addition operations that are easier to implement in hardware and more efficient, thereby significantly reducing the time complexity and hardware overhead of frequency domain transformation.

[0007] Specifically, the technical problem to be solved by the present invention is to address the shortcomings of the prior art. In the first aspect, the present invention provides a millimeter-wave radar range domain target detection method based on complex binary transformation, including: S1, constructing a training sample dataset containing different target distances and noise intensities.

[0008] S2. Construct the parameter model of the complex binary transform (CVBT) network.

[0009] S3. The loss function is constructed using the GrandNorm strategy. The gradient vanishing problem in the backpropagation of the binary neural network is solved by the pass-through estimator. The CVBT network parameter model is trained using the training sample dataset of S1 to obtain the trained CVBT network parameter model.

[0010] S4. The radar echo is processed by complex binary transformation using the trained CVBT network parameter model to obtain the range spectrum. The range spectrum is then used to perform constant false alarm rate (CFAR) detection to obtain the specific location of the target.

[0011] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transformation according to the present invention, the process of constructing the training sample dataset in step S1 is as follows: S11, select radar echo signals under noise-free conditions at different target distances, and then add noise with different signal-to-noise ratios to them to obtain noisy radar echo signals.

[0012] S12. Perform Fast Fourier Transform (FFT) on the noise-free and noisy radar echo signals respectively to obtain the corresponding noise-free range spectrum and noisy range spectrum.

[0013] S13. Using noiseless radar echo signals, noiseless range spectra, noisy radar echo signals, and noisy range spectra together to form a set of training samples, repeat the above process to construct a training sample dataset of 10,000 sets containing different target distances and noise intensities.

[0014] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transformation according to the present invention, the CVBT network parameter model includes a dual-path symmetric complex binary network structure, including a positive CVBT and an inverse CVBT. The radar echo is passed through the positive CVBT to obtain the range spectrum, and the obtained range spectrum is restored through the inverse CVBT to obtain the inverse echo. The radar echo is frequency-domain transformed by the positive CVBT to complete the frequency analysis. The inverse CVBT is used to make the transformation reversible.

[0015] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transform according to the present invention, the radar signal is used to obtain the range spectrum through positive CVBT, and the radar echo signal is represented as follows: s r The operation matrix of positive CVBT is W b_CVBT The actual part and the imaginary part are represented as follows: ;in, .

[0016] S r_CVBT The distance spectrum is obtained by positive CVBT. To represent matrix multiplication, positive CVBT is represented as: The inverse echo is obtained by reconstructing the range spectrum using the inverse CVBT; the operation matrix of the inverse CVBT matrix is... W b_ICVBT The real part and the imaginary part can be represented as: .

[0017] S t_ICVBT The inverse echo obtained by inverse CVBT is represented as: .

[0018] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transformation according to the present invention, the CVBT network parameter training process uses the GrandNorm strategy to construct a loss function to optimize the training process of the transformation matrix. The constructed loss function consists of two parts: echo loss and range spectrum loss, specifically expressed as follows: ;in, S r_GT , s t_GT These are the distance spectrum and inverse echo obtained by FFT and inverse FFT under noise-free conditions, respectively.

[0019] echo loss weight w range and distance spectrum loss weights w echoThe network is adaptively adjusted based on the dynamic changes in the gradients corresponding to the two loss functions during training. The calculation relationship is expressed as follows: .

[0020] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transformation according to the present invention, the binary neural network of the CVBT network model adopts the straight-through estimator (STE) strategy during the backpropagation training process.

[0021] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transform according to the present invention, the STE strategy quantizes continuous real values ​​into discrete forms represented only by binary {0,1}, approximately estimating the derivative of the binary function, thereby ensuring gradient stability and effective propagation during network training. Its binarization process can be expressed as: .

[0022] In one embodiment of the millimeter-wave radar range domain target detection method based on complex binary transformation according to the present invention, the process of training the CVBT network parameter model using the training sample dataset of S1 in step S3 is as follows: the CVBT network parameter model constructed in step S2 is used to train the training sample dataset of step S1, wherein the initial learning rate is set to 0.01, and the learning rate is reduced to half of the original value after every 30 iterations. After completing 200 iterations of training, the CVBT network parameters can be obtained.

[0023] Compared with existing technologies, the positive effects of this invention are as follows: The millimeter-wave radar target range detection method based on complex binary transform proposed in this invention effectively realizes frequency domain analysis of the signal by introducing a binary complex neural network to train the transform parameters. Compared with traditional FFT-based frequency domain analysis methods, this method is the first to use a binary neural network to train parameters for frequency domain transformation instead of FFT for frequency domain analysis, realizing the conversion of complex complex multiplication to simple real number addition during the frequency domain transformation process. This conversion significantly reduces the computational complexity of the frequency domain analysis process, greatly reduces the time and hardware resources required for computation, and has obvious technological innovation and practical application advantages. Attached Figure Description

[0024] Figure 1 This is a flowchart of the millimeter-wave radar target range detection method based on complex binary transformation in this invention.

[0025] Figure 2 The image shows a normalized comparison simulation result of the CVBT method and the traditional FFT method in the embodiments of this invention. Detailed Implementation

[0026] The following describes in detail exemplary embodiments according to the present application with reference to the accompanying drawings. In traditional radar signal processing, frequency domain analysis typically requires a large number of Fast Fourier Transforms, involving numerous complex complex multiplication operations, resulting in excessive computation time and high hardware resource consumption. Therefore, there is an urgent need to propose a frequency domain analysis method to effectively reduce the computational complexity and hardware requirements in radar signal processing. This invention aims to fully utilize the characteristics and advantages of binary weights in Binary Neural Networks (BNNs) to enable computation in an addition manner more suitable for hardware implementation. This fundamentally transforms the complex complex multiplication operations in traditional frequency domain analysis into more efficient addition operations that are easier to implement in hardware, thereby significantly reducing the time complexity and hardware overhead of frequency domain transformation.

[0027] To overcome the problems of high computational complexity and large hardware resource consumption in traditional millimeter-wave radar signal processing, this invention proposes a deep learning-based complex binary transform (CVBT) method. By converting the complex complex multiplication operations in the traditional Fourier transform into addition operations mainly based on binary operations, the computational complexity of frequency domain analysis is significantly reduced, thereby achieving efficient target range detection of millimeter-wave radar.

[0028] This invention is achieved using the following technical solutions: Figure 1 The flowchart shown is a millimeter-wave radar target range detection method based on deep learning complex binary transform (CVBT) according to the present invention. As can be seen from the figure, the millimeter-wave radar target range detection method based on deep learning complex binary transform (CVBT) proposed in this invention includes two parts: a weight training stage and a signal processing stage.

[0029] 1. Weight training phase: such as Figure 1 As shown in Part I, the weight training steps are as follows: Step 1: Constructing the training network: To enable the designed transformation matrix to possess reversible transformation characteristics similar to FFT, i.e., to perform both forward frequency domain analysis and signal reconstruction through inverse transformation, this method constructs a dual-path symmetric complex binary network structure, specifically including forward CVBT and inverse CVBT. The radar echo is processed through forward CVBT to obtain the range spectrum, and the obtained range spectrum is reconstructed through inverse CVBT to obtain the inverse echo.

[0030] The range spectrum of the radar signal can be obtained by passing it through a positive CVBT, and the radar echo signal is represented as follows: s r The operation matrix of positive CVBT is W b_CVBT The real part and the imaginary part can be represented as follows: ;in, .

[0031] S r_CVBT The distance spectrum is obtained by positive CVBT. To represent matrix multiplication, positive CVBT can be represented as: The inverse CVBT can be used to reconstruct the range spectrum and obtain the inverse echo. The operation matrix of the inverse CVBT matrix is... W b_ICVBT The real part and the imaginary part can be represented as follows: 。

[0032] S t_ICVBT The inverse echo obtained by inverse CVBT can be expressed as: .

[0033] By using positive CVBT, the radar echo can be frequency-domain transformed to complete frequency analysis; by using inverse CVBT, the transformation becomes reversible, thus simulating the effect of FFT simulation more accurately.

[0034] Step 2: Establishing the Loss Function and Backpropagation: To effectively obtain a transformation matrix with better performance, this method employs the GrandNorm strategy to construct the loss function during network training, thereby optimizing the transformation matrix training process. The constructed loss function consists of two parts: echo loss and range spectrum loss, specifically expressed as follows: ;in, S r_GT , s t_GT These are the distance spectrum and inverse echo obtained by FFT and inverse FFT under noise-free conditions, respectively. Echo loss weights. w range and distance spectrum loss weights w echo The network is adaptively adjusted based on the dynamic changes in the gradients corresponding to the two loss functions during training. The calculation relationship can be expressed as follows: .

[0035] To address the vanishing gradient problem in backpropagation training of binary neural networks, this method employs a Straight-Through Estimator (STE) strategy, effectively mitigating the gradient propagation difficulties caused by the non-differentiability of the binary function. The STE strategy quantizes continuous real values ​​into a discrete form represented only by binary {0,1}, approximating the estimation of the derivative of the binary function, thus ensuring gradient stability and effective propagation during network training. Its binarization process can be expressed as: .

[0036] Step 3: Construct the training dataset and train CVBT: First, select radar echo signals under noise-free conditions at a specific distance, and then apply different levels of noise interference to obtain noisy radar echo signals. Next, perform Fast Fourier Transform (FFT) on the noise-free and noisy radar echo signals respectively to obtain the corresponding noise-free range spectrum and noisy range spectrum. Use the noise-free radar echo signal, noise-free range spectrum, noisy radar echo signal, and noisy range spectrum together to form a training sample set. Repeat the above process to construct a training sample dataset of 10,000 sets containing different target distances and noise intensities. Train the network model constructed in Steps 1 and 2 on this dataset, with an initial learning rate set to 0.01, and the learning rate is decayed to half its original value after every 30 iterations. After 200 iterations of training, the CVBT network parameters can be obtained.

[0037] 2. Signal processing stage: such as Figure 1 As shown in Part II, the CVBT network parameters are obtained through weight training. The radar echo is processed by complex binary transformation to obtain the range spectrum. Constant False Alarm Rate (CFAR) is used to determine the specific location of the target, thus achieving efficient target range detection.

[0038] Example: The following example illustrates the entire process: The effect of the present invention can be illustrated by the following simulation experiment: 1. Simulation conditions: Platform configuration: CPU: AMD Ryzen 9 7940HX 5.20GHz; RAM: 32G.

[0039] Operating system: Windows 11.

[0040] Simulation software: MATLAB 2024a.

[0041] 2. Simulation parameter settings: carrier frequency 77GHz, pulse repetition period 18.77μs, distance unit number 512, signal bandwidth 150MHz.

[0042] 3. Simulation Content: Based on the simulation parameters set above, the target distance was set to 100m, and Gaussian white noise with a signal-to-noise ratio of 0dB was added. The CVBT proposed in this method was used to perform frequency domain analysis on the echo, and the results were compared with those of the traditional FFT. The normalized comparison simulation results are as follows: Figure 2 As shown in the figure, the horizontal axis represents distance, and the vertical axis represents the amplitude of the echo after normalization via FFT and CVBT. The figure demonstrates that the method of this invention can effectively replace FFT for frequency domain analysis.

[0043] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0044] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0045] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in this invention may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A range domain target detection method for millimeter-wave radar based on complex binary transform, characterized in that, include: S1. Construct a training sample dataset containing different target distances and noise intensities; S2. Construct a complex binary transform network parameter model; S3. Construct a loss function using the GrandNorm strategy, solve the gradient vanishing problem in backpropagation of the binary neural network using a pass-through estimator, and train the CVBT network parameter model using the training sample dataset obtained in step S1 to obtain the trained CVBT network parameters; S4. Use the trained CVBT network parameter model to process the radar echo through complex binary transformation to obtain the range spectrum, and then use the range spectrum to perform constant false alarm rate (CFAR) detection to obtain the specific location of the target.

2. The millimeter-wave radar range domain target detection method based on complex binary transformation as described in claim 1, characterized in that, The process of constructing the training sample dataset in step S1 is as follows: S11, Select radar echo signals under noise-free conditions at different target distances, and then add noise with different signal-to-noise ratios to them to obtain noisy radar echo signals; S12, Perform fast Fourier transform on the noise-free and noisy radar echo signals respectively to obtain the corresponding noise-free range spectrum and noisy range spectrum; S13, Use the noise-free radar echo signal, noise-free range spectrum, noisy radar echo signal, and noisy range spectrum together to form a set of training samples, and repeat the above process to construct 10,000 sets of training sample datasets containing different target distances and noise intensities.

3. The millimeter-wave radar range domain target detection method based on complex binary transformation as described in claim 1, characterized in that, Step S2 constructs the CVBT network parameter model. The CVBT network parameter model includes a dual-path symmetric complex binary network structure, comprising a forward CVBT and an inverse CVBT. The radar echo is processed through the forward CVBT to obtain the range spectrum, which is then reconstructed using the inverse CVBT to obtain the inverse echo. The radar echo undergoes frequency domain transformation using the forward CVBT to complete frequency analysis. The inverse CVBT ensures the transformation is reversible; the radar signal is processed through the forward CVBT to obtain the range spectrum, and the radar echo signal is represented as... s r The operation matrix of positive CVBT is W b_CVBT The actual part and the imaginary part are represented as follows: ;in, ；S r_CVBT The distance spectrum is obtained by positive CVBT. To represent matrix multiplication, positive CVBT is represented as: The inverse echo is obtained by reconstructing the range spectrum using the inverse CVBT; the operation matrix of the inverse CVBT matrix is... W b_ICVBT The actual part and the imaginary part are represented as follows: ; S t_ICVBT The inverse echo obtained by inverse CVBT is represented as: .

4. The millimeter-wave radar range domain target detection method based on complex binary transformation as described in claim 1, characterized in that, Step S3 involves training the CVBT network model parameters. The training process includes constructing a loss function using the GrandNorm strategy to optimize the training of the transformation matrix; and using a pass-through estimator to solve the gradient vanishing problem in the backpropagation of the binary neural network.

5. The millimeter-wave radar range domain target detection method based on complex binary transform as described in claim 4, characterized in that, The constructed loss function consists of two parts: return loss and range spectrum loss, as shown below: ; in, S r_GT , s t_GT These are the distance spectrum and inverse echo obtained by FFT and inverse FFT under noise-free conditions, respectively. echo loss weight w range and distance spectrum loss weights w echo The network is adaptively adjusted based on the dynamic changes in the gradients corresponding to the two loss functions during training. The calculation relationship is expressed as follows: .

6. The millimeter-wave radar range domain target detection method based on complex binary transform as described in claim 4, characterized in that, The binary neural network parameter model of CVBT employs a pass-through estimator strategy during backpropagation training. The STE strategy quantizes continuous real values ​​into discrete forms represented only by binary {0,1}, approximating the estimation of the derivative of the binary function, thereby ensuring gradient stability and effective propagation during network training. Its binarization process is expressed as follows: .

7. The millimeter-wave radar range domain target detection method based on complex binary transformation as described in claim 1, characterized in that, The process of training the CVBT network parameters using the training sample dataset of S1 in step S3 is as follows: the CVBT network model constructed in step S2 is used to train the training sample dataset of step S1, wherein the initial learning rate is set to 0.01, and the learning rate is reduced to half of the original value after every 30 iterations. After completing 200 iterations of training, the CVBT network parameters are obtained.

Images