Peak value extraction method and system based on coherent Doppler laser radar echo signals

Abstract

The invention discloses a peak value extraction method and system based on coherent Doppler laser radar echo signals, and belongs to the technical field of laser wind finding radars, the echo signals are converted into intermediate frequency signals and sampled, and sampled echo signals are obtained; performing frequency domain conversion on the sampled echo signal, and performing preliminary noise reduction to obtain an echo signal power spectrum; truncating and splicing the echo signal power spectrum to obtain a two-dimensional power spectrum; inputting the two-dimensional power spectrum into a pre-trained convolutional neural network for peak feature extraction to obtain a peak frequency; and obtaining Doppler frequency shift of the echo signal based on the peak frequency, and inverting the radial wind speed based on the Doppler frequency shift. The spatial continuity of the wind speed is fully considered, and when the convolutional neural network is adopted to process the two-dimensional power spectrum, part of the peak submerged in the noise can be highlighted again due to the influence of the nearby distance gate peak, so that the peak extraction success rate of the region is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of laser wind radar, and in particular relates to a peak value extraction method and system based on a coherent Doppler laser radar echo signal. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] The detection of atmospheric wind fields is of great significance for climate change prediction, extreme weather disaster warning, improving the efficiency of wind energy resource utilization, and ensuring aerospace safety. Early wind measuring devices such as mechanical anemometers, sounding balloons, wind towers, microwave radars, etc. often have problems such as small detection range, high cost, inconvenient movement, insufficient accuracy, and large environmental impact. In recent years, wind laser radars with advantages such as strong spatiotemporal resolution, high measurement accuracy, long detection distance, and strong anti-interference ability have been widely used in real-time detection of three-dimensional wind fields in the atmosphere. Laser wind radar is a technology based on the laser Doppler effect. By measuring the frequency shift of the backscattered laser beam, the wind speed in the atmospheric flow field can be calculated, and multi-dimensional monitoring of different heights and wind speeds in the atmosphere can be achieved.

[0004] The principle of coherent Doppler laser wind radar is to verify, analyze and invert the Doppler frequency shift of the echo signal using a data processing algorithm. The key point is to accurately obtain the peak frequency of the echo signal power spectrum. However, the signal-to-noise ratio of the laser wind radar echo signal is generally negatively correlated with the detection distance. The farther the distance, the lower the signal-to-noise ratio, and the more difficult it is to find the correct peak. The radial wind speed inverted from an incorrect peak value will have errors, and even extremely exaggerated fallacies. The traditional solution is to set a signal-to-noise ratio threshold, eliminate the radial wind speed calculated by the range gate with a signal-to-noise ratio lower than the threshold, and then supplement it with interpolation, but this is obviously not convincing enough and it is difficult to determine a suitable threshold.

[0005] Therefore, how to improve the success rate of peak extraction in areas where the signal-to-noise ratio of the laser wind radar echo signal is low is a technical problem that needs to be solved urgently. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a peak extraction method and system based on coherent Doppler lidar echo signal, which utilizes the spatial continuity of wind speed to truncate the one-dimensional power spectrum of the radar echo signal in units of range gate and splice it into a two-dimensional power spectrum; and utilizes the spatial continuity of wind speed to use a convolutional neural network to extract the peak features of the two-dimensional power spectrum, thereby extracting the peak frequency.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] In a first aspect, the present invention provides a peak extraction method based on a coherent Doppler laser radar echo signal, comprising:

[0009] Acquire the echo signal of the laser radar, convert the echo signal into an intermediate frequency signal and sample it to obtain a sampled echo signal;

[0010] Performing frequency domain conversion on the sampled echo signal and performing preliminary noise reduction to obtain an echo signal power spectrum;

[0011] Truncating and splicing the echo signal power spectrum to obtain a two-dimensional power spectrum;

[0012] Inputting the two-dimensional power spectrum into a pre-trained convolutional neural network to extract peak features and obtain peak frequency;

[0013] A Doppler frequency shift of the echo signal is obtained based on the peak frequency, and a radial wind speed is inverted based on the Doppler frequency shift.

[0014] A further technical solution is to obtain the echo signal power spectrum as follows:

[0015] Performing fast Fourier transform on the sampled echo signal to obtain a transformed frequency domain signal;

[0016] Calculating the power of each frequency component according to the frequency spectrum of the frequency domain signal to obtain the power spectrum density of the echo signal;

[0017] Through pulse accumulation and averaging, the power spectrum noise is preliminarily reduced to obtain the one-dimensional power spectrum of the echo signal.

[0018] A further technical solution is to obtain a two-dimensional power spectrum as follows:

[0019] The one-dimensional power spectrum of the echo signal is cut off in units of range gates to obtain the power spectrum of each range gate;

[0020] The truncated power spectra of the range gates are vertically spliced ​​into a two-dimensional power spectrum.

[0021] According to a further technical solution, the convolutional neural network includes, in sequence, an input layer, a convolutional layer, a ReLU layer, a maximum pooling layer, a convolutional layer, a ReLU layer, a maximum pooling layer, a fully connected layer, a ReLU layer, a fully connected layer, a ReLU layer and a regression layer.

[0022] A further technical solution is to obtain the peak frequency as follows:

[0023] The two-dimensional power spectrum is input into the trained convolutional neural network for processing to obtain the reconstructed target two-dimensional power spectrum;

[0024] The data of each range gate in the target two-dimensional power spectrum is normalized to highlight the peak of each range gate and extract the peak frequency of each range gate.

[0025] A further technical solution is to invert the radial wind speed as follows:

[0026] The Doppler frequency shift of each range gate signal is obtained by subtracting the modulation frequency of the radar signal from the peak frequency of each range gate;

[0027] The radial wind speed is finally calculated through the relationship between Doppler frequency shift and radial wind speed.

[0028] In a further technical solution, the relationship is:

[0029]

[0030] Among them, f d is the Doppler shift, v is the radial wind speed, and λ is the wavelength.

[0031] In a second aspect, the present invention provides a peak extraction system based on a coherent Doppler laser radar echo signal, comprising:

[0032] The signal sampling module is configured to: obtain the echo signal of the laser radar, convert the echo signal into an intermediate frequency signal and sample it to obtain a sampled echo signal;

[0033] A frequency domain conversion module is configured to: perform frequency domain conversion on the sampled echo signal, and obtain an echo signal power spectrum after performing preliminary noise reduction;

[0034] A signal splicing module is configured to: truncate and splice the echo signal power spectrum to obtain a two-dimensional power spectrum;

[0035] A peak extraction module is configured to: input the two-dimensional power spectrum into a pre-trained convolutional neural network to extract peak features and obtain peak frequency;

[0036] The wind speed inversion module is configured to obtain a Doppler frequency shift of an echo signal based on the peak frequency, and invert radial wind speed based on the Doppler frequency shift.

[0037] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the peak extraction method based on the coherent Doppler lidar echo signal as described in the first aspect.

[0038] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the peak extraction method based on the coherent Doppler lidar echo signal as described in the first aspect are implemented.

[0039] One or more of the above technical solutions have the following beneficial effects:

[0040] The present invention utilizes the spatial continuity of wind speed to truncate the one-dimensional power spectrum of the radar echo signal in units of range gates and then splice it into a two-dimensional power spectrum; and utilizes the spatial continuity of wind speed to extract the peak features of the two-dimensional power spectrum using a convolutional neural network, thereby extracting the peak frequency.

[0041] In low signal-to-noise ratio areas, some peaks of the power spectrum of the laser wind radar echo signal are submerged in the noise, and one-dimensional power spectrum peak extraction is difficult to be effective. The present invention fully considers the spatial continuity of wind speed, that is, the change of wind speed in a continuous space is also continuous and will not occur suddenly. Therefore, when the present invention uses a convolutional neural network to process the two-dimensional power spectrum, some peaks submerged in the noise will be re-highlighted due to the influence of the nearby range gate peak, thereby improving the success rate of peak extraction in this area.

[0042] When the traditional scheme extracts the peak value, due to the limitation of the lidar hardware, there is often a minimum step in the change of the peak frequency. This step will generally result in an error of less than 0.5m / s in the radial wind speed inversion. However, the peak frequency of the power spectrum after convolution in the present invention does not have such a minimum step, and there is no radial wind speed error caused by this.

[0043] In the present invention, the peak value in the low signal-to-noise ratio area will be re-highlighted due to the influence of the nearby range gate peak value, so that the peak frequency of the area extracted in this way has higher credibility than the traditional interpolation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0045] Figure 1 It is a flow chart of a peak extraction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0047] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0048] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0049] Embodiment 1

[0050] like Figure 1 As shown, this embodiment discloses a peak extraction method based on a coherent Doppler laser radar echo signal, the method comprising the following steps:

[0051] S1: Acquire the echo signal of the laser radar, convert the echo signal into an intermediate frequency signal and sample it to obtain a sampled echo signal;

[0052] S101: The receiving module of the coherent Doppler laser wind measurement radar receives the signal backscattered by the atmospheric aerosol particles. Since there is a relative speed between the aerosol particles and the radar, a corresponding Doppler frequency shift is generated in the backscattered signal.

[0053] S102: The laser radar echo signal and the local oscillator signal are input into the coupler together. The output result of the coupler is converted into an intermediate frequency electrical signal through a balanced detector. The intermediate frequency signal contains the Doppler frequency shift described in step S101.

[0054] S103: Use the acquisition card to sample the intermediate frequency electrical signal to obtain a sampled echo signal. The Nyquist sampling theorem must be met during sampling.

[0055] S2: performing frequency domain conversion on the sampled echo signal and performing preliminary noise reduction to obtain an echo signal power spectrum;

[0056] S201: Perform a fast Fourier transform (FFT) on the sampled echo signal to obtain a transformed frequency domain signal. This step converts the time domain signal into a frequency domain signal.

[0057] S202: Calculate the power of each frequency component according to the obtained frequency spectrum of the frequency domain signal to obtain the power spectrum density of the echo signal.

[0058] S203: Preliminarily reduce the power spectrum noise by pulse accumulation and averaging, and obtain the one-dimensional power spectrum of the echo signal. Specifically, the wavelength of the radar output light is 1550 nanometers, the repetition frequency is 10kHz, and the repetition frequency of the echo signal is also 10kHz. By accumulating and averaging these 10k power spectra, the noise can be preliminarily reduced.

[0059] S3: truncating and splicing the power spectrum of the echo signal to obtain a two-dimensional power spectrum;

[0060] S301: The obtained one-dimensional power spectrum of the laser radar echo signal is cut off by range gate to obtain the power spectrum of each range gate. A single range gate of the laser wind measurement radar used in this embodiment contains 112 sampling points, which means that the power spectrum of a single range gate contains 112 power density data. Peak search is to find the correct peak value in these 112 data.

[0061] S302: vertically splicing the truncated power spectrum data of each range gate into a two-dimensional power spectrum. The spatial continuity of the wind speed is utilized, and the peak positions of the spliced ​​two-dimensional power spectrum are continuous.

[0062] S4: inputting the two-dimensional power spectrum into a pre-trained convolutional neural network to extract peak features and obtain peak frequency;

[0063] S401: Construct and train convolutional neural network layers to obtain a trained convolutional neural network. Specifically: set the input layer size to 112*50; add two convolutional layers, use 3*3 convolution kernels, the first convolutional layer has 16 filters, the second convolutional layer has 32 filters, and both convolutional layers use 'same' to keep the feature map size unchanged; add a ReLU layer after each convolutional layer; add a maximum pooling layer to each of the two activation layers, use a 2x2 window, and a step size of 2; add two fully connected layers, the first fully connected layer, 128 neurons, the second fully connected layer, 64 neurons; add a ReLU activation layer to each of the two fully connected layers; set a regression layer.

[0064] It can be seen that the convolutional neural network includes input layer, convolution layer, ReLU layer, maximum pooling layer, convolution layer, ReLU layer, maximum pooling layer, fully connected layer, ReLU layer, fully connected layer, ReLU layer and regression layer in sequence.

[0065] Select power spectrum data with low noise and clear peak features as training data XTrain, and the data set size is 112*50*N, where N is the number of samples. Specify the training options, the maximum number of training iterations is 20, the size of each batch is 64, and the initial learning rate is 0.001. Use the training data XTrain to train the convolutional neural network, extract the peak features of the two-dimensional power spectrum, and obtain the peak frequency.

[0066] In this embodiment, the size of the training data set XTrain is 112*50*N, where N is the number of samples. In this embodiment, N=100, each sample contains 50 range gates, each range gate contains 112 sampling points, and the original echo signal power spectrum of the radar contains 128 range gates. Considering that the actual performance of the radar is difficult to detect the high-altitude wind speed beyond 3 kilometers, only 50 range gate data are selected for the experiment. A total of four ReLU activation layers are added to the convolutional neural network, and the ReLU nonlinear function is applied to increase the nonlinearity of the network; two maximum pooling layers are also added to reduce the size of the feature map to prevent overfitting while retaining important features; the fully connected layer is used to integrate the convolution layer and the pooling layer to extract local features from the input image to form a global feature representation. Compared with the initial input data, the predicted two-dimensional power spectrum data processed by the convolutional neural network has more prominent peak features, further reduced noise, improved power spectrum smoothness, and increased peak extraction success rate. The collected echo information is processed to obtain a one-dimensional power spectrum. This process is completed in the signal processing module inside the radar, and the process of peak extraction of the one-dimensional power spectrum is performed on Matlab.

[0067] S402: inputting the two-dimensional power spectrum to be measured into the trained convolutional neural network for processing to obtain a reconstructed target two-dimensional power spectrum;

[0068] In this embodiment, the two-dimensional power spectrum data to be tested is test data XTest, and its size is 112*50*M, where M is the number of two-dimensional power spectra to be tested. The trained convolutional neural network is used to process the input two-dimensional power spectrum data to be tested XTest to obtain the reconstructed target two-dimensional power spectrum data YPred.

[0069] S403: normalizing the data of each range gate in the target two-dimensional power spectrum, highlighting the peak of each range gate, and extracting the peak frequency of each range gate.

[0070] S5: Obtaining a Doppler frequency shift of the echo signal based on the peak frequency, and inverting the radial wind speed based on the Doppler frequency shift.

[0071] S501: Obtain the Doppler frequency shift of each range gate signal by subtracting the modulation frequency obtained by the radar signal in the transmitting module from the extracted peak frequency of each range gate.

[0072] S502: According to the Doppler frequency shift, the relative speed of the aerosol particles relative to the laser radar, that is, the radial wind speed, is finally calculated through the following relationship between the Doppler frequency shift and the radial wind speed. The relationship is expressed as:

[0073]

[0074] Among them, f d is the Doppler shift, v is the radial wind speed, and λ is the wavelength.

[0075] The coherent laser wind measurement radar mentioned in this embodiment is a coherent Doppler laser wind measurement radar that uses a pulsed laser with a wavelength of 1550nm, a pulse width of 400ns, and a pulse repetition frequency of 10KHz as the output light. The radar system includes a laser emission module, a receiving module, and a signal processing module. The seed source of the laser emission module excites the laser to generate linearly polarized light with a center frequency of f0, which is divided into an emission signal and a local oscillator signal after passing through a beam splitter. The laser emission pulse is modulated by acousto-optic modulation to a frequency of f0+f AOM The intermediate frequency pulse light is then amplified and emitted. The echo signal will produce f compared to the emitted signal. d The Doppler frequency shift of the backscattered echo signal received by the receiving module is f d +f0+f AOM The intermediate frequency signal described in S1 is the beat frequency signal of the echo signal and the local oscillator signal, which is converted into a frequency of f by a balanced detector. d +f AOM The basic principle of laser radar wind measurement is to obtain the Doppler frequency shift f d , thereby calculating the radial wind speed. The peak extraction method of the present invention is mainly completed in the signal processing module of the radar system. The signal processing module mainly includes signal acquisition, signal processing, wind field inversion and other parts. The signal sampling frequency is 1GHz, the number of sampling points is 112, the number of distance gates is 128, and each distance gate represents a detection distance of 60 meters.

[0076] The present invention fully considers the spatial continuity of wind speed, and uses spectrum splicing to perform data analysis and processing on radar signal power in order to extract the power spectrum peak; and further performs data processing on the two-dimensional power spectrum of the radar echo signal, extracts the two-dimensional power spectrum peak features through a convolutional neural network, and thus extracts the power spectrum peak frequency.

[0077] Embodiment 2

[0078] This embodiment discloses a peak extraction system based on a coherent Doppler laser radar echo signal, comprising:

[0079] The signal sampling module is configured to: obtain the echo signal of the laser radar, convert the echo signal into an intermediate frequency signal and sample it to obtain a sampled echo signal;

[0080] A frequency domain conversion module is configured to: perform frequency domain conversion on the sampled echo signal, and obtain an echo signal power spectrum after performing preliminary noise reduction;

[0081] A signal splicing module is configured to: truncate and splice the echo signal power spectrum to obtain a two-dimensional power spectrum;

[0082] A peak extraction module is configured to: input the two-dimensional power spectrum into a pre-trained convolutional neural network to extract peak features and obtain peak frequency;

[0083] The wind speed inversion module is configured to obtain a Doppler frequency shift of an echo signal based on the peak frequency, and invert radial wind speed based on the Doppler frequency shift.

[0084] Embodiment 3

[0085] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of embodiment 1 when executing the program.

[0086] Embodiment 4

[0087] The purpose of this embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method of embodiment 1 are performed.

[0088] The steps involved in the apparatus of the above embodiments 3 and 4 correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0089] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0091] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A peak extraction method based on coherent Doppler laser radar echo signal, characterized in that: include: Acquire the echo signal of the laser radar, convert the echo signal into an intermediate frequency signal and sample it to obtain a sampled echo signal; Performing frequency domain conversion on the sampled echo signal and performing preliminary noise reduction to obtain an echo signal power spectrum; Truncating and splicing the echo signal power spectrum to obtain a two-dimensional power spectrum; Inputting the two-dimensional power spectrum into a pre-trained convolutional neural network to extract peak features and obtain peak frequency; A Doppler frequency shift of the echo signal is obtained based on the peak frequency, and a radial wind speed is inverted based on the Doppler frequency shift.

2. The peak extraction method based on the coherent Doppler laser radar echo signal according to claim 1, characterized in that: The power spectrum of the echo signal is obtained as follows: Performing fast Fourier transform on the sampled echo signal to obtain a transformed frequency domain signal; Calculating the power of each frequency component according to the frequency spectrum of the frequency domain signal to obtain the power spectrum density of the echo signal; Through pulse accumulation and averaging, the power spectrum noise is preliminarily reduced to obtain the one-dimensional power spectrum of the echo signal.

3. The peak extraction method based on the coherent Doppler laser radar echo signal according to claim 2, characterized in that: The two-dimensional power spectrum is obtained as follows: The one-dimensional power spectrum of the echo signal is cut off in units of range gates to obtain the power spectrum of each range gate; The truncated power spectra of the range gates are vertically spliced ​​into a two-dimensional power spectrum.

4. The peak extraction method based on the coherent Doppler laser radar echo signal according to claim 1, characterized in that: The convolutional neural network includes an input layer, a convolution layer, a ReLU layer, a maximum pooling layer, a convolution layer, a ReLU layer, a maximum pooling layer, a fully connected layer, a ReLU layer, a fully connected layer, a ReLU layer and a regression layer in sequence.

5. The peak extraction method based on the coherent Doppler laser radar echo signal according to claim 1, characterized in that: The peak frequency is obtained as follows: The two-dimensional power spectrum is input into the trained convolutional neural network for processing to obtain the reconstructed target two-dimensional power spectrum; The data of each range gate in the target two-dimensional power spectrum is normalized to highlight the peak of each range gate and extract the peak frequency of each range gate.

6. The peak extraction method based on the coherent Doppler laser radar echo signal according to claim 1, characterized in that: The inverted radial wind speed is specifically: The Doppler frequency shift of each range gate signal is obtained by subtracting the modulation frequency of the radar signal from the peak frequency of each range gate; The radial wind speed is finally calculated through the relationship between Doppler frequency shift and radial wind speed.

7. The peak extraction method based on the coherent Doppler laser radar echo signal according to claim 6, characterized in that: The relationship is: Among them, f d is the Doppler shift, v is the radial wind speed, and λ is the wavelength.

8. A peak extraction system based on coherent Doppler laser radar echo signal, characterized in that: include: The signal sampling module is configured to: obtain the echo signal of the laser radar, convert the echo signal into an intermediate frequency signal and sample it to obtain a sampled echo signal; A frequency domain conversion module is configured to: perform frequency domain conversion on the sampled echo signal, and obtain an echo signal power spectrum after performing preliminary noise reduction; A signal splicing module is configured to: truncate and splice the echo signal power spectrum to obtain a two-dimensional power spectrum; A peak extraction module is configured to: input the two-dimensional power spectrum into a pre-trained convolutional neural network to extract peak features and obtain peak frequency; The wind speed inversion module is configured to obtain a Doppler frequency shift of an echo signal based on the peak frequency, and invert radial wind speed based on the Doppler frequency shift.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the peak extraction method based on the coherent Doppler laser radar echo signal as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the peak extraction method based on the coherent Doppler lidar echo signal as described in any one of claims 1 to 7 are implemented.

Images