Laser radar signal processing method combining improved sparrow search algorithm and dynamic threshold VMD
By improving the sparrow search algorithm and combining dynamic thresholding (VMD) with wavelet transform, the problem of traditional VMD algorithms relying on manual parameter settings is solved, achieving efficient denoising and accurate signal separation of lidar signals, and improving the efficiency and accuracy of signal processing.
Patent Information
- Application Number
- CN202511410247.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2025-12-26
AI Technical Summary
Traditional VMD algorithms rely on manual experience to set parameters in lidar signal processing, resulting in low efficiency and insufficient denoising accuracy. They are difficult to effectively separate signals from noise, and their performance is limited, especially under complex weather conditions.
By combining an improved sparrow search algorithm and dynamic thresholding (VMD), an iterative optimization is performed by constructing a fitness function, and modal segmentation is performed by dynamically adjusting the threshold. Wavelet transform is then used for fine denoising, achieving adaptive optimization of VMD parameters and accurate distinction between signal and noise.
It improves the signal-to-noise ratio and denoising accuracy, enhances the inversion accuracy and detection range of lidar signals, and has strong adaptability, making it suitable for complex noise environments.
Smart Images
Figure CN121208792A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of lidar signal processing technology, and specifically to a lidar signal processing method that combines an improved sparrow search algorithm and dynamic threshold VMD. Background Technology
[0002] LiDAR, especially 1550nm superconducting nanowire single-photon detector (SNSPD) LiDAR systems, is an important tool for atmospheric sounding and plays a crucial role in environmental monitoring, climate research, and other fields. However, in practical applications, especially under complex weather conditions such as rain or strong sunlight, photon signals attenuate during long-distance transmission, resulting in extremely weak echo signals received by the equipment. Simultaneously, noise photons from background light and the equipment itself become particularly prominent, mixing with the effective signal and severely impacting the system's signal-to-noise ratio (SNR). This, in turn, reduces the accuracy of signal parameter inversion and shortens the effective detection range.
[0003] To address the aforementioned noise problem, various signal denoising algorithms have been employed in existing technologies, such as Wavelet Transform (WT), Empirical Mode Decomposition (EMD), and Variational Mode Decomposition (VMD). Among these methods, VMD has attracted attention due to its superior performance in processing nonlinear and non-stationary lidar echo signals.
[0004] However, traditional VMD algorithms have the following obvious drawbacks in practical applications: Parameter selection relies on human experience, leading to inefficiency: The denoising effect of the VMD algorithm largely depends on the setting of two key parameters: the number of decomposition modes K and the quadratic penalty factor α. In existing technologies, these two parameters usually need to be manually set based on the researcher's experience. This process is not only time-consuming and inefficient, but also makes it difficult to find the globally optimal parameter combination, limiting the algorithm's adaptability and processing efficiency.
[0005] Insufficient noise reduction accuracy can easily lead to mode aliasing: Inappropriate parameter selection can cause incomplete or excessive signal decomposition, resulting in mode aliasing. That is, different signal components are incorrectly classified into the same mode, or the same signal component is decomposed into different modes. This will seriously affect the subsequent noise identification and removal effect, and will make it impossible to effectively separate useful low-frequency signals from noise.
[0006] Limited ability to process wideband signals: Traditional VMD algorithms are limited in performance when processing wideband signals containing complex background noise, making it difficult to achieve ideal noise reduction results.
[0007] Therefore, how to achieve adaptive optimization of VMD algorithm parameters and improve its signal denoising accuracy and robustness in complex noise environments is a technical problem that urgently needs to be solved in the field of lidar signal processing. Summary of the Invention
[0008] To address the aforementioned shortcomings in the existing technology, this invention provides a lidar signal processing method that combines an improved sparrow search algorithm and dynamic threshold VMD.
[0009] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD includes the following steps: Acquire the echo signal from the lidar; A fitness function is constructed by fusing dynamic spectral amplitude difference and signal smoothness. Based on the constructed fitness function, an improved sparrow search algorithm is used to iteratively optimize the echo signal to obtain the optimal variational mode decomposition parameters. Variational mode decomposition is performed on the echo signal based on the optimal variational mode decomposition parameters to obtain multiple mode components; A dynamic threshold mechanism is used to perform spectral peak segmentation on each modal component, select effective modal components, and reconstruct the signal to obtain the echo signal after preliminary denoising. Wavelet transform is performed on the echo signal after preliminary denoising to obtain the final denoised signal.
[0010] Furthermore, the fitness function is constructed by fusing the dynamic spectral amplitude difference and signal smoothness as follows: D=|fs|+|ms| Where D is the fitness, f is the number of points in the spectrum of the noisy signal whose peak intensity exceeds the preset basic threshold, s is the theoretical signal characteristic value, and m is the number of noise points with random fluctuations in the time domain signal.
[0011] Furthermore, based on the constructed fitness function, an improved sparrow search algorithm is used to iteratively optimize the echo signal to obtain the optimal variational mode decomposition parameters, including: Within the preset search range of variational mode decomposition parameters, a set number of parameter combinations are randomly generated as the initial sparrow population, with each parameter combination corresponding to one individual sparrow. Substitute each set of parameters into the VMD model to decompose the original echo signal, and then calculate the fitness value of each sparrow individual through the fitness function. Sort the current global best fitness and global worst fitness. Perform differentiated position updates based on the Sparrow role; The iteration terminates when the number of iterations reaches the maximum number of iterations, or when the change in the optimal fitness value after a set number of consecutive iterations is less than a preset threshold, and the optimal variational mode decomposition parameters corresponding to the current global optimal fitness are output.
[0012] Furthermore, variational mode decomposition is performed on the echo signal based on the optimal variational mode decomposition parameters to obtain multiple mode components, including: Substitute the optimal variational mode decomposition parameters into the VMD constrained variational model; The VMD-constrained variational model was iteratively calculated using the alternating direction multiplier method to obtain multiple modal components; among them, the low-frequency IMF contains the effective signal, and the high-frequency IMF contains background noise and shot noise.
[0013] Furthermore, the use of a dynamic threshold mechanism for spectral peak segmentation of each modal component includes: Perform a fast Fourier transform on each modal component to obtain the corresponding amplitude-frequency spectrum; The position where the amplitude value is greater than the adjacent frequency points on the left and right is defined as the peak value of the spectrum. The spectrum of all modal components is traversed, the local peak intensity of each spectrum is counted, and the global peak intensity of the original echo signal spectrum is calculated. The local peak intensity of each spectrum is used as a dynamic threshold and compared with a set proportion of the global peak intensity. When the local peak intensity is less than the set proportion of the global peak intensity, the dynamic threshold is corrected to the set proportion of the global peak intensity.
[0014] Furthermore, the screening of effective modal components includes: The peak intensity of the spectrum of each modal component is compared with the dynamic threshold. If all local peak intensities of a modal component are less than the dynamic threshold, it is determined to be a pure noise mode and is removed. If at least one local peak intensity of a modal component is greater than the dynamic threshold, it is determined to be a dominant mode of effective signal and the energy proportion exceeding the dynamic threshold is calculated. If the energy proportion is greater than the set proportion threshold, the modal component is retained; otherwise, the modal component is removed.
[0015] Further, signal reconstruction is performed to obtain the preliminary denoised echo signal, including: The selected effective modal components are linearly superimposed in their original time-domain order to obtain the echo signal after preliminary denoising.
[0016] Furthermore, the signal-to-noise ratio and normalized mean square error of the echo signal after preliminary denoising are calculated, and the calculation results are compared with the original echo signal to verify the denoising effect. If the denoising requirements are met, the echo signal after preliminary denoising is retained; otherwise, the dynamic threshold mechanism is readjusted to perform spectral peak segmentation on each modal component.
[0017] Furthermore, wavelet transform is performed on the initially denoised echo signal to obtain the final denoised signal, including: By selecting a wavelet basis that is compatible with the lidar signal, the echo signal after preliminary denoising is decomposed into multiple layers to obtain the approximation coefficients and detail coefficients of each layer. A soft threshold is applied to the detail coefficients of each layer; The processed detail coefficients and approximation coefficients are subjected to inverse wavelet transform to obtain the final denoised signal.
[0018] Furthermore, the signal-to-noise ratio, normalized mean square error, and smoothness of the final denoised signal are calculated, and the calculation results are compared with the original echo signal to verify the denoising effect. If the denoising requirements are met, the final denoised signal is retained; otherwise, the dynamic threshold mechanism is readjusted to perform spectral peak segmentation on each modal component.
[0019] The present invention has the following beneficial effects: This invention addresses the inefficiencies and unstable decomposition accuracy of traditional Variational Mode Decomposition (VMD) due to its reliance on manually selected parameters (number of modes K, quadratic penalty factor α). It achieves adaptive optimization of VMD parameters through an improved Sparrow Search Algorithm (FSSA) based on a new fitness function. This not only avoids dependence on human experience but also converges faster than the standard Sparrow Search Algorithm (SSA), efficiently finding the optimal parameter combination. Furthermore, it addresses the bottleneck of traditional algorithms such as Wavelet Transform (WT), VMD, and VMD-WT in suppressing the superposition of background noise and shot noise in complex weather conditions by fusing... The Dynamic Peak Thresholding (TVMD) algorithm based on time-frequency analysis accurately distinguishes between signal and noise modes. Combined with multi-scale analysis by WT, it further suppresses residual noise. Experiments show that its signal-to-noise ratio (SNR) is significantly improved and the normalized mean square error (NMSE) is reduced compared to VMD, which is significantly better than traditional algorithms. At the same time, the algorithm has stronger robustness and adaptability, and can stably retain useful signals under different noise intensities. It effectively improves the signal inversion accuracy and effective detection range of 1550nm SNSPD lidar, providing reliable technical support for real-time applications such as long-term unattended environmental monitoring. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of a LiDAR signal processing method that combines an improved sparrow search algorithm and dynamic threshold VMD. Detailed Implementation
[0021] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0022] The shortcomings of existing technologies necessitate addressing the following technical issues: Firstly, Variational Mode Decomposition (VMD) relies on manually selecting the number of modes K and the secondary penalty factor α, which is not only inefficient but also prone to mode aliasing or omission of key frequency components due to parameter deviations, becoming a fundamental bottleneck restricting VMD's denoising performance. Secondly, under complex weather conditions (such as rainy days or sunny days with strong background light), traditional algorithms such as Wavelet Transform (WT), VMD, and VMD-WT struggle to effectively suppress the superposition interference of background noise (solar infrared radiation, raindrop scattering) and shot noise, resulting in limited improvement in the signal-to-noise ratio (SNR) of the lidar echo signal and insufficient reduction in normalized mean square error (NMSE), directly impacting the signal parameter inversion accuracy and detection range of the 1550nm SNSPD lidar. Thirdly, VMD has low efficiency in processing broadband signals and cannot effectively separate low-frequency useful signals from noise, easily leading to the loss of useful information. Finally, the standard Sparrow Search (SSA) algorithm, when used for VMD parameter optimization, suffers from insufficient population diversity, fixed role ratios, and slow convergence speed, making it difficult to efficiently obtain optimal parameters.
[0023] The main objectives of this invention are to address the aforementioned technical problems one by one: The primary objective is to achieve adaptive optimization of VMD parameters. This is achieved by proposing an improved Sparrow Search Algorithm (FSSA) based on a new fitness function (integrating dynamic spectral amplitude difference and signal smoothness). This algorithm dynamically adjusts the proportion of sparrow roles to balance global exploration and local development, automatically searching for the optimal K and α, thus avoiding the drawbacks of manual parameter selection. The core objective is to improve the denoising performance of lidar echo signals under complex weather conditions. This is achieved by constructing a joint FSTV-WT algorithm. First, a VMD based on dynamic peak threshold (TVMD) is used—setting a threshold based on signal spectral characteristics to distinguish signal and noise modal components. Then, the multi-scale analysis capability of WT is combined to suppress residual noise, ultimately achieving improved SNR and reduced NMSE. The secondary objective is to optimize the broadband signal processing and low-frequency signal separation capabilities of VMD. By leveraging the "decomposition-spectral peak calculation-dynamic filtering" process of TVMD, broadband noise is efficiently processed while accurately retaining useful low-frequency signals. The auxiliary objective is to improve the convergence speed of the parameter optimization algorithm. Through the improved design of FSSA, the convergence efficiency is significantly accelerated, meeting the real-time detection requirements of lidar.
[0024] The basic idea of this invention is to propose a multi-stage signal denoising method that combines adaptive parameter optimization, dynamic threshold filtering, and multi-scale analysis. This method automatically obtains the optimal parameters of variational mode decomposition (VMD) through an improved intelligent optimization algorithm, then uses the optimized VMD for preliminary denoising, and finally performs refined processing through wavelet transform.
[0025] like Figure 1 As shown, an embodiment of the present invention provides a lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD, comprising the following steps S1 to S5: S1. Obtain the echo signal from the lidar; In this embodiment, step S1 acquires the original echo signal and initializes the algorithm parameters. First, the input original signal is the echo signal of a 1550nm superconducting nanowire single-photon detector (SNSPD) lidar. This signal contains useful backscattered information, but is affected by complex weather interference, exhibiting non-stationary and strong noise characteristics. The signal photon count range is 200-6000, and the detection distance covers 0-12km. Secondly, the key parameters of the algorithm were initialized: For the improved Sparrow Search Algorithm (FSSA), the population size was set to 30 (representing the number of VMD parameter combinations to be optimized), the maximum number of iterations was set to 100 (to balance optimization accuracy and efficiency), the initial value of the discoverer ratio (responsible for global exploration) was set to a high level, and the warning value R2∈[0,1] and the safety value ST∈[0.5,1.0] were set; For Variational Mode Decomposition (VMD), the parameter search range was set to the number of modes K∈[1,20] and the quadratic penalty factor α∈[1,6000] (based on the effective range verified by previous experiments, to avoid parameter overflow leading to decomposition failure); at the same time, the basic parameters of wavelet transform (WT) were initialized.
[0026] S2. A fitness function is constructed by fusing dynamic spectral amplitude difference and signal smoothness. Based on the constructed fitness function, an improved sparrow search algorithm is used to iteratively optimize the echo signal to obtain the optimal variational mode decomposition parameters. In this embodiment, step S2 constructs a fitness function to evaluate the merits of the parameters. This step aims to establish a quantitative index to achieve an objective evaluation of the merits of the VMD parameter (K, α) combination, which is the core guiding principle for FSSA optimization. The fitness function is designed as a dual-objective fusion form, specifically: D=|fs|+|ms| Where D represents fitness, f represents the number of points in the noisy signal spectrum where the peak intensity exceeds a preset basic threshold (reflecting the degree of noise interference on the signal spectrum, which needs to be determined through prior statistical analysis of the spectrum of similar LiDAR signals); m represents the number of randomly fluctuating noise points in the time-domain signal (calculated by the first-order difference of the signal to quantify the signal smoothness); and s represents the theoretical signal characteristic value (derived based on the number of peak points in the spectrum and the time-domain smoothness of the noise-free ideal LiDAR echo signal, or obtained through statistical fitting of a large number of clean signals). The physical meaning of this function is: to evaluate the ability of VMD to retain the signal spectrum characteristics under the parameter combination through the "spectral amplitude difference term (|fs|)," and to evaluate the noise suppression effect of the parameter combination through the "signal smoothness term (|ms|)." The smaller the function value, the more efficient the VMD can separate noise while retaining the useful signal (scattering-related low-frequency components), meeting the dual requirements of "signal integrity - noise suppression" in LiDAR signal denoising, and avoiding parameter optimization bias caused by a single indicator.
[0027] S3. Perform variational mode decomposition on the echo signal based on the optimal variational mode decomposition parameters to obtain multiple mode components; In this embodiment, step S3 executes the improved Sparrow Search Algorithm (FSSA) iterative optimization to output the optimal parameter combination. This step, through the adaptive iteration of FSSA, achieves automatic search for the optimal VMD parameters (K, α), solving the efficiency and accuracy bottlenecks of traditional VMD which relies on manual parameter selection. The specific process is as follows: 1. Population initialization: Within the preset search range of K and α, 30 sets of parameter combinations are randomly generated as the initial sparrow population, with each set of parameters corresponding to 1 "sparrow individual"; 2. Fitness Evaluation: Substitute each set of parameters into the VMD algorithm to decompose the original signal, then calculate the fitness value of each individual using the fitness function, and sort them to obtain the current global optimal fitness f. g (corresponding to the optimal parameter X) best The worst global fitness f w (corresponding to the worst-case parameter X) worst ); 3. Population Dynamic Update: Implement differentiated position update strategies based on sparrow roles (discoverer, follower, vigilant). 4. Termination judgment: When the number of iterations reaches 100, or the change in the optimal fitness value in 10 consecutive iterations is less than the preset threshold, the iteration terminates and outputs the K and α parameter combination corresponding to the current global optimal fitness. This combination can avoid mode aliasing and over-decomposition, and ensure the optimal VMD decomposition effect.
[0028] This step performs VMD decomposition on the original signal based on optimal parameters to obtain multiple modal components. The specific process is as follows: VMD Model Construction: Substitute the optimal parameters K (number of modes) and α (quadratic penalty factor) into the constrained variational model of VMD. Iterative solution of IMF: Iterative calculation is performed using the alternating direction multiplier method (ADMM): First, fix u k Solve for the center frequency of each mode, and then fix w. k Update u k Finally, update λi; when the change in IMF between two adjacent iterations is less than the preset tolerance, the decomposition terminates; Modal component output: Finally, K IMF components are obtained, each corresponding to information in different frequency ranges of the original signal. Among them, the low-frequency IMF mainly contains the useful backscattered signal, and the high-frequency IMF mainly contains background noise and shot noise, realizing the initial separation of signal and noise in the frequency domain.
[0029] S4. A dynamic threshold mechanism is used to perform spectral peak segmentation on each modal component, select effective modal components, and reconstruct the signal to obtain the echo signal after preliminary denoising. In this embodiment, step S4 dynamically calculates and sets the peak threshold for mode filtering based on the signal spectrum characteristics. This step designs a dynamic threshold mechanism for the output IMF component to accurately distinguish between "useful signal modes and noise modes," which is the core innovation of the improved VMD (TVMD). The specific calculation and setting logic is as follows: Spectrum analysis preprocessing: Perform Fast Fourier Transform (FFT) on each IMF component to obtain its amplitude-frequency spectrum; smooth the spectrum through a sliding window to suppress peak misjudgment caused by high-frequency line fluctuations; Peak identification: Define "spectral peak" as the position where the amplitude value is simultaneously greater than the adjacent frequency points on the left and right. Traverse the spectrum of all IMFs, count the local peak intensity of each spectrum, and calculate the global peak intensity of the original signal spectrum. Threshold verification and adjustment: Compare the calculated T with the global peak intensity P global Compare 1 / 3 of T, if T <P global / 3, then T is corrected to P global / 3, to avoid the useful signal mode being misjudged as noise due to the threshold being too low; the final output dynamic threshold T can both cover the upper limit of the peak value of the noise mode and ensure that the peak value of the useful signal mode is fully preserved.
[0030] This step filters and reconstructs the modal components, completing the initial denoising. Based on a dynamic threshold, this step filters and reconstructs the IMF components, completing the "coarse denoising" process and providing a high-quality signal foundation for subsequent fine denoising. The specific process is as follows: Modal selection logic: Compare the spectral peak intensity of each IMF component with the dynamic threshold T one by one: If all local peak intensities of an IMF are less than T, it is determined to be a "pure noise mode" (such as a high-frequency IMF, mainly containing raindrop scattering and shot noise) and is eliminated; if an IMF has at least one local peak intensity greater than T, it is determined to be a "useful signal-dominated mode" (such as a low-frequency IMF, containing the core information of scattering) and is retained; for "mixed modes" (some peaks exceed T, some are below T, mostly mid-to-high frequency IMFs), the energy ratio of the peaks exceeding the threshold is further calculated (total energy of peaks exceeding the threshold / total energy of the IMF). If the ratio is greater than 50% (ensuring that useful signal energy dominates), the mode is retained; otherwise, it is eliminated. Signal reconstruction: All retained "dominant modes of useful signals" are linearly superimposed in their original time-domain order to obtain a preliminary denoised signal; during the reconstruction process, the phase information of each IMF must be kept unchanged (to avoid phase distortion causing errors in the calculation of the lidar signal delay time, which would affect the range inversion). Preliminary denoising verification: The signal-to-noise ratio (SNR) and normalized mean square error (NMSE) of the preliminary denoised signal are calculated and compared with the original signal to verify the denoising effect. Typically, it is required that the SNR is improved and the NMSE is reduced after preliminary denoising. If the target is not met, return to step 5 to fine-tune the hyperparameter M and re-execute the screening and reconstruction to ensure that the preliminary denoised signal meets the input requirements for subsequent fine processing.
[0031] S5. Perform wavelet transform on the echo signal after preliminary denoising to obtain the final denoised signal.
[0032] In this embodiment, step S5 performs wavelet transform fine denoising on the initially denoised signal to output the final signal. This step utilizes the multi-scale analysis capability of WT to suppress residual high-frequency noise in the initially denoised signal (such as a small amount of shot noise not separated by TVMD and electronic device noise), achieving "fine denoising" and outputting a high-quality signal that can be directly used for parameter inversion. The specific process is as follows: Wavelet decomposition: Select a wavelet basis that is suitable for the lidar signal, and decompose the initial denoised signal into 3 to 5 layers to obtain the approximate coefficients (corresponding to the useful low-frequency components of the signal, with concentrated energy) and detail coefficients (corresponding to residual high-frequency noise, with dispersed energy) of each layer. Thresholding: Soft thresholding is applied to the detail coefficients of each layer (to avoid signal abrupt changes caused by hard thresholding). The threshold calculation adopts the Stein unbiased risk estimation method. Approximate coefficients are not thresholded to ensure that low-frequency useful signals are fully preserved. Signal reconstruction: Perform inverse wavelet transform on the processed approximation coefficients and detail coefficients to obtain the final denoised signal; after reconstruction, the signal needs to be normalized (map the amplitude range to the photon counting interval of the original signal to avoid numerical overflow affecting the subsequent lidar equation calculation). Final Verification and Output: Calculate the SNR, NMSE, and smoothness of the final signal (f(n) is the original signal, f'(n) is the final signal). According to the experimental standards in the document, the following must be met: compared with the original signal, the SNR is improved, the NMSE is reduced, and the smoothness is reduced. After verification, the final denoised signal is output. This signal can be directly substituted into the lidar equation to calculate the backscattering coefficient β(λ,R) and the extinction coefficient α(λ,R), providing accurate photon counting data support for atmospheric concentration inversion.
[0033] In summary, the FSTV-WT denoising method proposed in this invention has several significant advantages: It addresses the low efficiency and unstable decomposition accuracy problems caused by the reliance on manually selected parameters (mode number K, quadratic penalty factor α) in traditional variational mode decomposition (VMD). It achieves adaptive optimization of VMD parameters through an improved sparrow search algorithm (FSSA) based on a new fitness function, avoiding reliance on human experience and achieving faster convergence than the standard sparrow search algorithm (SSA), efficiently finding the optimal parameter combination. Furthermore, it addresses the difficulty of suppressing background noise and other noise in complex weather conditions using traditional algorithms such as wavelet transform (WT), VMD, and VMD-WT. The bottleneck of shot noise superposition interference is addressed by using Dynamic Peak Thresholding (TVMD) based on time-frequency analysis to accurately distinguish between signal and noise modes. Furthermore, multi-scale analysis by Wt.T. further suppresses residual noise. Experiments show that its signal-to-noise ratio (SNR) is significantly improved compared to VMD, and its normalized mean square error (NMSE) is reduced, demonstrating superior performance compared to traditional algorithms. Simultaneously, this algorithm exhibits stronger robustness and adaptability, stably preserving useful signals under varying noise intensities. This effectively improves the signal inversion accuracy and effective detection range of the 1550nm SNSPD lidar, providing reliable technical support for real-time applications such as long-term unattended environmental monitoring.
[0034] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.
[0035] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0036] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0037] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
[0038] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.
Claims
1. A lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD, characterized in that, Includes the following steps: Acquire the echo signal from the lidar; A fitness function is constructed by fusing dynamic spectral amplitude difference and signal smoothness. Based on the constructed fitness function, an improved sparrow search algorithm is used to iteratively optimize the echo signal to obtain the optimal variational mode decomposition parameters. Variational mode decomposition is performed on the echo signal based on the optimal variational mode decomposition parameters to obtain multiple mode components; A dynamic threshold mechanism is used to perform spectral peak segmentation on each modal component, select effective modal components, and reconstruct the signal to obtain the echo signal after preliminary denoising. Wavelet transform is performed on the echo signal after preliminary denoising to obtain the final denoised signal.
2. The lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, The fitness function is constructed by fusing dynamic spectral amplitude difference and signal smoothness as follows: D=|fs|+|ms| Where D is the fitness, f is the number of points in the spectrum of the noisy signal whose peak intensity exceeds the preset basic threshold, s is the theoretical signal characteristic value, and m is the number of noise points with random fluctuations in the time domain signal.
3. The lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, Based on the constructed fitness function, an improved sparrow search algorithm is used to iteratively optimize the echo signal to obtain the optimal variational mode decomposition parameters, including: Within the preset search range of variational mode decomposition parameters, a set number of parameter combinations are randomly generated as the initial sparrow population, with each parameter combination corresponding to one individual sparrow. Substitute each set of parameters into the VMD model to decompose the original echo signal, and then calculate the fitness value of each sparrow individual through the fitness function. Sort the current global best fitness and global worst fitness. Perform differentiated position updates based on the Sparrow role; The iteration terminates when the number of iterations reaches the maximum number of iterations, or when the change in the optimal fitness value after a set number of consecutive iterations is less than a preset threshold, and the optimal variational mode decomposition parameters corresponding to the current global optimal fitness are output.
4. The lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, Variational mode decomposition (VMD) of the echo signal is performed based on the optimal VMD parameters to obtain multiple mode components, including: Substitute the optimal variational mode decomposition parameters into the VMD constrained variational model; The VMD-constrained variational model was iteratively calculated using the alternating direction multiplier method to obtain multiple modal components; among them, the low-frequency IMF contains the effective signal, and the high-frequency IMF contains background noise and shot noise.
5. The lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, The use of a dynamic threshold mechanism for spectral peak segmentation of each modal component includes: Perform a fast Fourier transform on each modal component to obtain the corresponding amplitude-frequency spectrum; The position where the amplitude value is greater than the adjacent frequency points on the left and right is defined as the peak value of the spectrum. The spectrum of all modal components is traversed, the local peak intensity of each spectrum is counted, and the global peak intensity of the original echo signal spectrum is calculated. The local peak intensity of each spectrum is used as a dynamic threshold and compared with a set proportion of the global peak intensity. When the local peak intensity is less than the set proportion of the global peak intensity, the dynamic threshold is corrected to the set proportion of the global peak intensity.
6. The lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, Valid modal components include: The peak intensity of the spectrum of each modal component is compared with the dynamic threshold. If all local peak intensities of a modal component are less than the dynamic threshold, it is determined to be a pure noise mode and is removed. If at least one local peak intensity of a modal component is greater than the dynamic threshold, it is determined to be a dominant mode of effective signal and the energy proportion exceeding the dynamic threshold is calculated. If the energy proportion is greater than the set proportion threshold, the modal component is retained; otherwise, the modal component is removed.
7. The lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, Signal reconstruction is performed to obtain the preliminary denoised echo signal, including: The selected effective modal components are linearly superimposed in their original time-domain order to obtain the echo signal after preliminary denoising.
8. A lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, Calculate the signal-to-noise ratio and normalized mean square error of the echo signal after preliminary denoising, and compare the calculation results with the original echo signal to verify the denoising effect. If the denoising requirements are met, retain the echo signal after preliminary denoising; otherwise, readjust the dynamic threshold mechanism to perform spectral peak segmentation on each modal component.
9. A lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, The echo signal after preliminary denoising is subjected to wavelet transform to obtain the final denoised signal, including: By selecting a wavelet basis that is compatible with the lidar signal, the echo signal after preliminary denoising is decomposed into multiple layers to obtain the approximation coefficients and detail coefficients of each layer. A soft threshold is applied to the detail coefficients of each layer; The processed detail coefficients and approximation coefficients are subjected to inverse wavelet transform to obtain the final denoised signal.
10. A lidar signal processing method combining an improved sparrow search algorithm and dynamic threshold VMD according to claim 1, characterized in that, Calculate the signal-to-noise ratio, normalized mean square error, and smoothness of the final denoised signal, and compare the calculation results with the original echo signal to verify the denoising effect; if the denoising requirements are met, retain the final denoised signal. Otherwise, the dynamic threshold mechanism is readjusted to perform spectral peak segmentation on each modal component.