Adaptive unscented Kalman filter laser radar signal processing method and system

Abstract

The invention provides a self-adaptive unscented Kalman filter laser radar signal processing method and system, and belongs to the field of laser radar signal processing. The method comprises the following steps: constructing a laser radar detection physical model containing thick-tail distribution noise characteristics; performing initialization operation; inputting an original noisy signal sequence, and carrying out iterative filtering processing on the original noisy signal sequence by detecting distance units; in each iteration loop, dual adaptive unscented Kalman filtering is synchronously executed; for the processed observation data of all the detection distance units, outputting a final posteriori state vector sequence of all the detection distance units, and reconstructing to obtain a high-fidelity laser radar echo signal curve; and extracting the high-fidelity optical parameters, and performing inversion on the high-fidelity optical parameters to obtain an inversion result. According to the method, non-Gaussian characteristics of environmental noise can be accurately represented and adapted, high-fidelity extraction and reconstruction of weak echo signals under the background of strong interference and strong attenuation are realized, and the accuracy and robustness of laser radar detection are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of laser radar signal processing, and particularly relates to a self-adaptive unscented Kalman filter laser radar signal processing method and system. BACKGROUND

[0002] As an active optical remote sensing technology, laser radar has been widely used in active remote sensing detection, target tracking and dynamic system monitoring, etc. due to its high spatial and temporal resolution, high precision and other advantages. In these applications, accurately extracting effective information from the weak echo signal returned from the receiver is the key to determining the performance of the entire detection system.

[0003] The prior art mainly has the following several insurmountable bottlenecks when processing such signals: First, the failure of the traditional noise model. Most classic signal filtering algorithms, such as moving average filtering, Wiener filtering and standard Kalman filtering, are based on the basic assumption that the system noise follows a Gaussian distribution. However, strong background interference has significant non-Gaussian and non-stationary characteristics, often manifesting as strong pulses with large amplitude and randomly appearing, resulting in a large number of "wild values" in the measurement data. The filter based on the Gaussian assumption cannot effectively describe the noise distribution with "heavy tail" characteristics, and when processing these "wild values", it often incorrectly counts them as effective signals, resulting in distorted filtering results and even completely destroying the effective information of the original signal.

[0004] Second, the problem of high nonlinearity of the laser radar equation. The laser radar equation describing the propagation of laser in the atmosphere or water is a strong nonlinear equation containing an exponential decay term and a complex scattering term. Traditional linear filtering methods cannot be applied. To solve the nonlinear problem, the extended Kalman filter (EKF) is introduced, but EKF performs first-order Taylor expansion on the nonlinear function through the Jacobian matrix to achieve linearization approximation, which itself will introduce a large truncation error, and the accuracy is limited in the strong nonlinear scene.

[0005] Third, the limitations of existing nonlinear filtering methods. Unscented Kalman filter (UKF) uses unscented transformation and Sigma point sampling strategy, without explicit linearization, and has higher approximation accuracy for the statistical characteristics of nonlinear systems, which is a better choice for processing laser radar equations. However, the standard UKF algorithm still assumes that the process noise and measurement noise follow a Gaussian distribution. Therefore, although the standard UKF can handle nonlinear problems well, its performance will also decline sharply when faced with the above non-Gaussian strong interference with heavy tail characteristics, and the accuracy and stability of the filtering result cannot be guaranteed, and even the filter may diverge.

[0006] In summary, the prior art lacks a signal processing method that can effectively deal with the "strong nonlinearity" of the laser radar equation and the "non-Gaussian thick tail" of the real harsh environment noise. SUMMARY

[0007] In view of the above problems in the prior art, the present application provides an adaptive unscented Kalman filtering laser radar signal processing method and system, which solves the problem that the laser radar signal processing method in the prior art is mostly based on Gaussian noise assumption, and is difficult to effectively process strong non-Gaussian and non-stationary background noise, resulting in extremely low signal-to-noise ratio of echo signal, distortion of effective signal, and insufficient precision of physical parameter inversion.

[0008] To achieve the above purpose, the technical scheme adopted by the present application is as follows: an adaptive unscented Kalman filtering laser radar signal processing method, comprising the following steps: S1, constructing a laser radar detection physical model containing thick-tailed distribution noise characteristics; S2, initializing the state vector and covariance matrix based on the laser radar detection physical model; S3, based on the initialization result, inputting the original noisy signal sequence, and performing iterative filtering processing of the original noisy signal in the detection distance unit; S4, in each iteration loop of S3, simultaneously performing double adaptive unscented Kalman filtering until all observation data of the detection distance unit are processed, outputting the posterior state vector sequence of all detection distance units, and reconstructing the high-fidelity laser radar echo signal curve based on the posterior state vector sequence; S5, extracting high-fidelity optical parameters from the high-fidelity laser radar echo signal curve, and inverting the high-fidelity optical parameters to obtain the inversion result, thereby completing the processing of the laser radar signal.

[0009] Further, the laser radar detection physical model is constructed as follows: Construct a state equation for describing the dynamic evolution of physical parameters with time and distance; Based on the state equation, an observation equation based on the laser radar equation is constructed; Based on the state equation and the observation equation, the construction of the laser radar detection physical model is completed.

[0010] Further, the expression of the observation equation is as follows: ; ; Wherein, represents the laser radar measurement value, represents the measurement function, represents the observation noise, denotes a detection range, denotes a system constant, denotes a backscattering coefficient, denotes a lidar wavelength ， denotes an extinction coefficient, s denotes an integral variable, denotes k a state vector at time t, denotes k a state vector at time t-1, denotes k a control vector at time t, denotes a thick-tailed process noise, denotes a state transition function, d denotes a differential symbol.

[0011] Further, the thick-tailed process noise and the observation noise are both modeled by a thick-tailed probability distribution, both subject to a T distribution, and the expression of the probability density function of the T distribution is as follows: ; wherein, denotes the probability density function of the T distribution, denotes a Gamma function, denotes a degree of freedom parameter, denotes t a random variable of the T distribution.

[0012] Further, the S3 is specifically: based on the initialization result, inputting an original noisy signal sequence; based on the original noisy signal sequence, performing state estimation on the original noisy signal by using adaptive unscented Kalman filtering, wherein, for the state vector , for the macro parameter layer, a sliding window strategy is adopted for estimation, and for the micro parameter layer, high-frequency real-time tracking estimation is performed, and iterative filtering processing of the detection range unit is completed.

[0013] Further, the S4 is specifically: S401, in each iteration cycle of S3, synchronously performing adaptive unscented Kalman filtering time update, and predicting a prior state of a current detection range unit; S402, based on the prediction result, estimating and updating a degree of freedom parameter in a lidar detection physical model online; S403. Real-time tracking of the observation residuals of the state estimation, and dynamic adjustment of the unscented transformation scaling factor for the next iteration cycle based on the covariance of the observation residuals. Specifically, when the laser beam passes through the region with the largest parameter gradient change, the distribution range of the Sigma points is automatically shrunk; in relatively uniform regions inside the fog, the sampling range is expanded. S404, Update the degree-of-freedom parameters As a key factor, the dynamic scaling factor is calculated, and an adaptive unscented Kalman filter measurement update is performed. The measurement noise is adjusted using the dynamic scaling factor, and the posterior state is obtained by combining the current actual observation value of the lidar. S405. Determine whether all detection range units have been processed. If yes, output the posterior state vector sequence of all detection range units and proceed to S406. Otherwise, return to S401. S406. Based on the posterior state vector sequence, the high-fidelity lidar echo signal curve is reconstructed.

[0014] Furthermore, the expression for the dynamic adjustment of the covariance is as follows: ; in, Indicates in k At time t, the unscented transform scaling factor adjusted by adaptive unscented Kalman filtering. Let represent the initial value of the unscented transformation κ. Indicates in k The current observation value of the lidar at any time. Indicates in k The time step is the predicted observation value calculated by the filter based on the state at the previous time step. This represents the variance of the observation noise.

[0015] The present invention also provides an adaptive unscented Kalman filter lidar signal processing system, comprising: The first processing module is used to construct a physical model for lidar detection that includes the characteristics of thick-tailed distributed noise. The second processing module is used to initialize the state vector and covariance matrix based on the physical model of lidar detection. The third processing module is used to input the original noisy signal sequence based on the initialization result and perform iterative filtering processing on the original noisy signal unit by detection range. The fourth processing module is used to synchronously execute dual adaptive unscented Kalman filtering in each iteration loop of the third processing module until the observation data of all detection range units are processed, output the final posterior state vector sequence of all detection range units, and reconstruct the high-fidelity lidar echo signal curve based on the final posterior state vector sequence. The fifth processing module is configured to extract high-fidelity optical parameters from the high-fidelity laser radar echo signal curve, and perform inversion on the high-fidelity optical parameters to obtain an inversion result, thereby completing processing of the laser radar signal.

[0016] The present application has the following advantages: 1. The laser radar signal processing method based on the thick-tailed distribution can accurately characterize and adapt to the non-Gaussian characteristics of the environmental noise, realize high-fidelity extraction and reconstruction of weak echo signals in a strong interference and strong attenuation background, and significantly improve the accuracy and robustness of laser radar detection.

[0017] 2. The signal reconstruction accuracy is high, and the fidelity is good. The thick-tailed T distribution is used to accurately model strong pulse and non-Gaussian noise, which fundamentally overcomes the defect that the traditional Gaussian noise assumption does not conform to the physical reality. This makes the algorithm able to effectively distinguish between weak real signals and extreme noise "wild values", thereby realizing high-fidelity reconstruction of the original echo signal and providing a high-quality data basis for accurate inversion of subsequent physical parameters.

[0018] 3. The algorithm is robust and has good environmental adaptability. The dual adaptive adjustment mechanism of the freedom parameter and the filter scaling factor is designed, so that the filter can adjust its parameters online according to real-time data, and intelligently adapt to dynamically changing noise environment and signal characteristics. Whether in a relatively uniform fog area or in a fog boundary with dramatic parameter changes, the algorithm can remain stable and efficient, significantly improving the robustness of the system in complex and variable conditions.

[0019] 4. The strong nonlinearity is effectively processed, and the physical model application is more reliable. For the high nonlinearity of the laser radar equation in thick fog, the UKF and the layered filtering architecture used in the present application avoid the linearization error of traditional methods and can more accurately process the complex propagation behavior of light signals in strong scattering media. This ensures the reliable application of the physical model in the filtering process, ultimately improving the accuracy of the inverted physical parameters. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The method flowchart of the present application.

[0021] Figure 2 The system structure schematic diagram of the present application. DETAILED DESCRIPTION

[0022] The following describes a specific embodiment of the present application so that those skilled in the art can understand the present application, but it should be clear that the present application is not limited to the scope of the specific embodiment, and that all applications that utilize the concept of the present application are within the scope of the present application as defined and determined by the appended claims, and that all such variations are obvious to those skilled in the art.

[0023] Embodiment 1 The basic idea of the present application is to solve the two technical challenges of high nonlinearity of lidar signals and strong non-Gaussian noise by proposing a technical solution that deeply integrates physical models, advanced filtering theory, and adaptive noise modeling. To address the high nonlinearity of the lidar equation in strong attenuation media, the present application uses the unscented Kalman filter (UKF) algorithm, which accurately transfers the statistical properties of the nonlinear system through its Sigma sampling strategy, thereby avoiding the large errors introduced by the traditional extended Kalman filter (EKF) due to linearization. More core, to solve the problem of strong pulse interference, the present application innovatively discards the traditional Gaussian noise assumption and introduces a heavy-tailed probability distribution (specifically, T distribution) that can flexibly represent extreme "wild values" to mathematically model the system and observation noise. In addition, the present application also designs a set of double adaptive adjustment mechanism, through online estimation of the degree of freedom parameter of T distribution and dynamic adjustment of the scaling factor of UKF, so that the whole filtering process can intelligently adapt to the dynamic changes of environmental noise and signal characteristics. Through the above scheme, the present application aims to reconstruct the weak effective echo signal from the original data full of strong interference with high fidelity, laying a solid foundation for subsequent accurate inversion.

[0024] As Figure 1 shown, the present application provides an adaptive unscented Kalman filter lidar signal processing method, the implementation method of which is as follows: S1, a lidar detection physical model containing heavy-tailed distribution noise characteristics is constructed, which is specifically: A state equation is constructed to describe the dynamic evolution of physical parameters over time and distance; based on the state equation, an observation equation based on the lidar equation is constructed; based on the state equation and the observation equation, the construction of the lidar detection physical model is completed.

[0025] In this embodiment, this step aims to establish a nonlinear state space model (i.e. lidar detection physical model) that can accurately describe the lidar detection process. First, a state equation is constructed to describe the dynamic evolution of physical parameters over time and distance: wherein, x represents a state vector, which includes heavy-tailed noise parameters, extinction coefficient, liquid water content, backscattering coefficient, etc.,​ denotes the thick-tailed process noise, subject to T-distribution, covariance matrix is dynamically updated by a multi-scale strategy.

[0026] Secondly, the observation equation based on the laser radar equation is established: ; where, denotes the laser radar measurement (usually photon counting, photon time, etc.), denotes the measurement function, denotes the observation noise, denotes the detection distance, denotes the system constant, denotes the backscattering coefficient, denotes the laser radar wavelength ， denotes the extinction coefficient, s denotes the integral variable (0 to any point on the path of the detection distance r ), denotes k the state vector at time t, denotes k the state vector at time t-1, denotes k the control vector at time t, denotes the thick-tailed process noise, denotes the state transition function, d denotes the differential symbol.

[0027] In the formula, contains photon counting, photon time, etc., and the covariance matrix is determined by the joint of photon counting Poisson statistics and detector noise. Similarly, thick-tailed noise modeling is adopted.

[0028] In this embodiment, thick-tailed process noise and observation noise are both modeled by thick-tailed probability distribution, specifically T-distribution. T-distribution can flexibly capture the thick-tailed characteristics of noise by introducing a degree of freedom parameter , and its probability density function can be expressed as: ; where, denotes the probability density function of T-distribution, denotes the Gamma function, denotes the degree of freedom parameter, denotes t a random variable of T-distribution.

[0029] S2, based on the laser radar detection physical model, the state vector and the covariance matrix are initialized; In this embodiment, before the filtering starts, the state vector and the related covariance matrix are initialized. The initial value can be set according to historical experience data, weather model prediction value or a reasonable default value.

[0030] S3, based on the initialization result, input the original noisy signal sequence, and perform iterative filtering processing on the original noisy signal in the detection distance unit, which is specifically: Based on the initialization result, input the original noisy signal sequence; based on the original noisy signal sequence, use adaptive unscented Kalman filtering to estimate the state of the original noisy signal, wherein, for the state vector , for the macro parameter layer, a sliding window strategy is used for estimation, and for the micro parameter layer, high-frequency real-time tracking estimation is performed, to complete the iterative filtering processing of the detection distance unit.

[0031] In this embodiment, the hierarchical unscented Kalman filter iterative calculation is as follows: Since the laser radar equation has high nonlinearity, this embodiment uses unscented Kalman filtering for state estimation to avoid the linearization error of traditional extended Kalman filtering. In order to cope with the characteristics of different parameter scales, this step uses a hierarchical filtering architecture. For different characteristic parameters in the state vector , different strategies are used for processing: macro parameter layer: for slowly changing parameters (such as average laser radar ratio), a sliding window smoothing strategy is used for robust estimation. Micro parameter layer: for parameters that change quickly (such as distance-by-distance extinction / backscattering coefficient profile), high-frequency real-time tracking estimation is performed.

[0032] S4, in each iteration cycle of S3, a double adaptive unscented Kalman filter is executed synchronously until all observation data of the detection distance unit are processed, and the posterior state vector sequence of all detection distance units is output, and based on the posterior state vector sequence, a high-fidelity laser radar echo signal curve is reconstructed, and the implementation method is as follows: S401, in each iteration cycle of S3, an adaptive unscented Kalman filter time update is executed synchronously to predict the prior state of the current detection distance unit; S402, based on the prediction result, the degree of freedom parameter in the laser radar detection physical model is estimated and updated online. In this embodiment, the degree of freedom parameter obtained by S402 online estimation and update is applied to the S404 measurement update step in the current iteration period. The specific application mode is: the updated degree of freedom parameter As a key factor, a dynamic scaling factor is used to calculate the measurement noise R in the current S404 step to handle outliers, and this is used to adjust the measurement noise covariance matrix at that moment. When the estimation of S402 indicates the presence of strong impulse noise (i.e., the degree of freedom parameters...) (Reduce), measure the noise covariance matrix This will be temporarily increased, thereby reducing the Kalman gain.

[0033] S403. Real-time tracking of the observation residuals of the state estimation, and dynamic adjustment of the unscented transformation scaling factor for the next iteration cycle based on the covariance of the observation residuals. Specifically, when the laser beam passes through the region with the largest parameter gradient change, the distribution range of the Sigma points is automatically shrunk; in relatively uniform regions inside the fog, the sampling range is expanded. In this embodiment, this step adjusts the scaling factor based on the observed residuals. It will be used in the next iteration cycle (i.e. k In the "Time Update (Prediction)" step (i.e., S401 of the next iteration) at time +1, its specific function is to control the generation of the unscented transformation sigma point Sigma in this step. When the adjustment result of S403 indicates that the current state may be in a region of drastic parameter changes (scaling factor)... (If the sigma is reduced), the sigma points in the next cycle will be more tightly distributed around the state mean to improve the approximation accuracy for local strong nonlinearities; conversely, in uniform regions (if the sigma points are increased), the next set of sigma points will be more widely distributed to enhance the stability of the filter.

[0034] S404, Update the degree-of-freedom parameters As a key factor, the dynamic scaling factor is calculated, and an adaptive unscented Kalman filter measurement update is performed. The measurement noise is adjusted using the dynamic scaling factor, and the posterior state is obtained by combining the current actual observation value of the lidar. S405. Determine whether all detection range units have been processed. If yes, output the posterior state vector sequence of all detection range units and proceed to S406. Otherwise, return to S401. S406. Based on the posterior state vector sequence, the high-fidelity lidar echo signal curve is reconstructed.

[0035] In this embodiment, dual adaptive adjustment is performed during the iteration process, as follows: To enable the filtering algorithm to intelligently adapt to complex dynamic environments, two adaptive adjustment mechanisms are executed simultaneously in each iteration of S3: the noise model adaptively employs a maximum likelihood estimation algorithm to estimate and update the degree of freedom parameters of the T-distribution noise model online based on the latest observation data. When strong impulse noise is detected, the degree of freedom parameter value is automatically reduced, enhancing the model's tolerance for extreme "wild values"; when the environmental noise approaches a Gaussian distribution, the degree of freedom parameter value is increased, optimizing the filtering performance.

[0036] Filtering process adaptation (scaling factor adjustment): Real-time tracking of the observation residuals of state estimation, and dynamic adjustment of the scaling factor of the unscented transformation according to the covariance of the residuals When the laser beam passes through a region with a dramatic change in the parameter gradient, the Sigma distribution range is automatically contracted to improve the approximation accuracy of the local nonlinear process; in the relatively uniform region inside the fog body, the sampling range is expanded to enhance stability. Its dynamic adjustment can be expressed as: ; where, represents the unscented transformation scaling factor adjusted by the adaptive unscented Kalman filter at k time, represents the initial value of the unscented transformation κ, represents the current observation value of the lidar at k time, represents the predicted observation value predicted by the filter based on the state at the last time at k time, represents the observation noise variance.

[0037] In this embodiment, S3 and S4 are executed in a loop until all the observation data of the detection distance units are processed. The final output posterior state vector sequence , which contains the atmospheric optical parameters after accurate filtering. Based on this sequence, a high signal-to-noise ratio and high fidelity lidar echo signal curve can be reconstructed without distortion.

[0038] S5, extract high-fidelity optical parameters from the high-fidelity lidar echo signal curve, and perform inversion on the high-fidelity optical parameters to obtain an inversion result, completing the processing of the lidar signal.

[0039] In this embodiment, the high-fidelity optical parameters (such as extinction coefficient and backscattering coefficient profiles) obtained by S5, which have eliminated strong noise interference, are used to reconstruct a high-fidelity lidar echo signal curve through mature physical or statistical inversion algorithms (such as the klett method or the fernald method).

[0040] Embodiment 2 As Figure 2As shown, the present application provides an adaptive unscented Kalman filter lidar signal processing system for performing the adaptive unscented Kalman filter lidar signal processing according to any one of embodiments 1, comprising: A first processing module for constructing a lidar detection physical model containing the noise characteristics of a thick-tailed distribution; A second processing module for initializing the state vector and covariance matrix based on the lidar detection physical model; A third processing module for inputting the original noisy signal sequence based on the initialization result, and performing iterative filtering processing of the original noisy signal for each detection distance unit; A fourth processing module for synchronously performing double adaptive unscented Kalman filtering in each iteration cycle of the third processing module until all observation data of the detection distance unit is processed, outputting the final posterior state vector sequence of all detection distance units, and reconstructing a high-fidelity lidar echo signal curve based on the final posterior state vector sequence; A fifth processing module for extracting high-fidelity optical parameters from the high-fidelity lidar echo signal curve, and inverting the high-fidelity optical parameters to obtain an inversion result, thereby completing the processing of the lidar signal.

[0041] The adaptive unscented Kalman filter lidar signal processing system provided by the present application can perform the adaptive unscented Kalman filter lidar signal processing method described in method embodiment 1, and has similar implementation principles and beneficial effects, which will not be repeated here.

[0042] In this embodiment, the present application can divide the functional units according to the adaptive unscented Kalman filter lidar signal processing method, for example, each function can be divided into a functional unit, or two or more functions can be integrated into a processing unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit. It should be noted that the division of units in the present application is illustrative, and is only a logical division. Actual implementation can have another division method.

[0043] In order to realize the principles and beneficial effects of the method, the adaptive unscented Kalman filter lidar signal processing system in the embodiment comprises hardware structures and / or software modules corresponding to the functions. Those skilled in the art should easily realize that, in combination with the schematic units and algorithm steps described in the embodiments disclosed in the present application, the present application can be realized in the form of hardware and / or a combination of hardware and computer software. Whether a certain function is realized in the form of hardware or computer software is driven depends on the specific application and design constraints of the technical solution, and different methods can be used to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

Claims

1. An adaptive unscented Kalman filter lidar signal processing method, characterized in that, Includes the following steps: S1. Construct a physical model for lidar detection that includes the characteristics of thick-tailed distributed noise; S2. Based on the physical model of lidar detection, initialize the state vector and covariance matrix; S3. Based on the initialization results, input the original noisy signal sequence and perform iterative filtering processing on the original noisy signal using the detection range unit; S4. In each iteration of S3, dual adaptive unscented Kalman filtering is executed synchronously until all the observation data of all detection range units are processed, the posterior state vector sequence of all detection range units is output, and the high-fidelity lidar echo signal curve is reconstructed based on the posterior state vector sequence. S5. Extract high-fidelity optical parameters from the high-fidelity lidar echo signal curve, and invert the high-fidelity optical parameters to obtain the inversion result, thus completing the processing of the lidar signal.

2. The adaptive unscented Kalman filter lidar signal processing method according to claim 1, characterized in that, The physical model for lidar detection is constructed as follows: Construct state equations to describe the dynamic evolution of physical parameters with time and distance; Based on the state equation, an observation equation based on the lidar equation is constructed; Based on the state equation and observation equation, the physical model of lidar detection was constructed.

3. The adaptive unscented Kalman filter lidar signal processing method according to claim 2, characterized in that, The expression for the observation equation is as follows: ； ； in, This represents the measurement value from the lidar. Represents the measurement function. Indicates observation noise. Indicates the detection distance. Represents system constants. Represents the backscattering coefficient. Indicates the wavelength of lidar ， Indicates the extinction coefficient. s Represents the integral variable. express k The state vector at time t, express k The state vector at time -1 express k Control vector at time, This indicates noise in the thick-tailed process. Represents the state transition function. d Represents the differential symbol.

4. The adaptive unscented Kalman filter lidar signal processing method according to claim 3, characterized in that, Thick-tail process noise and observation noise All models are based on a heavy-tailed probability distribution and follow a T-distribution. The expression for the probability density function of the T-distribution is as follows: ； in, This represents the probability density function of the T-distribution. Represents the Gamma function. Represents the degree of freedom parameter. express t A random variable with a distribution.

5. The adaptive unscented Kalman filter lidar signal processing method according to claim 1, characterized in that, Specifically, S3 is: Based on the initialization results, input the original noisy signal sequence; Based on the original noisy signal sequence, an adaptive unscented Kalman filter is used to estimate the state of the original noisy signal, whereby the state vector is... For the macroscopic parameter layer, a sliding window strategy is used for estimation, while for the microscopic parameter layer, high-frequency real-time tracking estimation is performed to complete the iterative filtering processing of the detection range unit.

6. The adaptive unscented Kalman filter lidar signal processing method according to claim 4, characterized in that, Specifically, S4 is: S401. In each iteration of S3, the adaptive unscented Kalman filter time update is executed synchronously to predict the prior state of the current detection range cell. S402. Based on the prediction results, estimate and update the degree-of-freedom parameters in the physical model of lidar detection online. ; S403. Real-time tracking of the observation residuals of the state estimation, and dynamic adjustment of the unscented transformation scaling factor for the next iteration cycle based on the covariance of the observation residuals. Specifically, when the laser beam passes through the region with the largest parameter gradient change, the distribution range of the Sigma points is automatically shrunk; in relatively uniform regions inside the fog, the sampling range is expanded. S404, Update the degree-of-freedom parameters As a key factor, the dynamic scaling factor is calculated, and an adaptive unscented Kalman filter measurement update is performed. The measurement noise is adjusted using the dynamic scaling factor, and the posterior state is obtained by combining the current actual observation value of the lidar. S405. Determine whether all detection range units have been processed. If yes, output the posterior state vector sequence of all detection range units and proceed to S406. Otherwise, return to S401. S406. Based on the posterior state vector sequence, the high-fidelity lidar echo signal curve is reconstructed.

7. The adaptive unscented Kalman filter lidar signal processing method according to claim 6, characterized in that, The expression for the dynamic adjustment of covariance is as follows: ； in, Indicates in k At time t, the unscented transform scaling factor adjusted by adaptive unscented Kalman filtering. Let represent the initial value of the unscented transformation κ. Indicates in k The current observation value of the lidar at any time. Indicates in k The time step is the predicted observation value calculated by the filter based on the state at the previous time step. This represents the variance of the observation noise.

8. An adaptive unscented Kalman filter lidar signal processing system, used to perform the adaptive unscented Kalman filter lidar signal processing according to any one of claims 1-7, characterized in that, include: The first processing module is used to construct a physical model for lidar detection that includes the characteristics of thick-tailed distributed noise. The second processing module is used to initialize the state vector and covariance matrix based on the physical model of lidar detection. The third processing module is used to input the original noisy signal sequence based on the initialization result and perform iterative filtering processing on the original noisy signal unit by detection range. The fourth processing module is used to synchronously execute dual adaptive unscented Kalman filtering in each iteration loop of the third processing module until the observation data of all detection range units are processed, output the final posterior state vector sequence of all detection range units, and reconstruct the high-fidelity lidar echo signal curve based on the final posterior state vector sequence. The fifth processing module is used to extract high-fidelity optical parameters from the high-fidelity lidar echo signal curve, and to invert the high-fidelity optical parameters to obtain the inversion result, thus completing the processing of the lidar signal.

Images