Signal optimization method, device and computer equipment for polarized light compass

Abstract

The application relates to a signal optimization method, device and computer equipment of a polarized light compass. The method comprises the following steps: acquiring an original heading angle signal collected by a polarized light compass in a driving device, and extracting a heading global feature from the original heading angle signal; wherein the heading global feature comprises a steady-state feature reflecting long-term smooth changes of the heading of the driving device and a transient feature reflecting environmental interference and working condition mutation; performing differential domain analysis on the heading global feature to obtain a monitoring statistical quantity; determining noise posterior probabilities corresponding to different noise working conditions according to the monitoring statistical quantity and a pre-constructed noise prior model; performing state division on the original heading angle signal according to the noise posterior probabilities corresponding to the different noise working conditions to obtain heading state data with different noise contents; and optimizing the original heading angle signal according to the heading state data with different noise contents. The method can improve the directional accuracy of the polarized light compass.

Description

Technical Field

[0001] This application relates to the field of polarized light navigation technology, and in particular to a signal optimization method, apparatus and computer equipment for a polarized light compass. Background Technology

[0002] Polarization navigation systems are fully autonomous navigation systems that rely on polarized light from the sky to calculate the heading information of a carrier. These systems require no external information and do not radiate energy, making them widely used in aerospace, aviation, maritime, and other military fields. The polarization compass is a key component of polarization navigation systems; its measurement accuracy directly determines the performance of the entire system. Therefore, improving the measurement accuracy of the polarization compass has always been a key research focus in the field of polarization navigation.

[0003] Existing methods for improving the orientation accuracy of polarization compasses in harsh environments primarily focus on processing polarization angle images acquired by polarization imaging systems. These methods can be divided into two categories: one involves smoothing the pixels of the polarization angle image, a method limited by the size of the local smoothing window; the other, developed with the advancements in machine learning, especially deep learning, employs neural networks. However, neural network training requires vast amounts of data, and the atmospheric polarization angle images used for polarization navigation are constantly changing, causing the sample data to fluctuate in real time. This results in weak generalization ability of machine learning, ultimately limiting the effective improvement of polarization compass orientation accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide a signal optimization method, apparatus, and computer equipment for polarized optical compasses that can improve the orientation accuracy of polarized optical compasses, addressing the aforementioned technical problems.

[0005] In a first aspect, this application provides a signal optimization method for a polarization compass, including:

[0006] The original heading angle signal collected by the polarization compass in the driving equipment is acquired, and the global heading feature is extracted from the original heading angle signal; wherein, the global heading feature includes steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions;

[0007] Differential domain analysis is performed on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics for characterizing the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics for characterizing the deviation of the heading characteristics of the driving equipment under steady operating conditions relative to the noise-free state.

[0008] Based on the monitoring statistics and the pre-built noise prior model, the noise posterior probability corresponding to different noise conditions is determined.

[0009] Based on the noise posterior probability corresponding to different noise conditions, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

[0010] The original heading angle signal is optimized based on heading state data with different noise levels.

[0011] Secondly, this application also provides a signal optimization device for a polarization compass, comprising:

[0012] The feature extraction module is used to acquire the original heading angle signal collected by the polarization compass in the driving equipment, and extract the heading global features from the original heading angle signal; wherein, the heading global features include steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions;

[0013] The statistical analysis module is used to perform differential domain analysis on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics used to characterize the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics used to characterize the deviation of the heading characteristics of the driving equipment under stable operating conditions relative to the noise-free state.

[0014] The probability determination module is used to determine the posterior probability of noise corresponding to different noise conditions based on the monitoring statistics and the pre-built noise prior model.

[0015] The data determination module is used to divide the original heading angle signal into states according to the noise posterior probability corresponding to different noise conditions, so as to obtain heading state data with different noise contents.

[0016] The signal optimization module is used to optimize the original heading angle signal based on heading state data with different noise contents.

[0017] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0018] The original heading angle signal collected by the polarization compass in the driving equipment is acquired, and the global heading feature is extracted from the original heading angle signal; wherein, the global heading feature includes steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions;

[0019] Differential domain analysis is performed on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics for characterizing the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics for characterizing the deviation of the heading characteristics of the driving equipment under steady operating conditions relative to the noise-free state.

[0020] Based on the monitoring statistics and the pre-built noise prior model, the noise posterior probability corresponding to different noise conditions is determined.

[0021] Based on the noise posterior probability corresponding to different noise conditions, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

[0022] The original heading angle signal is optimized based on heading state data with different noise levels.

[0023] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0024] The original heading angle signal collected by the polarization compass in the driving equipment is acquired, and the global heading feature is extracted from the original heading angle signal; wherein, the global heading feature includes steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions;

[0025] Differential domain analysis is performed on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics for characterizing the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics for characterizing the deviation of the heading characteristics of the driving equipment under steady operating conditions relative to the noise-free state.

[0026] Based on the monitoring statistics and the pre-built noise prior model, the noise posterior probability corresponding to different noise conditions is determined.

[0027] Based on the noise posterior probability corresponding to different noise conditions, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

[0028] The original heading angle signal is optimized based on heading state data with different noise levels.

[0029] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0030] The original heading angle signal collected by the polarization compass in the driving equipment is acquired, and the global heading feature is extracted from the original heading angle signal; wherein, the global heading feature includes steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions;

[0031] Differential domain analysis is performed on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics for characterizing the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics for characterizing the deviation of the heading characteristics of the driving equipment under steady operating conditions relative to the noise-free state.

[0032] Based on the monitoring statistics and the pre-built noise prior model, the noise posterior probability corresponding to different noise conditions is determined.

[0033] Based on the noise posterior probability corresponding to different noise conditions, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

[0034] The original heading angle signal is optimized based on heading state data with different noise levels.

[0035] The aforementioned signal optimization method, apparatus, and computer equipment for polarized optical compasses achieve full-domain, comprehensive characterization of signal variation patterns by separating steady-state and transient features of heading. By constructing static and dynamic dual-monitoring statistics through differential domain analysis, long-term steady-state offsets and instantaneous environmental disturbances can be accurately captured simultaneously. The combination of a noise prior model and probabilistic inference enables quantitative identification of noise conditions, overcoming the technical bottleneck of traditional methods' inability to finely distinguish heading states with different noise levels. Finally, hierarchical adaptive signal optimization is implemented based on heading state differences, maximizing the preservation of effective details of the true heading while efficiently suppressing inherent device noise and complex interference noise introduced by harsh environments such as dust, aerosols, and fog. Furthermore, this application requires no massive training samples, exhibits strong generalization ability, high real-time performance, and strong environmental adaptability, significantly improving the heading measurement accuracy and output stability of polarized optical compasses under complex and harsh conditions. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating a signal optimization method for a polarization compass in one embodiment.

[0038] Figure 2 This is a flowchart illustrating the process of extracting global heading features in one embodiment;

[0039] Figure 3 This is a flowchart illustrating the process of extracting global heading features in another embodiment;

[0040] Figure 4 This is a flowchart illustrating the process of determining monitoring statistics in one embodiment;

[0041] Figure 5 This is a flowchart illustrating the process of determining monitoring statistics in another embodiment;

[0042] Figure 6 This is a flowchart illustrating the process of determining the posterior probability of noise corresponding to different noise conditions in one embodiment.

[0043] Figure 7 This is a flowchart illustrating the process of determining heading status data with different noise levels in one embodiment.

[0044] Figure 8 This is a flowchart illustrating the optimization of the original heading angle signal in one embodiment;

[0045] Figure 9 This is a flowchart illustrating the signal optimization method for a polarization compass in another embodiment;

[0046] Figure 10 This is a structural block diagram of a signal optimization device for a polarization compass in one embodiment;

[0047] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] The signal optimization method for polarized optical compasses provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown is described. The signal optimization method for the polarization compass provided in this embodiment is executed by a computer device, which can be a server or a terminal device with powerful computing capabilities.

[0050] In one exemplary embodiment, such as Figure 1 As shown, a signal optimization method for a polarization compass is provided. Taking the application of this method to a server as an example, the method includes the following steps.

[0051] S101, acquire the original heading angle signal collected by the polarized light compass in the driving equipment, and extract the global heading feature from the original heading angle signal.

[0052] The driving equipment refers to the carrier equipment equipped with a polarized optical compass, including but not limited to vehicle-mounted, shipborne, airborne, and drone navigation carriers, which rely on the polarized optical compass to complete autonomous heading calculation. The polarized optical compass is a passive autonomous navigation sensing element mounted on the driving equipment. It calculates the carrier's heading angle based on the atmospheric polarization mode, and is not constrained by satellite or ground base station signals. However, it is susceptible to interference from sandstorms, fog, aerosols, and turbulent conditions, outputting a noisy original heading.

[0053] The original heading angle signal is a continuous time-series angle data sampled and output in real time by the polarized light compass. It is mixed with the actual heading changes of the carrier, the inherent temperature drift noise of the device, the noise of the harsh environment, and the sudden noise of the carrier's turbulence. It has not been noise-reduced and corrected.

[0054] The overall heading characteristics include steady-state characteristics reflecting the long-term stable changes in the heading of the driving equipment and transient characteristics reflecting environmental disturbances and sudden changes in operating conditions. Overall heading characteristics are a complete set of features extracted from the original heading angle signal, divided into steady-state and transient characteristics, fully covering the slow-changing and sudden disturbance patterns of heading. Steady-state characteristics are low-frequency features reflecting the long-term, gradual changes in the actual heading of the driving equipment, representing the true driving / navigation trend of the vehicle. The rate of change is slow and mainly affected by the slow drift of components. Transient characteristics are short-term, high-frequency fluctuations caused by severe environmental disturbances, sudden changes in vehicle operating conditions, and transient electromagnetic interference; most of these are noise components.

[0055] Optionally, time-series heading angle data can be continuously collected using a polarized optical compass at a fixed sampling frequency to construct an original heading time-series dataset. Furthermore, joint time-frequency domain feature modeling is performed on the original heading signal to uncover the inherent variation law of the signal. Through feature filtering and decomposition mechanisms, steady-state features and transient features are separated and combined into a global heading feature.

[0056] S102, perform differential domain analysis on the global heading characteristics to obtain monitoring statistics.

[0057] The differential domain analysis involves performing differential operations on the time-series data of the entire feature domain to uncover the temporal changes in features. Monitoring statistics include dynamic statistics characterizing the instantaneous fluctuations in the heading characteristics of the driving equipment, and static statistics characterizing the deviation of the heading characteristics of the driving equipment under stable operating conditions relative to a noise-free state. Static statistics quantify the deviation of the steady-state heading characteristics from an ideal noise-free benchmark under stable operating conditions, and are used to measure systematic baseline drift error. Dynamic statistics quantify the instantaneous fluctuations of transient characteristics, characterizing the strength of short-term random disturbances caused by sudden environmental changes and turbulent operating conditions.

[0058] Optionally, simple feature values ​​can only describe the current state and cannot reflect the dynamic changes in the signal. Therefore, the dimensionality of the entire heading feature can be split into dominant trend features and residual disturbance features; static statistics can be calculated based on the steady-state values ​​of the features to measure the baseline offset error under steady-state conditions; differential operations can be performed on the feature time-series data to obtain the instantaneous changes of the features; dynamic statistics can be constructed based on the changes to quantify the intensity of sudden disturbances, and finally a set of dynamic and static joint monitoring statistics can be formed.

[0059] In one alternative embodiment, differential methods may be used. The statistical modeling method performs time-series differencing on steady-state and transient characteristics, constructs a covariance matrix, and solves for standardized static and dynamic statistics, accurately adapting to the weak perturbation and strong abrupt change characteristics of polarization signals.

[0060] Another optional embodiment can also use the trend residual statistical method to fit the long-term trend line of the heading characteristics. The trend deviation is a static statistic, and the instantaneous fluctuation amplitude of the residual is a dynamic statistic, which is suitable for slow-moving vehicle scenarios.

[0061] S103. Based on the monitoring statistics and the pre-built noise prior model, determine the noise posterior probability corresponding to different noise conditions.

[0062] The noise prior model is a probabilistic model with pre-calibrated parameters based on measured data from multiple scenarios, storing the statistical distribution patterns under low, medium, and high noise conditions. The noise posterior probability is the real-time probability value of the original signal belonging to low, medium, and high noise conditions, respectively, under the current static and dynamic statistical observation conditions.

[0063] Optionally, noise disturbances corresponding to different harsh environments (dust storms, dense fog, strong light scattering) exhibit fixed statistical distribution patterns. Therefore, by using a pre-constructed noise prior model and combining it with real-time monitoring statistics, a probabilistic inference algorithm can be employed to quantify the confidence probability of the current signal being in low-noise, medium-noise, and high-noise conditions, thereby achieving quantitative discrimination of noise status. Specifically, the static and dynamic monitoring statistics obtained in real time are input into the noise prior model, and the model outputs the intra-class and inter-class probabilities corresponding to each noise condition. Combining the preset prior probabilities of the conditions, the posterior probability of the original heading signal corresponding to different noise conditions at each moment is obtained through iterative calculation using the probabilistic inference formula.

[0064] For example, based on the Bayesian criterion, four types of monitoring statistics can be integrated to calculate the posterior probability of multiple working conditions, which has high discrimination accuracy, strong robustness, and is adaptable to complex and harsh environments.

[0065] S104. Based on the noise posterior probability corresponding to different noise conditions, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

[0066] Among them, the heading status data is segmented heading data after being divided according to the posterior probability classification, and is divided into three categories according to the noise ratio: low noise effective data, medium noise mixed data, and high noise interference data.

[0067] Optionally, different posterior probabilities correspond to different signal-to-noise ratios. By using the maximum probability criterion, the original heading time-series signal is classified point by point and frame by frame, and the continuous original signal is divided into heading state data with multiple noise levels, thereby achieving fine differentiation of noise states and providing a basis for subsequent differentiated optimization.

[0068] S105 optimizes the original heading angle signal based on heading state data with different noise levels.

[0069] Optionally, the noise reduction requirements for heading signals with different noise levels vary significantly. Low-noise signals require detail preservation and weak noise reduction, while high-noise signals require strong interference suppression and preservation of valid information. Therefore, the corresponding optimization parameters and processing intensity can be adaptively matched for heading status data with different noise levels. Light fidelity filtering is applied to low-noise valid signals, moderate noise reduction optimization is applied to medium-noise mixed signals, and strong noise reduction and feature reconstruction are applied to high-noise interference signals. The segmented optimized signals are then spliced ​​and fused to output a globally optimized high-precision heading signal.

[0070] The aforementioned signal optimization method for polarized optical compasses achieves full-domain, comprehensive characterization of signal variation patterns by separating steady-state and transient features of the heading. It constructs static and dynamic dual-monitoring statistics through differential domain analysis, enabling simultaneous and accurate capture of long-term steady-state offsets and instantaneous environmental disturbances. Combining a noise prior model and probabilistic inference, it achieves quantitative identification of noise conditions, overcoming the technical bottleneck of traditional methods' inability to finely distinguish heading states with different noise levels. Finally, it implements hierarchical adaptive signal optimization based on heading state differences, maximizing the preservation of effective details of the true heading while efficiently suppressing inherent device noise and complex interference noise introduced by harsh environments such as dust, aerosols, and fog. Furthermore, this application requires no massive training samples, exhibits strong generalization ability, high real-time performance, and strong environmental adaptability, significantly improving the heading measurement accuracy and output stability of polarized optical compasses under complex and harsh conditions.

[0071] Optionally, in one embodiment, such as Figure 2 As shown, a method for extracting global heading features is provided, which specifically includes the following steps:

[0072] S201, Based on the original heading angle signal, construct the time-frequency domain statistics matrix corresponding to the original heading angle signal.

[0073] Among them, the time-frequency domain statistics matrix is ​​a multi-dimensional data matrix that integrates time-domain statistical features and frequency-domain statistical features. By sliding sampling the original heading signal, low-order and high-order statistics such as mean, variance, skewness, kurtosis, and fluctuation range are extracted to fully characterize the time-domain fluctuation characteristics and frequency-domain distribution characteristics of the signal.

[0074] In this embodiment, the original heading angle signal is a one-dimensional time-series signal, and single-dimensional data cannot fully distinguish between slow real changes and high-frequency noise interference. Therefore, by using a sliding window for segmented sampling, multi-dimensional statistical characteristics can be extracted from the time domain and frequency domain respectively, and a statistical matrix integrating time and frequency information can be constructed. This realizes the transformation of shallow observation data into a deep statistical feature space, providing high-dimensional input for subsequent feature decoupling modeling.

[0075] Specifically, a fixed-length sliding window can be set for the continuous raw heading angle time-series signal, and the signal can be segmented frame by frame. Further, for the time-domain signal within each window, statistical parameters such as mean, variance, skewness, kurtosis, and range are calculated to characterize the time-domain stationarity and fluctuation characteristics of the signal. Next, a Fourier transform is performed on the signal within the window to obtain the frequency domain sequence, and multi-dimensional statistics in the frequency domain are calculated to characterize the frequency distribution characteristics of the signal. Finally, the time-frequency domain statistical parameters of all windows are concatenated according to the time-series dimension to form a complete time-frequency domain statistical matrix.

[0076] S202, construct a linear transformation model based on the transpose of the time-frequency domain statistics matrix.

[0077] The linear transformation model characterizes the mapping relationship between the original heading angle signal and the slow feature space. This model is a matrix transformation model that maps the high-dimensional observation space of the original heading signal to the slow feature space. By fixing the linear mapping relationship, it achieves feature decoupling, dimensional compression, and extraction of intrinsic patterns from the mixed signal. The slow feature space is a feature space constructed with the speed of signal temporal change as a constraint. The dimension with the most gradual change in this space corresponds to the actual heading trend of the carrier, while the dimension with drastic change corresponds to environmental and operational noise interference.

[0078] Optionally, time-frequency domain statistical matrices suffer from dimensional coupling and information redundancy, making it impossible to directly distinguish between steady-state effective features and transient noise features. Therefore, the data dimension can be reconstructed by matrix transposition, establishing a fixed linear transformation mapping relationship to project high-dimensional statistical space data onto an independent slow feature space, thereby achieving decoupling and separation of mixed features.

[0079] Specifically, the constructed time-frequency domain statistical matrix can be transposed to reconstruct the data arrangement structure; the generalized eigenvalue equation can be solved based on the transposed matrix to train the optimal weight coefficient matrix; finally, a linear transformation model can be built through the weight matrix to establish a one-to-one mapping relationship between the time-frequency statistical space and the slow feature space, thereby realizing the linear projection of the original mixed signal onto the pure feature space.

[0080] For example, it can be adopted The method of high-order slow feature linear transformation, which solves the linear transformation matrix based on high-order statistical constraints, enhances the ability to extract weak perturbation and slow drift features and is suitable for the harsh environment of low signal-to-noise ratio polarization compass.

[0081] S203, based on the linear transformation model, determine the global heading characteristics.

[0082] Optionally, a linear transformation model can be invoked to spatially map the time-frequency domain statistical matrix to obtain multi-dimensional initial slow features. Based on preset constraints of zero mean, unit variance, and feature decorrelation, the target slow features with the smallest time-series change rate are selected to preserve the core inherent laws of the signal. Furthermore, the target slow features are dimensionally split, with the gently changing dominant components as steady-state features and the residual fluctuation components as transient features. The two types of features are merged to form a complete heading global feature.

[0083] In this embodiment, by constructing a time-frequency domain statistical matrix, the time-domain stationary characteristics and frequency-domain distribution characteristics of the original heading signal are comprehensively explored. Then, through matrix transpose and linear transformation model construction, the high-dimensional redundant data is decoupled and dimensionality reduced, and the accurate mapping relationship from the original mixed signal to the slow feature space is established. Finally, through constraint screening and feature decomposition, the heading global features containing steady-state and transient features are obtained, which fully cover the actual heading change law of the driving equipment and the environmental and operating condition interference characteristics.

[0084] Optionally, in an exemplary embodiment, such as Figure 3 As shown, a method for determining the global heading characteristics is provided, which specifically includes the following steps:

[0085] S301, based on a linear transformation model, performs feature space mapping on the original heading angle signal to obtain multi-dimensional initial slow features.

[0086] Among them, the multi-dimensional initial slow features characterize the inherent temporal variation law of the original heading angle signal in different dimensions. That is, the multi-dimensional initial slow features are a set of multi-dimensional feature sequences output after mapping. Each feature characterizes a certain inherent temporal evolution law of the original heading. The fast and slow change characteristics are different and include slow-changing effective components and fast-changing interference components.

[0087] Optionally, the original heading angle signal can be input into the constructed linear transformation model, and window-by-window calculations can be performed according to the linear transformation model to finally obtain multi-dimensional initial slow features.

[0088] For example, in the embodiments of this application, The linear transformation model can be expressed as:

[0089] (1)

[0090] in, This is the transpose of the time-frequency domain statistics matrix; yes The corresponding number Group of linear transformation mapping operators; For timing / sliding window index labels; For the first Dimensional slowness characteristics.

[0091] Specifically, a sliding time window can be added to the raw heading angle signal, and the index can be used to... Data is extracted frame by frame; time-frequency domain statistics are calculated for each extracted frame, and a time-frequency domain statistics matrix is ​​constructed. Then transpose to get Call the first Group linear operators Perform linear operations on the transpose matrix and output the i-th element corresponding to the current window. Dimensional slow features After completing frame preprocessing, high-frequency disturbances are initially stripped away and initial slow-varying components are extracted within each time slice.

[0092] Each slow feature exist The middle is linearly combined with all input data:

[0093] (2)

[0094] in, Represents the coefficient matrix The transpose of the matrix; The original input observation time-frequency data matrix consists of multiple sets of time-frequency domain observation samples. Each row corresponds to the observation samples under different time windows, each column of the matrix corresponds to the observation variables of different frequency points or observation channels, and the matrix elements are the raw observation data without feature mapping processing; This represents the initial slow feature.

[0095] coefficient matrix Obtained by solving the generalized eigenvalue problem:

[0096] (3)

[0097] in, Input observation matrix The time difference matrix, ; The time set averaging operator represents the mathematical expectation of a time series sample. Represents the eigenvalue matrix, eigenvalues . The coefficient matrix corresponds to .

[0098] S302, determine the target slow feature from the multi-dimensional initial slow features.

[0099] Among them, the target slow feature has the smallest rate of change over time; the rate of change over time is the magnitude of change in the feature value within a unit sampling interval, and the smaller the value, the more smoothly the feature changes over time. The target slow feature is the core feature selected from all the initial slow features with the smallest rate of change over time, which carries the information of the actual long-term change in the heading of the driving equipment.

[0100] Optionally, the actual heading evolves slowly with an extremely low rate of change, while the features corresponding to environmental noise fluctuate dramatically with a large rate of change. Therefore, by quantifying the temporal change rate of each initial slow feature, the feature with the smallest rate of change can be selected as the target slow feature, thus preserving the effective heading information.

[0101] Specifically, the temporal change index of each initial slow feature can be calculated one by one, and the average amplitude of the difference between adjacent time points is often used to represent the rate of change. By comparing the change rate values ​​of all dimensions, the feature sequence corresponding to the minimum value is selected as the target slow feature.

[0102] S303 performs state decomposition on the target slow features to obtain steady-state and transient features.

[0103] Optionally, the target slow feature time series is used as input and split using a trend stripping method: the smoothed trend term is used as the steady-state feature, and the residual after subtracting the smoothed trend from the original sequence is used as the transient feature. The two are combined to form the heading global feature.

[0104] In this embodiment, multidimensional initial slow features are generated by linear transformation to complete feature space mapping. Target slow features are then selected based on the rate of change, and steady-state and transient features are obtained through component decomposition, thus fully achieving the decoupling and separation of effective heading information and interference noise. Compared to traditional single-filter decomposition methods, this application autonomously selects features based on the inherent temporal variation law of the data, without the need for manually fixing filter parameters, and adaptively adapts to various operating conditions such as vehicle-mounted, ship-mounted, and airborne applications.

[0105] Optionally, in an exemplary embodiment, such as Figure 4As shown, a method for determining monitoring statistics is provided, which specifically includes the following steps:

[0106] S401 divides the entire heading domain features to obtain the dominant features used to characterize the overall heading change trend and the residual features used to characterize signal disturbance noise.

[0107] Among them, the dominant feature is the main component extracted from the full-domain heading feature, which corresponds to the long-term change trend of the actual heading of the carrier and is the main body of effective heading information; the residual feature is the remaining component after removing the dominant trend from the full-domain feature, which is composed of disturbances such as device noise, sand / fog interference, and operating condition turbulence, and represents various types of random noise.

[0108] Optionally, the heading global feature fusion incorporates the true heading trend and superimposed noise. Therefore, by separating the main body and the residual through trend stripping, effective information and disturbance noise can be decoupled, laying the foundation for the separate calculation of the two types of statistics. Specifically, the heading global feature sequence composed of steady-state and transient states can be used as input, and a trend extraction algorithm can be used to separate the gentle main trend as the dominant feature; the difference between the original global features and the dominant feature sequence is the residual feature.

[0109] For example, the model constructed using the above formula (2) extracts slow features. :

[0110] (4)

[0111] in, Indicates the number of dominant features; It represents the dominant feature and can extract the overall trend of changes in process heading data; This represents the residual characteristic and is considered noise or fluctuation. This represents the total spatial dimension of the slow features, that is, the total number of dimensions of all initial slow features.

[0112] S402. Based on the dominant characteristics and residual characteristics, determine the monitoring statistics.

[0113] Optionally, the dominant feature focuses on long-term, slow drift and is used to solve for static statistics; the residual feature focuses on instantaneous random jumps and is used to solve for dynamic statistics after time-series differencing, forming a dual-indicator monitoring system of static and dynamic indicators. Specifically, the original numerical values ​​of the dominant feature can be used to obtain static statistics representing baseline shift; time-series differencing can be performed on the dominant feature and residual feature at adjacent time points, and dynamic statistics representing instantaneous disturbances can be obtained based on the fluctuation amplitude and dispersion of the difference sequence; the static and dynamic statistics together form a complete set of monitoring statistics.

[0114] In this embodiment, by hierarchically splitting the features of the entire heading domain, the dominant features representing the true heading and the residual features representing interference noise are accurately separated, overcoming the difficulty of distinguishing between effective information and noise aliasing; then, based on the two types of features, static and dynamic dual monitoring statistics are constructed respectively to achieve separate quantitative monitoring of long-term baseline drift error and instantaneous change noise.

[0115] Optionally, in an exemplary embodiment, such as Figure 5 As shown, a method for determining monitoring statistics is provided, which specifically includes the following steps:

[0116] S501, steady-state characteristic analysis is performed on the dominant characteristics and residual characteristics to obtain static statistics.

[0117] Among them, static statistics are quantitative indicators obtained through steady-state characteristic analysis. They are used to characterize the baseline deviation and system drift of the heading characteristics relative to the ideal noiseless state under steady operating conditions, and reflect the slow-varying systematic error.

[0118] Optionally, the dominant feature exhibits long-term baseline drift, and the residual feature exhibits steady-state bias error. Therefore, in this embodiment, steady-state distribution analysis can be performed on both types of features to extract time-series steady-state deviation characteristics, quantify the systematic offset error of the entire signal, and generate static statistics for monitoring baseline error under operating conditions. Specifically, time-series sequences of the dominant feature and residual feature can be collected over a continuous period. Steady-state feature analysis methods such as steady-state mean deviation, steady-state covariance distribution, and baseline offset distance can be used to calculate the overall deviation of the features from the ideal noise-free benchmark. By fusing the trend deviation of the dominant feature and the steady-state bias of the residual feature, a static statistic for characterizing the system drift error can be obtained.

[0119] For example, based on the dominant features and residual features extracted by formula (4) above, the process of determining the static statistics can be expressed as follows:

[0120] (5)

[0121] in, This represents the static statistics corresponding to the dominant feature; This represents the static statistics corresponding to the residual characteristics.

[0122] S502 performs temporal difference processing on the dominant features and residual features to obtain transient change information.

[0123] Among them, transient change information is the instantaneous fluctuation, abrupt change amplitude and dynamic offset data obtained after time series difference, which reflects the dynamic fluctuation and sudden disturbance characteristics of the feature at each moment.

[0124] Optionally, steady-state analysis can only characterize long-term shifts and cannot capture instantaneous abrupt changes and temporal fluctuations. Therefore, by simultaneously performing temporal differencing on the dominant and residual features, fixed steady-state trends can be eliminated, while the details of instantaneous jumps, short-term fluctuations, and abrupt changes can be amplified to extract pure transient dynamic change information. Specifically, the temporal sequences of the dominant and residual features can be differentially analyzed at adjacent sampling times to obtain the temporal changes of the two sets of features. These changes eliminate long-term steady-state shifts and retain only the instantaneous fluctuations, abrupt jumps, and dynamic disturbances at each moment, summarizing to form complete transient change information of the features.

[0125] For example, based on the dominant features and residual features extracted by the above formula (4), the process of performing time-series differencing on the dominant features and residual features can be expressed as follows:

[0126] (6)

[0127] in, Difference features that are the dominant characteristics; It is the difference feature of the residual feature.

[0128] S503, determine dynamic statistics based on transient change information.

[0129] Optionally, statistical operations can be performed on the transient differential change information of the dominant feature and residual feature to extract quantitative parameters such as fluctuation amplitude, dispersion, and abrupt change intensity, and fuse them to obtain dynamic statistics that can characterize the instantaneous disturbance level of the signal in real time, thereby realizing the quantitative monitoring of environmental interference, operating condition fluctuations, and sudden noise.

[0130] For example, based on the above formula (6), the dynamic statistic can be expressed as:

[0131] (7)

[0132] in, The dynamic statistics corresponding to the dominant feature; This is the dynamic statistic corresponding to the residual characteristics.

[0133] In this embodiment, by performing steady-state analysis on the dominant features and residual features to solve the static statistics, the slow-varying errors such as system drift and baseline offset generated by the long-term operation of the polarization compass can be accurately quantified, making up for the shortcomings of traditional methods that cannot monitor steady-state deviations. By using time-series differential processing to remove steady-state trends and extract pure transient change information, the details of instantaneous disturbances caused by environmental interference and sudden changes in operating conditions can be effectively amplified. Then, by quantifying the transient information, dynamic statistics are obtained, enabling accurate measurement of short-term random noise and sudden disturbances.

[0134] Optionally, in an exemplary embodiment, such as Figure 6 As shown, a method for determining the posterior probability of noise corresponding to different noise conditions is provided, which specifically includes the following steps:

[0135] S601 inputs the monitoring statistics into the pre-built noise prior model to obtain the intra-class conditional probability and inter-class conditional probability under different noise conditions.

[0136] Among them, noise conditions are signal operating states categorized according to interference strength, divided into low-noise, medium-noise mixed, and high-noise strong interference conditions, corresponding to different environmental, turbulence, and meteorological disturbance levels, respectively. Intra-class conditional probability is the probability that, given a specific noise condition, the real-time monitored statistics conform to the statistical distribution of that condition, characterizing the degree of matching between the current observed index and the data distribution within the corresponding condition. Inter-class conditional probability is the probability of distribution overlap between different noise conditions, used to correct discrimination errors caused by blurred condition boundaries and to compensate for probability confusion at the boundaries between multiple noise classes.

[0137] In this embodiment, the noise prior model incorporates the distribution patterns of statistical quantities for each working condition. By inputting real-time dynamic and static combined statistical quantities and comparing them with the inherent distribution of each working condition, the model calculates the matching probability (intra-class conditional probability) of the current observed data falling within each type of working condition and the crossover probability (inter-class conditional probability) of different working conditions overlapping, providing intermediate parameters for the posterior solution.

[0138] Based on this, the monitoring statistics are used as observation samples and imported into the noise prior model. The noise prior model retrieves the pre-stored distribution parameters for low, medium, and high noise conditions, calculates the fit between the current statistics and the distribution of the corresponding condition for each condition, and outputs the intra-class conditional probability for each condition. Based on the overlapping intervals of the distributions of various conditions, the inter-class conditional probability of mutual interference between different conditions is calculated for subsequent boundary error compensation.

[0139] For example, the prior model uses a multidimensional Gaussian distribution to fit various noise conditions. After inputting multidimensional monitoring statistics, it simultaneously solves the intra-class and inter-class conditional probabilities based on the Gaussian probability density function. This model has the highest fit with the dynamic and static statistical system mentioned above and the best accuracy in identifying harsh environments. For example, according to the first... Events corresponding to each heading angle state Calculate the intra-class conditional probability and the inter-class conditional probability:

[0140] (8)

[0141] in, and They represent the first Intra-class conditional probability and inter-class conditional probability of each heading angle state; Indicates the first The monitoring statistics for each heading angle state, and its corresponding statistical limit is: .

[0142] S602, based on the preset prior probability, and the intra-class conditional probability and inter-class conditional probability under different noise conditions, determine the noise posterior probability corresponding to different noise conditions.

[0143] The preset prior probability is the probability of various noise conditions occurring naturally, obtained from historical big data statistics during the engineering calibration phase. It does not change with real-time sampling data and is a fixed prior parameter.

[0144] Optionally, based on the fundamental Bayesian probability criterion, the posterior probability of each noise condition is calculated in real time by using the natural occurrence probability (prior probability) of the operating condition as a benchmark and combining it with the two types of conditional probabilities for weighted correction calculation, eliminating errors caused by cross-interference of operating conditions. Specifically, the prior probabilities of low, medium, and high noise operating conditions are retrieved from pre-calibrated fixed values; the intra-class conditional probability is used as the main matching coefficient, and the inter-class conditional probability is used as the cross-correction coefficient; according to the Bayesian conversion rules, the calculation is corrected item by item to obtain the final posterior probability of noise for each noise operating condition.

[0145] With the first Events corresponding to each heading angle state For example, the posterior probability of noise It can be calculated using the following formula:

[0146] (9)

[0147] in, Indicates an event The corresponding prior probability; Indicates non-event The corresponding prior probability; ; .

[0148] In this embodiment, by splitting the conditional probabilities within and between classes, the matching degree between the statistical quantity and the single noise condition is considered, and the discrimination bias caused by the overlap of different noise conditions is compensated, which significantly improves the accuracy of noise probability quantification in complex and variable scenarios such as sandstorms, dense fog, and motor vehicle bumps.

[0149] Optionally, in an exemplary embodiment, such as Figure 7 As shown, a method for determining heading state data with different noise levels is provided, which specifically includes the following steps:

[0150] S701 determines the target noise condition corresponding to the maximum noise posterior probability based on the noise posterior probability corresponding to different noise conditions.

[0151] Among them, the target noise condition is the type of noise condition with the largest posterior probability among all noise conditions at the same sampling time, including three categories: low noise condition, medium noise condition, and high noise condition.

[0152] Understandably, different posterior probabilities represent the degree of confidence that a signal belongs to a corresponding operating condition. The operating condition corresponding to the maximum probability is the actual noise operating condition that best matches the current signal. The maximum posterior criterion is used to complete the selection of the optimal operating condition.

[0153] Optionally, different posterior probabilities can be used to represent the credibility of the signal belonging to the corresponding working condition. The working condition corresponding to the maximum probability is the real noise working condition that best matches the current signal. The optimal working condition is selected by using the maximum posterior criterion.

[0154] S702 determines the noise category corresponding to the original heading angle signal based on the target noise conditions.

[0155] Among them, the noise category is a signal classification label obtained based on the target noise condition mapping, which is divided into low noise effective signal category, medium noise mixed signal category, and high noise interference signal category.

[0156] Optionally, a one-to-one mapping relationship between noise conditions and noise categories can be pre-established. The selected target noise conditions are directly matched with the corresponding noise classification labels to complete the signal attribute classification. Specifically, the preset mapping rules can be: low noise condition → low noise category, medium noise condition → medium noise category, high noise condition → high noise category; based on the determined target noise conditions, the noise category of the current sampling point signal is obtained by looking up the table.

[0157] S703 divides the original heading angle signal into states according to the noise category to obtain heading state data with different noise contents.

[0158] Optionally, the continuous long-time-series original heading signal can be divided into multiple segments using noise category labels marked at each point. Within each segment, the noise level is consistent, ultimately forming heading status data with differentiated noise content. Specifically, along the original heading time sequence, segments are made according to the noise category labels of each sampling point. Sampling points with consecutive noise categories are grouped together, resulting in three categories of heading status data: low noise, medium noise, and high noise.

[0159] For example, the current state Assign the heading angle state with the highest posterior probability. :

[0160] (10)

[0161] in, , and This represents heading status data with different noise levels.

[0162] In this embodiment, dynamic classification is based on measured probabilities, which is not bound by manually set parameters and can adaptively adjust the division boundary according to changing environments such as sandstorms, dense fog, and carrier turbulence. The noise attribute is quantitatively classified through working condition-category mapping, which effectively avoids the defect that a single threshold cannot adapt to multi-gradient noise. The noise level boundary of the divided heading status data is clear.

[0163] Optionally, in an exemplary embodiment, such as Figure 8 As shown, a method for optimizing the original heading angle signal is provided, which specifically includes the following steps:

[0164] S801 determines the elasticity factor of each heading state data based on the data characteristics of heading state data with different noise contents.

[0165] Among them, data characteristics are the statistical attributes of a single set of heading state data, including serial correlation coefficient, variance, signal-to-noise ratio, amplitude fluctuation, etc. The elasticity factor is an adjustable parameter used to adaptively control the noise reduction intensity of singular spectrum analysis. The higher the noise, the larger the elasticity factor value and the stronger the noise reduction; the lower the noise, the smaller the elasticity factor, preserving the original heading details to the maximum extent.

[0166] Optionally, the time-domain features (signal-to-noise ratio, autocorrelation coefficient, and signal variance) of each type of heading state data can be extracted separately and substituted into a preset mapping relationship. The higher the noise content and the lower the signal-to-noise ratio, the larger the value of the elasticity factor obtained; conversely, the elasticity factor of low-noise data is smaller. The independent elasticity factors corresponding to the three types of data are determined one by one.

[0167] For example, elasticity factor One method to determine this is to first base it on the length of the LFS (Length of FragmentSeries). Calculate the difference threshold :

[0168] (11)

[0169] in, It refers to the number of sampling points for a single segment of heading state data, i.e., the total length of a single set of low-noise / medium-noise / high-noise segmented time-series data after noise classification, used to link elasticity factors and control... Embedded window length; Indicates the first One sampling point.

[0170] Furthermore, according to Data obtained before and after LFS denoising and The difference between Less than Determine the standard threshold Calculate the correlation coefficient This correlation coefficient can effectively determine the correlation between the current heading angle data and all heading angle data:

[0171] (12)

[0172] in, This indicates the length of the original heading angle data; and Indicates the first The average of the individual sample signals and the total signal; and They represent the first The and the first A recognized heading angle signal.

[0173] Finally, based on the correlation coefficient between the current heading angle data and all heading angle data... and Calculate the elasticity factor :

[0174] (13)

[0175] S802, based on the elasticity factor of each heading state data, optimizes the parameters of the pre-constructed singular spectrum analysis model to obtain the optimized singular spectrum analysis model.

[0176] The pre-constructed singular spectrum analysis model is a basic SVD singular spectrum decomposition and denoising model. The model inherently has fixed parameters such as the decomposition order and threshold, without optimization based on real-time noise characteristics. The optimized singular spectrum analysis model is an improved model that uses the elasticity factor corresponding to each set of data to correct the decomposition threshold and the singular value truncation order (i.e.,...). Robust sliding singular spectrum model).

[0177] Optionally, for each type of heading state data, two key parameters of the singular spectrum model are modified using the corresponding elasticity factor: one is the singular value screening threshold, and the other is the effective feature component truncation order. When the elasticity factor increases, the threshold is increased and more noise components corresponding to small singular values ​​are discarded. When the elasticity factor decreases, the threshold is decreased and more subtle heading components are retained, ultimately generating a singular spectrum model specifically optimized for each type of data.

[0178] S803 employs an optimized singular spectrum analysis model to denoise heading state data with varying noise levels.

[0179] Optionally, low-noise heading data is lightly filtered using a model optimized with a small elasticity factor; medium-noise mixed data is denoised using a model with medium parameters for balanced denoising; and high-noise data is denoised by using a model optimized with a large elasticity factor to strongly filter out high-frequency noise, thus completing the denoising for each group of data.

[0180] S804 splices and reconstructs the denoised heading state data to obtain an optimized signal of the original heading angle signal.

[0181] Optionally, the denoised data from the low-noise, medium-noise, and high-noise groups are sequentially joined together according to the original signal time sequence to eliminate segmentation breaks and generate a continuous and uninterrupted optimized heading angle timing signal.

[0182] In this embodiment, the elasticity factor is adaptively obtained based on the characteristics of grouped data, and the singular spectrum model parameters are customized and optimized based on the elasticity factor to achieve graded differentiated singular spectrum denoising and time-series reconstruction. No manual on-site tuning of filter parameters is required, and the environmental adaptability is strong. The optimized heading signal obtained by the final splicing and reconstruction achieves the optimal balance between fidelity and denoising effect, significantly improving the heading output accuracy and curve stability of the polarization compass under varied and harsh working conditions.

[0183] Figure 9 This is a flowchart illustrating a signal optimization method for a polarized optical compass in another embodiment. Based on the above embodiments, this embodiment provides an optional example of a signal optimization method for a polarized optical compass. (Combined with...) Figure 9 The specific implementation process is as follows:

[0184] S901: Acquire the original heading angle signal collected by the polarization compass in the driving equipment, and construct the time-frequency domain statistical matrix corresponding to the original heading angle signal based on the original heading angle signal.

[0185] S902, construct a linear transformation model based on the transpose of the time-frequency domain statistics matrix.

[0186] The linear transformation model represents the mapping relationship between the original heading angle signal and the slow feature space.

[0187] S903, based on a linear transformation model, performs feature space mapping on the original heading angle signal to obtain multi-dimensional initial slow features.

[0188] Among them, the multi-dimensional initial slow features characterize the intrinsic temporal variation law of the original heading angle signal in different dimensions.

[0189] S904 determines the target slow feature from the multi-dimensional initial slow features.

[0190] Among them, the target slow feature has the smallest rate of change over time.

[0191] S905 performs state decomposition on the target slow features to obtain steady-state and transient features.

[0192] S906 divides the overall heading characteristics to obtain dominant features that characterize the overall heading trend and residual features that characterize signal disturbance noise.

[0193] S907 performs steady-state characteristic analysis on the dominant and residual characteristics to obtain static statistics.

[0194] S908 performs temporal difference processing on the dominant features and residual features to obtain transient change information.

[0195] S909, determine dynamic statistics based on transient change information.

[0196] S910 inputs the monitoring statistics into the pre-built noise prior model to obtain the intra-class conditional probability and inter-class conditional probability under different noise conditions.

[0197] S911, based on the preset prior probability, and the intra-class conditional probability and inter-class conditional probability under different noise conditions, determine the noise posterior probability corresponding to different noise conditions.

[0198] S912 determines the target noise condition corresponding to the maximum noise posterior probability based on the noise posterior probability corresponding to different noise conditions.

[0199] S913 determines the noise category corresponding to the original heading angle signal based on the target noise conditions.

[0200] S914 divides the original heading angle signal into states according to the noise category to obtain heading state data with different noise contents.

[0201] The S915 optimizes the original heading angle signal based on heading status data with different noise levels.

[0202] Optionally, based on the data characteristics of heading state data with different noise contents, the elasticity factor of each heading state data is determined; based on the elasticity factor of each heading state data, the parameters of the pre-constructed singular spectrum analysis model are optimized to obtain the optimized singular spectrum analysis model; the optimized singular spectrum analysis model is used to denoise the heading state data with different noise contents; the denoised heading state data are spliced ​​and reconstructed to obtain the optimized signal of the original heading angle signal.

[0203] The specific processes of S901-S915 described above can be found in the description of the above method embodiments. Their implementation principles and technical effects are similar, and will not be repeated here.

[0204] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0205] Based on the same inventive concept, this application also provides a signal optimization device for a polarized optical compass used to implement the signal optimization method for the polarized optical compass described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in the one or more polarized optical compass signal optimization device embodiments provided below can be found in the limitations of the polarized optical compass signal optimization method described above, and will not be repeated here.

[0206] In one exemplary embodiment, such as Figure 10 As shown, a signal optimization device 1000 for a polarization compass is provided, comprising: a feature extraction module 1010, a statistical analysis module 1020, a probability determination module 1030, a data determination module 1040, and a signal optimization module 1050, wherein:

[0207] The feature extraction module 1010 is used to acquire the original heading angle signal collected by the polarization compass in the driving equipment, and extract the heading global features from the original heading angle signal; wherein, the heading global features include steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions;

[0208] The statistical analysis module 1020 is used to perform differential domain analysis on the full-domain features of the heading to obtain monitoring statistics. Among them, the monitoring statistics include dynamic statistics that characterize the instantaneous fluctuation of the heading features of the driving equipment and static statistics that characterize the deviation of the heading features of the driving equipment under stable operating conditions from the noise-free state.

[0209] The probability determination module 1030 is used to determine the posterior probability of noise corresponding to different noise conditions based on the monitoring statistics and the pre-built noise prior model.

[0210] The data determination module 1040 is used to divide the original heading angle signal into states according to the noise posterior probability corresponding to different noise conditions, so as to obtain heading state data with different noise contents.

[0211] The signal optimization module 1050 is used to optimize the original heading angle signal based on heading state data with different noise contents.

[0212] In one embodiment, the feature extraction module 1010 includes:

[0213] The matrix construction unit is used to construct the time-frequency domain statistics matrix corresponding to the original heading angle signal based on the original heading angle signal.

[0214] The model building unit is used to construct a linear transformation model based on the transpose of the time-frequency domain statistics matrix; wherein, the linear transformation model represents the mapping relationship between the original heading angle signal and the slow feature space;

[0215] The feature extraction unit is used to determine the global features of the heading based on the linear transformation model.

[0216] In one embodiment, the feature extraction unit is specifically used for:

[0217] Based on the linear transformation model, the original heading angle signal is mapped to the feature space to obtain multi-dimensional initial slow features. The multi-dimensional initial slow features characterize the inherent temporal variation law of the original heading angle signal in different dimensions. From the multi-dimensional initial slow features, the target slow features are determined. The target slow features have the smallest time change rate. The target slow features are decomposed into steady-state features and transient features.

[0218] In one embodiment, the statistical analysis module 1020 includes:

[0219] The feature segmentation unit is used to segment the features of the entire heading domain to obtain the dominant features that characterize the overall trend of heading change and the residual features that characterize signal disturbance noise.

[0220] The statistical analysis unit is used to determine monitoring statistics based on dominant characteristics and residual characteristics.

[0221] In one embodiment, the statistical analysis unit is specifically used for:

[0222] Steady-state characteristic analysis is performed on the dominant and residual characteristics to obtain static statistics; time-series differencing is performed on the dominant and residual characteristics to obtain transient change information; and dynamic statistics are determined based on the transient change information.

[0223] In one embodiment, the probability determination module 1030 is specifically used for:

[0224] The monitoring statistics are input into a pre-built noise prior model to obtain the intra-class conditional probability and inter-class conditional probability under different noise conditions. Based on the preset prior probability and the intra-class conditional probability and inter-class conditional probability under different noise conditions, the noise posterior probability corresponding to different noise conditions is determined.

[0225] In one embodiment, the data determination module 1040 is specifically used for:

[0226] Based on the posterior probability of noise corresponding to different noise conditions, the target noise condition corresponding to the maximum posterior probability is determined; based on the target noise condition, the noise category corresponding to the original heading angle signal is determined; based on the noise category, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

[0227] In one embodiment, the signal optimization module 1050 is specifically used for:

[0228] Based on the data characteristics of heading state data with different noise levels, the elasticity factor of each heading state data is determined. Based on the elasticity factor of each heading state data, the parameters of the pre-constructed singular spectrum analysis model are optimized to obtain the optimized singular spectrum analysis model. The optimized singular spectrum analysis model is used to denoise the heading state data with different noise levels. The denoised heading state data are then spliced ​​and reconstructed to obtain the optimized signal of the original heading angle signal.

[0229] Each module in the aforementioned polarization compass signal optimization device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0230] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a signal optimization method for a polarization compass.

[0231] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0232] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the signal optimization method for a polarization compass provided in the above embodiments.

[0233] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the signal optimization method for the polarization compass provided in the above embodiment.

[0234] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the signal optimization method for a polarization compass provided in the above embodiments.

[0235] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0236] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0237] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0238] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for optimizing the signal of a polarized optical compass, characterized in that, The method includes: The original heading angle signal collected by the polarization compass in the driving equipment is acquired, and the global heading feature is extracted from the original heading angle signal; wherein, the global heading feature includes steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions; Differential domain analysis is performed on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics for characterizing the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics for characterizing the deviation of the heading characteristics of the driving equipment under steady operating conditions relative to the noise-free state. Based on the monitoring statistics and the pre-built noise prior model, the noise posterior probability corresponding to different noise conditions is determined. Based on the noise posterior probability corresponding to different noise conditions, the original heading angle signal is divided into states to obtain heading state data with different noise contents. The original heading angle signal is optimized based on heading state data with different noise levels.

2. The method according to claim 1, characterized in that, The extraction of global heading features from the original heading angle signal includes: Based on the original heading angle signal, construct the time-frequency domain statistics matrix corresponding to the original heading angle signal; A linear transformation model is constructed based on the transpose of the time-frequency domain statistics matrix; wherein the linear transformation model characterizes the mapping relationship between the original heading angle signal and the slow feature space; Based on the linear transformation model, the global heading characteristics are determined.

3. The method according to claim 2, characterized in that, The determination of the global heading characteristics based on the linear transformation model includes: Based on the linear transformation model, the original heading angle signal is mapped into a feature space to obtain multi-dimensional initial slow features; wherein, the multi-dimensional initial slow features characterize the inherent temporal variation law of the original heading angle signal in different dimensions. From the initial slow features in multiple dimensions, a target slow feature is determined; wherein the target slow feature has the smallest rate of change over time. The target slow feature is decomposed into state to obtain the steady-state feature and the transient feature.

4. The method according to any one of claims 1-3, characterized in that, The differential domain analysis of the global heading features yields monitoring statistics, including: The heading global features are divided to obtain the dominant features used to characterize the overall trend of heading change and the residual features used to characterize signal disturbance noise. Based on the dominant feature and the residual feature, the monitoring statistics are determined.

5. The method according to claim 4, characterized in that, The determination of monitoring statistics based on the dominant feature and the residual feature includes: Steady-state characteristic analysis is performed on the dominant feature and the residual feature to obtain static statistics; Temporal difference processing is performed on the dominant features and residual features to obtain transient change information; Based on the transient change information, determine the dynamic statistics.

6. The method according to any one of claims 1-3, characterized in that, The step of determining the posterior probability of noise corresponding to different noise conditions based on the monitoring statistics and the pre-constructed noise prior model includes: The monitoring statistics are input into a pre-built noise prior model to obtain the intra-class conditional probability and inter-class conditional probability under different noise conditions. Based on the preset prior probability, as well as the intra-class conditional probability and inter-class conditional probability under different noise conditions, the noise posterior probability corresponding to different noise conditions is determined.

7. The method according to any one of claims 1-3, characterized in that, The step of dividing the original heading angle signal into states based on the noise posterior probability corresponding to different noise conditions to obtain heading state data with different noise contents includes: Based on the posterior probability of noise corresponding to different noise conditions, determine the target noise condition corresponding to the maximum posterior probability of noise. Based on the target noise conditions, determine the noise category corresponding to the original heading angle signal; Based on the noise category, the original heading angle signal is divided into states to obtain heading state data with different noise contents.

8. The method according to any one of claims 1-3, characterized in that, The optimization of the original heading angle signal based on heading state data with different noise contents includes: Based on the data characteristics of heading state data with different noise contents, determine the elasticity factor of each heading state data. Based on the elasticity factor of each heading state data, the parameters of the pre-constructed singular spectrum analysis model are optimized to obtain the optimized singular spectrum analysis model. An optimized singular spectrum analysis model was used to denoise the heading state data with different noise levels. The denoised heading state data are spliced ​​and reconstructed to obtain the optimized signal of the original heading angle signal.

9. A signal optimization device for a polarized optical compass, characterized in that, The device includes: The feature extraction module is used to acquire the original heading angle signal collected by the polarization compass in the driving equipment, and extract the heading global features from the original heading angle signal; wherein, the heading global features include steady-state features reflecting the long-term stable change of the heading of the driving equipment and transient features reflecting environmental disturbances and sudden changes in operating conditions; The statistical analysis module is used to perform differential domain analysis on the global heading characteristics to obtain monitoring statistics; wherein, the monitoring statistics include dynamic statistics used to characterize the instantaneous fluctuation of the heading characteristics of the driving equipment and static statistics used to characterize the deviation of the heading characteristics of the driving equipment under stable operating conditions relative to the noise-free state. The probability determination module is used to determine the posterior probability of noise corresponding to different noise conditions based on the monitoring statistics and the pre-built noise prior model. The data determination module is used to divide the original heading angle signal into states according to the noise posterior probability corresponding to different noise conditions, so as to obtain heading state data with different noise contents. The signal optimization module is used to optimize the original heading angle signal based on heading state data with different noise contents.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images