A method and device for cleaning data of on-line calibration of three-axis magnetometer in marine carrier assembly state

Abstract

The application discloses a kind of marine carrier assembly state three-axis magnetometer online calibration data cleaning method and device, it is related to marine magnetic survey equipment calibration data processing technical field.The method includes obtaining the point cloud sample point set of marine magnetometer three-axis magnetic field observation;By white transformation, the point cloud sample point set is mapped to white space, and in white space, the instantaneous estimate value of spherical shell radius is estimated by direction binning;Then, according to the spherical shell radius, the noise scale is combined to determine the outer offset shell belt interval;Finally, after iterative updating ellipsoidal shell parameter convergence, the sample point set finally falling in the outer offset shell belt interval is used as calibration data.The application can effectively eliminate internal cluster, external outlier, and suppress non-uniform sampling interference, and then high-precision extract intrinsic shell structure from sensor raw data containing complex structure noise, to provide high-purity input data for subsequent sensor calibration calculation.

Description

Technical Field

[0001] This invention relates to the field of calibration data processing technology for marine magnetic surveying equipment, specifically to a method and apparatus for cleaning online calibration data of a triaxial magnetometer in the assembled state of a marine carrier. This method is applicable to situations where the magnetometer is installed on a marine carrier such as a shipborne platform, towed body, or underwater vehicle, and under conditions where the propulsion / power supply is operating, the carrier structure is magnetized, the external background field is difficult to guarantee purity, and attitude excitation is limited. It filters and controls the quality of abnormal samples caused by intra-shell clusters, extra-shell outliers, and non-uniform directional sampling in the calibration data collected. Background Technology

[0002] Magnetometers, accelerometers, and other triaxial sensors are core components of inertial measurement units, attitude and heading reference devices, and various motion sensing devices. Their measurement accuracy directly determines the performance of subsequent navigation, positioning, and control devices.

[0003] Taking a triaxial vector magnetometer as an example, under ideal conditions where the external magnetic field is approximately uniform and the sensor is fully excited in attitude, the endpoint of the sensor output vector in the three-dimensional coordinate system should approximately fall on a sphere centered at the origin. However, in practical applications, when the sensor is inevitably subjected to errors such as hard iron bias, soft iron distortion, and triaxial non-orthogonality, the aforementioned sphere will undergo translation and linear distortion, and the observation point set will usually present as an ellipsoidal shell structure in three-dimensional space. Therefore, using ellipsoid fitting to solve for and compensate for hard iron, soft iron, and non-orthogonal parameters is a relatively mature calibration path in engineering.

[0004] However, in in-situ calibration scenarios at the assembly level (device level) of marine carriers, the aforementioned ideal premises are often difficult to meet, leading to significant engineering obstacles in traditional ellipsoidal fitting calibration procedures. Specifically, marine magnetometers are typically installed on carriers such as shipboard platforms, towed bodies, or underwater vehicles for overall calibration. The calibration process often cannot achieve complete isolation from the carrier structure, power cables, and propulsion / power units. The start-up and shutdown of thrusters, motors, and pumps, as well as fluctuations in cable current, introduce short-term, clustered, and direction-dependent magnetic interference. Coupled with factors such as magnetization of the carrier structure and the difficulty in maintaining a pure environmental background field, the calibration data collected in the three-axis space will not only contain obvious outliers outside the ellipsoidal shell main structure but also clusters inside the shell, and may exhibit abnormal patterns such as overall shifts and local collapses. At the same time, due to the influence of sea state disturbances, attitude control limitations, towed stability, and mission trajectory constraints, the calibration attitude excitation is usually insufficient, and the observation direction coverage is prone to ring-shaped or local sector distribution, with a strongly non-uniform sampling density. Under the above conditions, traditional ellipsoidal fitting methods that rely on uniform attitude coverage or assume a relatively pure background field are prone to problems such as shape drift, excessive retention of outliers, or accidental deletion of effective shell points, thereby reducing the reliability and repeatability of in-situ calibration of the assembly state.

[0005] While existing point cloud statistical filtering and neighborhood smoothing are general cleaning methods that can suppress some noise, they typically do not explicitly utilize the geometric prior that "calibration data should satisfy ellipsoidal shell consistency." Furthermore, they struggle to achieve stable screening and quality control of valid calibration data in situations with uneven directional coverage, clustered outlier distributions, and a mixture of intra-shell clusters and valid shell points. Although ellipsoid fitting and error compensation calibration methods for triaxial sensors are relatively mature, there is still a significant gap in dedicated data cleaning techniques for device-level calibration of marine magnetometers. Moreover, traditional ellipsoid fitting methods are prone to drift and overfitting problems because they are not adapted to the aforementioned marine conditions.

[0006] Therefore, there is an urgent need for a data cleaning method for in-situ calibration of marine carriers in assembly state, which can effectively remove abnormal samples from the calibration data and output a high-quality dataset that meets the consistency of the ellipsoidal shell under conditions of impure background field, strong interference and poor attitude coverage, so as to provide reliable input for subsequent assembly state calibration parameter solution. Summary of the Invention

[0007] In view of the above problems, this invention proposes a method and apparatus for cleaning online calibration data of a triaxial magnetometer in the assembled state of a marine carrier. The aim is to utilize the prior geometric knowledge that sensor observation data should be distributed within an ellipsoidal shell, transforming the outlier cleaning problem into a robust joint estimation problem of ellipsoidal parameters and shell thickness. By simulating the adaptive expansion and conformation of a physical balloon to the shell under pressure, geometrically consistent effective interior points can be accurately extracted from point cloud data with complex contamination.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] Firstly, this embodiment provides a method for cleaning online calibration data of a triaxial magnetometer in the assembled state of a marine carrier, the steps of which include: Acquire triaxial magnetic field observation data from an ocean magnetometer, and treat the triaxial components at each moment as point cloud sample point sets in a three-dimensional coordinate system; By whitening conversion, the point cloud sample point set is mapped to the whitening space, so that the point cloud sample point set is approximately spherically distributed in the whitening space; In the whitened space, the instantaneous estimate of the spherical shell radius is estimated by directional binning; based on the spherical shell radius, the outer shell zone interval is determined in combination with the noise scale; The ellipsoidal shell parameters are iteratively updated, and after convergence, the set of sample points that finally fall within the outer shell zone is used as calibration data.

[0010] Preferably, the whitening conversion process includes: Let the ellipsoid shape matrix be a symmetric positive definite shape matrix. For the symmetric positive definite shape matrix Perform Cholesky decomposition;

[0011] In the formula, This represents a lower triangular invertible matrix. Indicates transpose; use Perform whitening conversion on the point cloud sample point set:

[0012] in, For the first time after albino A point cloud sample vector, For the original number A point cloud sample vector, It is the center of the ellipsoid.

[0013] Preferably, estimating the radius of the spherical shell by directional binning includes: In whitening space, based on unit direction vector

[0014]

[0015] In the formula, Indicates the first time after albinism A point cloud sample vector, This represents the unit direction vector.

[0016] The set of radii for each point inside the direction box The outer quantile is taken as the local shell radius estimate of the current orientation box; The instantaneous estimate of the spherical shell radius is obtained by taking the median of the local shell radii of all effective boxes. .

[0017] Preferably, the outer shell zone interval is determined based on the spherical shell radius and in conjunction with a noise scale, including: Based on the instantaneous estimate of the spherical shell radius Determine the shell distance of each point cloud A sample set is constructed based on the shell distance; Based on the sample set, the noise scale is estimated using the median absolute deviation (MAD). Determine the shell band half-thickness based on the noise scale. , These are adjustable bandwidth control parameters; The outer off-center shell zone is defined based on the shell zone half-thickness. ;in, This represents the instantaneous estimate of the radius of the spherical shell. and This indicates an adjustable bias parameter, which controls the inward contraction to eliminate internal cluster points and the outward expansion to encompass natural noise and minor disturbances.

[0018] Preferably, updating the ellipsoidal shell parameters includes determining a directionally balanced resultant pressure force based on the shell points, and updating the ellipsoid center using the resultant pressure force: For each orientation box, the normal vector of all shell points within the box. Based on shell consistency weight We perform a weighted average to obtain the resultant force within the box. ; The resultant pressure is obtained by averaging the resultant forces within all the boxes. , Based on the aforementioned pressure resultant force, the ellipsoid center is updated by mapping back to the original space through inverse whitening:

[0019] In the formula, Indicates the resultant force of pressure. L This represents a lower triangular invertible matrix. Indicates the update step size. This represents the center of the original ellipsoid.

[0020] Preferably, updating the ellipsoidal shell parameters includes iteratively updating the ellipsoidal shape matrix using Tyler-type M estimators combined with directional balance weights, and the update method is as follows:

[0021] in, This represents the updated ellipsoid shape matrix. For data dimensions, This represents the total number of effective weights after direction normalization. For an effective set of steering boxes, For the steering box index, Indicates the first The set of candidate shell point indices within each orientation box, i.e., the point cloud samples currently covered by the ellipsoidal shell. For the index of point cloud samples, For the original number A point cloud sample vector, Centered at the ellipsoid The effective weights after directional equilibrium are expressed as follows:

[0022] in, No. Shell consistency weights for individual point clouds. To be consistent with the sample Index of candidate point cloud samples located within the same orientation box. For the first Shell consistency weights for each candidate shell point and Belonging to the same direction box The expression is:

[0023] In the formula, Indicates the first The coordinate vector of a point cloud sample in whitening space This represents the instantaneous estimate of the spherical shell radius in whitened space. The noise scale parameter or shell thickness scale parameter in the direction of whitening radius is used to control the decay rate of the weighting function.

[0024] Preferably, when updating the ellipsoid shape matrix, further damping and trace normalization are applied:

[0025] In the formula, This represents the ellipsoid shape matrix of the current iteration. This represents the updated ellipsoid shape matrix. The damping factor is updated for shape and can be adaptively scaled by coverage gating; For matrix trace operations, The average eigenvalues ​​of the shape matrix are used to normalize the shape matrix to... This eliminates scale degrees of freedom and ensures stable updates.

[0026] Preferably, updating the ellipsoidal shell parameters includes adaptively adjusting the update step size and shell band parameters based on the directional distribution entropy. The directional distribution entropy calculation process is as follows: Count the number of candidate shell points in the direction box And define the probability distribution. ; Calculate directional entropy ; Define the number of effective coverage boxes ; And define soft coverage metrics ; Based on the aforementioned soft coverage index, the adaptive scaling ellipsoid center update step size and shape update damping factor are used.

[0027]

[0028] In the formula, This indicates the step size for updating the ellipsoid center. This represents the shape update damping factor. The reference center step size damping factor, As the reference shape damping factor, for The worse the coverage, the more conservative the center and shape updates. This indicates a soft coverage target.

[0029] Preferably, updating the ellipsoidal shell parameters includes exponentially sliding the ellipsoidal shell radius to suppress estimation fluctuations:

[0030] in, This represents the instantaneous estimated radius of the spherical shell in the t-th iteration; This represents the instantaneous estimated radius of the spherical shell in the t-th iteration. This represents the instantaneous estimated radius of the spherical shell in the (t-1)th iteration. This represents the smoothing factor.

[0031] Secondly, this embodiment also provides an online calibration data cleaning device for a triaxial magnetometer in the assembled state of a marine carrier. This device is used to implement the online calibration data cleaning method for a triaxial magnetometer in the assembled state of a marine carrier as described in any of the preceding claims. The device includes: The sensor module is used to acquire triaxial magnetic field observation data of the marine magnetometer and to treat the triaxial components at each time as a point cloud sample set in a three-dimensional coordinate system. The data cleaning device is used to map the point cloud sample point set to a whitening space through whitening transformation, and in the whitening space, to estimate the instantaneous value of the spherical shell radius by directional binning; to determine the outer shell zone interval based on the spherical shell radius and the noise scale; and to iteratively update the ellipsoidal shell parameters, and after convergence, to use the sample point set that finally falls within the outer shell zone interval as calibration data.

[0032] This invention provides a method and apparatus for cleaning online calibration data of a triaxial magnetometer in the assembled state of a marine carrier. By deeply integrating ellipsoidal shell geometric priors, robust statistical estimation, and adaptive engineering enhancement mechanisms, it can effectively remove internal clusters and external outliers, and suppress non-uniform sampling interference. This allows for the high-precision extraction of intrinsic shell structure from the original sensor data containing complex structural noise, providing high-purity input data for subsequent sensor calibration calculations. This overcomes the technical bottlenecks of insufficient robustness and poor adaptability to complex marine conditions in traditional cleaning methods.

[0033] This invention further constructs an integrated data cleaning device that deeply integrates hardware and software, organically coupling the hardware device's condition triggering gating, direction coverage measurement detection, and software algorithm quality control logic to achieve quality control throughout the entire data processing process. The multi-module architecture of the hardware device can complete the classification and storage, high-speed transmission, and local / remote interaction of data. The cleaned data can be directly ellipsoidal calibration and fitting and visualization on the hardware side, and can also achieve flexible data interaction with external terminals through multiple communication interfaces. This realizes end-to-end closed-loop processing of raw data from marine magnetometers, from hardware acquisition, transmission and storage, to software algorithm cleaning and calculation, to hardware calibration and analysis, result output, and external data interaction. This integrated device features comprehensive and targeted optimization design for complex marine conditions, achieving high synergy between hardware and software and seamless integration of various processing stages. Compared to traditional single hardware and software solutions, it boasts significant technical advantages such as strong robustness, high adaptability to operating conditions, complete processing flow, and superior ease of operation. It completely solves the industry pain points of hardware and software disconnect and fragmented data processing flow in traditional technologies, providing a highly reliable and adaptable integrated sensor data preprocessing solution for fields such as marine magnetometer calibration, magnetic attitude detection, and spatial positioning. It has outstanding engineering application value and technology promotion value. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of typical data distribution of a marine magnetometer under shipboard attitude-constrained conditions.

[0036] Figure 2 This is a schematic diagram of whitening transformation.

[0037] Figure 3 This is a schematic diagram of directional binning and equal-area binning.

[0038] Figure 4 This is a schematic diagram of the externally biased shell structure.

[0039] Figure 5 This is a schematic diagram for calculating the resultant force of directional equilibrium pressure.

[0040] Figure 6 The image shows the distribution of point sets before and after cleaning, and the fitting effect of the fitted ellipsoidal shell, in the actual data of the marine magnetometer used as an example.

[0041] Figure 7This is a schematic diagram of the calibration data cleaning device for a three-axis vector magnetometer for in-situ calibration of a marine carrier assembly, provided in an embodiment of the present invention.

[0042] Figure 8 This is a schematic diagram showing the connection between the data cleaning device, the marine magnetometer, and the computer provided in this embodiment of the invention.

[0043] Figure 9 This is a three-view schematic diagram of the data cleaning apparatus provided in an embodiment of the present invention.

[0044] Figure 10 This is a flowchart illustrating the data cleaning process for a three-axis vector magnetometer calibration method for in-situ calibration of marine carriers in the assembled state, as provided in this embodiment of the invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Under ideal, uniform geomagnetic field conditions, the error-free output of a triaxial magnetometer should be distributed across a spherical surface centered at the origin. However, in real-world data acquisition conditions on offshore platforms, short-term clustered magnetic interference can be introduced by factors such as thruster, motor pump, power cable current fluctuations, and relay switching. Simultaneously, influenced by heading control, pitch / roll limiting, towed body stability, and sea state disturbances, the magnetometer's attitude traversal is often insufficient, resulting in a ring-shaped or local sector-like distribution of directional coverage. This leads to a composite structure in the actual point cloud: "ellipsoidal shell, internal and external contamination, and non-uniform directional sampling." Consequently, traditional ellipsoidal fitting and outlier removal methods, which rely on the assumption of uniform sampling, are prone to shape drift, excessive retention of outliers, or accidental deletion of valid shell points. Figure 1 This is a schematic diagram of typical data distribution of an ocean magnetometer under shipboard attitude-constrained conditions. The original three-axis magnetic field observation points are distributed in an ellipsoidal shell in three-dimensional space, accompanied by internal cluster points and external outliers.

[0047] To this end, this invention discloses a method and apparatus for cleaning online calibration data of a triaxial magnetometer in the assembled state of a marine carrier. Compared with the prior art, the method of this invention does not rely on the assumption of data uniformity and has strong robustness to internal clusters, external sparse noise and non-holonomic attitude sampling, and is suitable for non-uniform sampling and complex noise in the ocean.

[0048] In one embodiment, a data cleaning method for triaxial vector magnetometer calibration in in-situ calibration of marine carriers is provided. This method treats the original triaxial observation sequence of the marine magnetometer as a three-dimensional point cloud and utilizes the physical prior that "observation points should approximately fall on an ellipsoidal shell" to model the cleaning process as a thin-shell consistency recovery problem under conditions of complex noise, internal and external contamination, and under-coverage orientation. Its core objective is to eliminate outliers and internal clusters in the original data, thereby removing non-uniform sampling interference and accurately extracting effective interior points of the ellipsoidal shell. This results in an ellipsoidal shell that is as thin as possible and covers the vast majority of effective observations, providing highly pure input data for the marine magnetometer ellipsoidal calibration and ensuring the accuracy of hard iron bias, soft iron distortion, and non-orthogonal error calculations during subsequent ellipsoidal calibration.

[0049] In a preferred embodiment, the method for cleaning marine magnetometer calibration data based on geometric constraints includes the following steps: S1. Acquire triaxial magnetic field observation data from the ocean magnetometer and treat the triaxial components at each moment as point cloud sample point sets in a three-dimensional coordinate system; S2. By whitening conversion, the point cloud sample point set is mapped to the whitening space, so that the point cloud sample point set is approximately spherically distributed in the whitening space; S3. In the whitened space, estimate the instantaneous value of the spherical shell radius by directional binning; based on the spherical shell radius, determine the outer shell zone interval in combination with the noise scale; S4. Iteratively update the ellipsoidal shell parameters, and after convergence, use the sample point set that finally falls within the outer shell zone as the calibration data.

[0050] The following is an illustration through specific embodiments: (I) Whitening modeling and unified dimensions; In a specific embodiment, in S2, to explicitly utilize the composite structure presented by the actual point cloud at the geometric level, this invention uses ellipsoidal shell constraints as a consistency criterion. Let the original spatial point be... The center of the ellipsoid is The ellipsoid shape matrix is ​​a symmetric positive definite matrix. Then the ideal shell point should approximately satisfy:

[0051] in The normalized spherical shell radius in the whitened space; in some implementations, to obtain more reliable initial values ​​and avoid the impact of a few extreme outliers in the ocean data on the stability of the initial iteration, after acquiring the 3D point cloud data, a lightweight robust axial pre-filter is first performed to remove obviously out-of-bounds peaks, and then the initial ellipsoid center is obtained by the Minimum Volume Enclosing Ellipsoid (MVEE) method. With quadratic matrix and order After trace normalization, the initial shape matrix is ​​used as the initial input for algorithm iteration. Since MVEE provides an envelope rather than a thin shell, this embodiment preferably shrinks the initial radius based on MVEE, so that the initial ellipsoid is located inside the point cloud. Then, through the above-mentioned shell radius smoothing and micro-step update, it gradually expands to fit the thin shell, thereby achieving the goal of "removing a small number of outliers and retaining the vast majority of shell consistency points" before ocean magnetometer calibration.

[0052] Furthermore, in order to transform the ellipsoidal shell estimation into a spherical shell problem that is easier to perform orientation statistics and robust radius estimation, the symmetric positive definite shape matrix is... Perform Cholesky decomposition;

[0053] In the formula, Represents a symmetric positive definite shape matrix. This represents a lower triangular invertible matrix. Indicates transpose; use Perform whitening transformation on the 3D point cloud and define the whitening coordinates:

[0054] in, For the first time after albino A point cloud sample vector, For the original number A point cloud sample vector, It is the center of the ellipsoid.

[0055] Then there is An ellipsoidal shell is equivalent to a radius of in white space. The spherical shell. Figure 2 This is a rendering of the effect of mapping the original spatial ellipsoidal shell to a whitened spatial approximate spherical shell through whitening.

[0056] In this embodiment, the radius in the whitened space With direction ( This represents the i-th point cloud data vector in the whitening space. This represents the point cloud index. The direction vector of the i-th point cloud in the whitening space is used as a unified dimension for subsequent "shell radius estimation, shell band construction, direction coverage evaluation and mechanical update".

[0057] (ii) Orientation parameterization and equal area binning In one specific embodiment, in the whitening space, the direction vector for each point... We use spherical coordinates to parameterize the direction. Let... Then the azimuth angle The angle between the projection of the direction vector onto the xy-plane and the positive x-axis is used to characterize the rotational position of the direction in the horizontal plane, and its value range is... ∈ [0, 2π). This can be represented as:

[0058] Celestial Angle The angle between the direction vector and the positive z-axis is used to characterize the degree of inclination of the direction relative to the vertical axis, and its value ranges from [value missing]. ∈ [0, π), can be represented as:

[0059] Furthermore, in S3, in order to stably estimate the shell radius even with under-coverage of the ocean attitude and the presence of a single-sided outward point, the shell radius is not directly estimated. Instead of uniform division, it is... Evenly divided into In the latitude zone whitening space, at the same time, with azimuth angle Evenly divided into Each sector, thus forming a total One steering wheel.

[0060] Schematic diagram of directional binning and equal-area binning Figure 3 The left figure shows the directional distribution of sample points in the whitened space near a unit sphere and their bin assignment (different colors are used to distinguish different orientation bin numbers), reflecting the non-uniformity of directional sampling under under-coverage conditions of ocean attitude. The right figure shows the partitioning effect of the adopted equal-area directional bin grid on a unit sphere. Through this directional binning mechanism, a "one bin, one vote" directional balance constraint can be achieved in subsequent shell radius estimation and parameter iterative updates, suppressing the bias caused by differences in sampling density, and improving robustness and stability under unilateral outlier and under-coverage conditions.

[0061] (III) Robust Shell Radius Estimation For the first The set of radii of points inside the azimuth box The outer quantile is taken as the local shell radius estimate of the current orientation box. :

[0062] in, ) represents the outer quartile parameter. In this embodiment, the local shell radius is taken as the upper quartile of the outer sample, i.e. , For the first Set of sample indices within each orientation box. Local shell radius. This reflects the outer edge of the shell in that direction; the purpose of using quantiles is to remain robust to the outer point on one side. The median of the local shell radii for all valid orientation boxes is used to obtain the instantaneous estimate of the global spherical shell radius. :

[0063] In this embodiment, the combination of "bin-quantile-median" makes the shell radius estimation no longer dominated by a small number of unilateral outliers, effectively avoiding interference from unilateral outliers, and is particularly suitable for the circular trajectory and local coverage data patterns common to offshore platforms.

[0064] (iv) Externally offset shell band structure Considering the noise dispersion and disturbance in ocean magnetometer data, this embodiment constructs an outer biased shell band (balloon wall) to define candidate shell points. The steps include: Based on the radius of the spherical shell Determine the shell distance of each point cloud;

[0065] A sample set is constructed based on the shell distance; in some implementations, | smallest The points form a set ; Based on the sample set Estimating noise scale using median absolute deviation: ; Determining the half thickness of the shell band based on noise scale , This is an adjustable coefficient.

[0066] To avoid interaction with shape damping factor In this embodiment, the bias coefficients of the inner and outer shells are denoted as follows: and Based on this, the outer bias shell zone is located:

[0067] Samples falling into this interval constitute the candidate shell point set. The schematic diagram of the offset shell band formed by the outer offset shell region in this embodiment is shown below. Figure 4 The left figure illustrates the process of determining the shell boundary. The horizontal axis represents the radius of each sample point in the whitening space, and the blue bars represent the radius distribution of all samples. The red dashed line indicates the currently estimated spherical shell radius. The two orange dashed lines represent the inner and outer boundaries of the shell zone, respectively. Points falling between these two orange lines are identified as "candidate shell points" and participate in subsequent pressure calculations. The right figure shows the results of the "candidate shell point" selection; red dots represent "candidate shell points" that satisfy the shell zone region criteria, and the translucent green spheres correspond to the radii. The reference spherical shell is used as a reference. The purpose of this step is to filter out a thin layer close to the spherical shell in the 3D point cloud using the shell band determined in the left image, and use it as the input set for subsequent shape updates.

[0068] (v) Parameter update (center update + shape update) In one specific embodiment, S4 includes the following regarding parameter updates: 1. Resultant force of directional equilibrium pressure and update of ellipsoid center First, a mechanical analogy of "balloon adapting to pressure and fitting" is used. Figure 5 This core update mechanism has been visualized. Figure 5 The left-middle diagram illustrates the resultant pressure force in directional equilibrium. Colored dots represent candidate shell points in the whitened space. Red arrows indicate the weighted average vector method for the shell points within each directional box, and the direction of the overall resultant pressure force obtained by averaging across all effective directional boxes. Specifically, for candidate shell points in the whitened space, the spherical shell normal and direction are aligned, and a unit normal vector is taken. .

[0069] To maximize the contribution of samples closer to the shell centerline, the shell consistency weight for the i-th candidate shell point is defined as:

[0070] In the formula, Indicates the first The coordinate vectors of the candidate shell points in the whitening space This represents the instantaneous estimate of the spherical shell radius in whitened space. The noise scale parameter represents the direction of the whitening radius.

[0071] First, the resultant force within the box is obtained by weighted averaging of the normal vectors of all shell points within each box. ;

[0072] Then, for all valid resultant forces within the box... The average pressure resultant force is obtained by averaging. This resultant force indicates the direction of the current center's offset in the whitened space:

[0073] In the formula, The set of valid bin indexes that meet the sample number threshold.

[0074] Finally, based on the aforementioned pressure resultant force, the ellipsoid center is updated by inverse whitening mapping back to the original space:

[0075] In the formula, L represents a lower triangular invertible matrix. To represent the step size for updating the ellipsoid center, coverage-gated adaptive scaling can be used, as described later.

[0076] Figure 5 The right-middle figure shows the final center update after the combined force calculation, where the gray translucent spherical shell represents the ideal shell layer of the whitened space, and the red markers indicate the center update. This indicates the direction and magnitude of the center's movement in this iteration. As the iteration progresses, if the center gradually approaches the true value, the distribution of shell points tends to be symmetrical, and the resultant pressure force... Gradually decreasing, central update volume Converging to near 0 indicates that the central estimate is stable.

[0077] 2. Tyler-type walk matrix update and ellipsoidal shape matrix update Given that a relatively stable initial envelope is often used in the calibration of marine magnetometers, this embodiment preferably uses a small step size and applies a trust region constraint to avoid the center being pulled off due to transient disturbances in local coverage data.

[0078] A Tyler-type scatter matrix update method, combined with directional balance weights, is used to iteratively update the ellipsoid shape matrix. Let the data dimension... And calculate using the current parameters:

[0079] To achieve "one ticket per box", the first... Weight normalization within each direction bin:

[0080] Then each valid steering box satisfies = 1. Define the number of valid coverage boxes:

[0081] Therefore, the shape matrix update adopts the following method:

[0082] This represents the updated ellipsoid shape matrix. The effective weights after directional equalization ensure that each orientation box makes a nearly consistent contribution to the overall update, thereby avoiding the dominance of shape updates by data-intensive sampling in a certain heading or attitude segment.

[0083] To prevent iterative oscillations and maintain scale consistency, this embodiment further employs damping and trace normalization:

[0084] In the formula, The damping factor is updated for shape and can be adaptively scaled by coverage gating; For matrix trace operations, The average eigenvalues ​​of the shape matrix are used to normalize the shape matrix to... This eliminates scale degrees of freedom and ensures stable updates.

[0085] (vi) Insufficiently covered soft gating The aforementioned center and shape updates can be linked with orientation coverage gating in marine conditions to avoid drift caused by shape indistinguishability under under-coverage conditions. That is, the update intensity of the shape matrix is ​​automatically adjusted based on the real-time detection results of orientation coverage.

[0086] In this embodiment, to specifically adapt to the actual acquisition characteristics of insufficient attitude coverage of marine platforms, directional distribution entropy is introduced as a coverage metric, and the update intensity and shell band parameters are adaptively adjusted when coverage is insufficient. Specific steps include: Count the number of candidate shell points in the direction box And define the probability distribution. ; Calculate directional entropy ; Define the number of effective coverage boxes ; And define soft coverage metrics ; The step size and shape damping factor are updated based on the adaptive scaling ellipsoid center according to the soft coverage index; when the ocean magnetometer sampling only covers local directions. It will decrease significantly. In this embodiment, it is preferable to automatically reduce the reference center step size damping factor. With reference shape damping factor This includes scaling the ellipsoid center update step size and shape update damping factor according to coverage:

[0087]

[0088] in , for The worse the coverage, the more conservative the center and shape updates become, causing the update process to transition from "rapid shape adjustment" to "robust shell radius fitting," thereby reducing the instantaneous external point traction on the shape caused by thruster start-stop, wave disturbances, etc.

[0089] (vii) Exponential moving average of shell radius; To suppress For fluctuations under marine disturbance conditions, this embodiment uses an exponential moving average to smooth the shell radius:

[0090] in, Let represent the instantaneous spherical shell radius estimate obtained in the t-th iteration from "directional binning - outer quantile - median"; This represents the smoothed estimate of the spherical shell radius, which serves as the input for the next iteration. This represents the smoothing factor.

[0091] To improve smoothness when directional coverage is insufficient or there are many outer offset points, the following definition is made:

[0092] in As the baseline smoothing factor, This is the lower limit. The outer proportion is defined as:

[0093] in This is the outer bias coefficient. This refers to the thickness of the shell band. This represents the entire point cloud dataset. If... If there are too many outliers on the outer side, the update rate will be further reduced. .

[0094] (viii) Convergence Criterion and Final Classification.

[0095] In some implementation schemes, the above process is performed iteratively until the center displacement occurs. With the magnitude of the resultant force Simultaneously, if the value is below a preset threshold or the maximum number of iterations is reached, after iteration convergence, outlier classification is performed based on the converged ellipsoid parameters and noise scale, and the whitening radius of each sample is determined. With convergence Comparison based on thin-shell consistency criterion Determine whether an interior point is an outlier. This is an adjustable bandwidth control parameter, equivalent to a multiple of the balloon wall thickness. Candidate points that meet the above consistency criteria are marked as retained, while those that do not are marked as rejected, and a retention mask label corresponding one-to-one with the original sample is generated.

[0096] Figure 6The middle left figure shows the point cloud classification visualization of the cleaning results, where blue points are candidate shell points that are retained because they meet the thin shell consistency criterion, and orange points are outliers or intrashell clusters that are removed because they do not meet the criterion. Figure 6 The right-hand image overlays the converged ellipsoidal shell geometric model onto the retained points. It can be seen that the retained points are mainly distributed near the surface of the ellipsoidal shell, forming a continuous thin-shell structure, verifying the effectiveness of the consistency criterion in extracting valid calibration data. The final output is the cleaned set of interior points. And a mask mark (mask_keep) corresponding one-to-one with the original sample is used for subsequent ellipsoid calibration and fitting of hard iron bias, soft iron distortion and non-orthogonal error.

[0097] The dataset cleaned by this invention can effectively solve problems such as interference, attitude limitations, and insufficient data coverage in the calibration process of marine magnetometers. It can maintain excellent cleaning effect and robustness in two-dimensional, three-dimensional, and sparse data scenarios, enabling subsequent magnetometer calibration to have high accuracy and stability. At the same time, the high-quality data output by this algorithm can also provide data support for attitude detection, spatial positioning, and other fields in marine magnetic surveying, achieving accuracy improvement in multiple fields.

[0098] Based on the same inventive concept, in another embodiment, a three-axis vector magnetometer calibration data cleaning device for in-situ calibration of marine carrier assembly is also provided. This device is used to ensure the stable implementation of the algorithm under complex marine conditions from the hardware level, maximize the advantages of the software algorithm and the performance of the hardware equipment, and make the cleaning and processing of marine magnetometer calibration data a complete, implementable, and highly adaptable integrated solution.

[0099] The cleaning device provided in this embodiment is highly compatible with the cleaning method for ocean magnetometer calibration data described above, and can achieve seamless integration of cleaning and calibration. It is suitable for marine platforms such as shipborne towed magnetometers, external magnetometers, and ocean buoy magnetic measurement units.

[0100] In some implementations, the device structure is as follows: Figure 7 As shown; it integrates a sensor module and a data cleaning device; such as Figure 8 As shown, the data cleaning device connects to the marine magnetometer via the magnetometer interface and transmits data back and interacts with control commands through a communication link with an external computer, thus forming a closed-loop processing structure for calibration data: "magnetometer - data cleaning device - computer".

[0101] The sensor module is used to accurately acquire raw observation data of the three-axis magnetic field from a marine magnetometer in a marine environment, and convert the three-dimensional magnetic field data into three-dimensional point cloud data. This module integrates a marine magnetometer, a magnetic compass, and a three-axis accelerometer. The data cleaning device includes a data storage module, a central processing module, and a calibration module. It provides a dedicated and highly adaptable hardware platform for calibrating the data cleaning algorithm, achieving deep hardware-software synergy and ensuring stable execution of the cleaning process under complex marine conditions. Specifically, the data storage module stores real-time observation data collected by the sensor module, as well as intermediate data, iteration parameters, and final cleaning results from the algorithm. It also supports data backtracking and secondary processing, facilitating hardware-level management of the entire data lifecycle. The central processing module includes a preprocessing and whitening modeling unit and a Balloon-Shell cleaning unit, used to run the entire data cleaning process. The calibration module performs ellipsoidal calibration fitting: using calibration software with the cleaned interior points as input, it performs robust ellipsoid fitting and calculates calibration parameters such as hard iron bias, soft iron distortion, and non-orthogonal errors. In some implementation schemes, the external structure and interface layout of the data cleaning device are as follows: Figure 9 As shown, Figure 9 The top view, front view, and side view of the device are given to illustrate the external structural features and relative positions of the device's casing shape, magnetometer interface, power port, communication port (such as Ethernet port), and status indicator lights.

[0102] In one specific implementation scheme, the sensor module completes the acquisition and conversion of multi-sensor data and stores it in the data storage module. The central processing module calls the preprocessing and whitening modeling unit to preprocess the original 3D point cloud data to obtain the initial input for algorithm iteration. Then, it performs Cholesky decomposition on the ellipsoid shape matrix and executes whitening transformation, transforming the original spatial ellipsoidal shell cleaning problem into a whitened spatial spherical shell estimation problem. The intermediate data of the transformation is backed up to the data storage module. In this implementation scheme, the preprocessing and whitening modeling unit has high-speed matrix operation capabilities, which can quickly complete decomposition and coordinate transformation, ensuring the efficiency and security of data flow. Furthermore, the Balloon-Shell cleaning unit is called to perform data purification steps and iteratively update the ellipsoidal shell and spherical shell parameters. After convergence, based on the final ellipsoid parameters, noise scale, and normalized radius threshold, the point cloud is divided into shell points and outliers, and the finally obtained shell point data is stored in the data storage module. Among them, the Balloon-Shell cleaning unit has optimized hardware operations for non-uniform ocean sampling scenarios, which can improve the efficiency and accuracy of orientation statistics. At the same time, it can be adjusted according to real-time changes in ocean conditions. and It has equal shell parameters and supports adjusting the iteration step size. Local or remote control can be implemented to improve the algorithm's adaptability to complex working conditions.

[0103] Preferably, the device also includes a data upload / return module for transmitting data to the data analysis / display module for visualization. In some implementations, raw data, cleaning results, and calibration parameters are read from the data storage module and uploaded to an external PC via USB / serial port / Wi-Fi; alternatively, it can receive control commands from an external terminal and transmit them to the central processing module to remotely adjust algorithm parameters, completing the entire closed-loop process. The data analysis / display module includes a visualization interface for displaying the original point cloud, the distribution of interior points after cleaning, the spatial distribution of the fitted ellipsoidal shell, and the fitting effect, facilitating viewing by staff. Calibration and visualization data are synchronously stored back.

[0104] In some implementations, the overall workflow of the device is described with reference to... Figure 10 ,include: Sensor module: Collects raw three-dimensional magnetic field data such as the direction and intensity of the magnetic field in the space where the vehicle is currently located, converts the raw three-dimensional magnetic field data into three-dimensional point cloud data, and transmits it to the data storage module; Data storage module: Stores the raw data to be cleaned; stores the cleaned data to be returned. Central processing module: Uses the Balloon-Shell data cleaning algorithm based on geometric constraints and robust statistics to clean the raw data; Calibration module: Uses the cleaned data to calibrate equipment such as magnetometers for subsequent use; Data upload / return module: transmits the cleaned data to the data analysis / display module and simultaneously stores it in the data storage module to support data backtracking and secondary processing; Data Analysis / Display Module: Visualizes the cleaned data.

[0105] The marine magnetometer calibration data cleaning device provided in this embodiment deeply integrates software and hardware in its data processing flow. Using the data cleaning device as the hardware carrier and equipped with the core data cleaning algorithm, it can realize the entire process of marine magnetometer calibration, from raw data acquisition and cleaning to calibration, as well as visualization and external interaction. Each hardware module and algorithm logic is highly compatible, and each processing link is seamlessly connected, which significantly improves the efficiency and accuracy of marine magnetometer calibration. It completely solves the industry pain points of software and hardware disconnect, poor adaptability, and fragmented process in traditional technologies, greatly improves the efficiency and stability of marine magnetometer calibration data processing, and thus provides highly reliable technical support for marine magnetic measurement engineering applications.

[0106] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0107] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for cleaning online calibration data of a triaxial magnetometer in the assembled state of a marine carrier, characterized by the following steps: include: Acquire triaxial magnetic field observation data from an ocean magnetometer, and treat the triaxial components at each moment as point cloud sample point sets in a three-dimensional coordinate system; By whitening conversion, the point cloud sample point set is mapped to the whitening space, so that the point cloud sample point set is approximately spherically distributed in the whitening space; In the whitened space, the instantaneous estimate of the spherical shell radius is estimated by directional binning; based on the spherical shell radius, the outer shell zone interval is determined in combination with the noise scale; The ellipsoidal shell parameters are iteratively updated, and after convergence, the set of sample points that finally fall within the outer shell zone is used as calibration data.

2. The method as described in claim 1, characterized in that, The whitening conversion process includes: Let the ellipsoid shape matrix be a symmetric positive definite shape matrix. For the symmetric positive definite shape matrix Perform Cholesky decomposition; In the formula, This represents a lower triangular invertible matrix. Indicates transpose; use Perform whitening conversion on the point cloud sample point set: in, For the first time after albino A point cloud sample vector, For the original number A point cloud sample vector, It is the center of the ellipsoid.

3. The method as described in claim 1, characterized in that, Estimating the radius of a spherical shell by directional binning includes: In the whitened space, equal-area binning is performed based on the unit direction vector; In the formula, Indicates the first time after albinism A point cloud sample vector, Indicates the first The unit direction vector of a point cloud; For the set of radii of points within each orientation box, the outer quantile is taken as the local shell radius estimate of the current orientation box; The instantaneous estimate of the spherical shell radius is obtained by taking the median of the local shell radii of all valid boxes.

4. The method as described in claim 1, characterized in that, Based on the spherical shell radius and in conjunction with the noise scale, the outer shell zone interval is determined, including: The shell distance of each point cloud is determined based on the instantaneous estimate of the spherical shell radius, and a sample set is constructed based on the shell distance. Based on the sample set, the noise scale is estimated using the median absolute deviation; the shell band half thickness is determined based on the noise scale. The outer off-center shell zone is defined based on the shell zone half-thickness. ;in, This represents the instantaneous estimate of the radius of the spherical shell. and This indicates an adjustable bias parameter, which controls the inward contraction to eliminate internal cluster points and the outward expansion to encompass natural noise and minor disturbances.

5. The method as described in claim 1, characterized in that, Updating the ellipsoidal shell parameters includes determining the directional equilibrium of the resultant pressure force based on the shell points, and using this resultant pressure force to update the ellipsoid center: For each orientation box, the weighted average of the normal vectors of all shell points inside the box is used to obtain the resultant force inside the box; The resultant pressure force is obtained by averaging the resultant forces within all the boxes. Based on the aforementioned pressure resultant force, the ellipsoid center is updated by mapping back to the original space through inverse whitening: In the formula, Indicates the resultant force of pressure. L This represents a lower triangular invertible matrix. This indicates the step size for updating the ellipsoid center. This represents the center of the original ellipsoid.

6. The method as described in claim 1, characterized in that, Updating the ellipsoidal shell parameters involves iteratively updating the ellipsoidal shape matrix using Tyler-type M-estimators combined with directional equilibrium weights. The update method is as follows: in, This represents the updated ellipsoid shape matrix. For data dimensions, This represents the total number of effective weights after direction normalization. For an effective set of steering boxes, For the steering box index, Indicates the first The set of candidate shell point indices within each orientation box, i.e., the point cloud samples currently covered by the ellipsoidal shell. For the index of point cloud samples, For the original number A point cloud sample vector, Centered at the ellipsoid The effective weights after directional equilibrium are expressed as follows: in, For the first Shell consistency weights for individual point clouds. To be consistent with the sample Index of candidate point cloud samples located within the same orientation box. For the first Shell consistency weights for each candidate shell point and Belonging to the same direction box The expression is: In the formula, Indicates the first The coordinate vector of a point cloud sample in whitening space This represents the instantaneous estimate of the spherical shell radius in whitened space. The noise scale parameter or shell thickness scale parameter in the direction of whitening radius is used to control the decay rate of the weighting function.

7. The method as described in claim 6, characterized in that, When updating the ellipsoid shape matrix, further damping and trace normalization are applied: In the formula, This represents the ellipsoid shape matrix of the current iteration, where the ellipsoid shape matrix is ​​a symmetric positive definite matrix. This represents the updated ellipsoid shape matrix. Update the damping factor for shape; For matrix trace operations, The average eigenvalues ​​of the shape matrix are used to normalize the shape matrix to... This eliminates scale degrees of freedom and ensures stable updates.

8. The method as described in claim 1, characterized in that, Updating the ellipsoidal shell parameters includes adaptively adjusting the ellipsoid center update step size and shape update damping factor based on the directional distribution entropy. The steps include: Statistically calculate the probability distribution of shell points in each orientation box. ; Calculate the directional entropy based on the aforementioned probability distribution. ; The number of effective covered boxes is determined based on the following conditions. And define soft coverage metrics ; Based on the aforementioned soft coverage index, the adaptive scaling ellipsoid center update step size and shape update damping factor are used. In the formula, This indicates the step size for updating the ellipsoid center. This represents the shape update damping factor. The reference center step size damping factor, As the reference shape damping factor, for The worse the coverage, the more conservative the center and shape updates. This indicates a soft coverage target.

9. The method as described in claim 1, characterized in that, Updating the ellipsoidal shell parameters involves exponentially sliding the ellipsoidal shell radius to suppress estimation fluctuations: in, This represents the instantaneous estimated radius of the spherical shell in the t-th iteration. This represents the instantaneous estimated radius of the spherical shell in the t-th iteration. This represents the instantaneous estimated radius of the spherical shell in the (t-1)th iteration. This represents the smoothing factor.

10. A data cleaning device for online calibration of a triaxial magnetometer in the assembled state of a marine carrier, characterized in that, The apparatus for implementing the online calibration data cleaning method for a triaxial magnetometer in the assembled state of a marine carrier as described in any one of claims 1-9 comprises: The sensor module is used to acquire triaxial magnetic field observation data of the marine magnetometer and to treat the triaxial components at each time as a point cloud sample set in a three-dimensional coordinate system. The data cleaning device is used to map the point cloud sample point set to a whitening space through whitening transformation, and in the whitening space, to estimate the instantaneous value of the spherical shell radius by directional binning; to determine the outer shell zone interval based on the spherical shell radius and the noise scale; and to iteratively update the ellipsoidal shell parameters, and after convergence, to use the sample point set that finally falls within the outer shell zone interval as calibration data.

Images