Multi-target full-process automatic rapid three-dimensional positioning method based on single-axis magnetic measurement data

Abstract

The invention belongs to the technical field of magnetic exploration, and particularly relates to a multi-target full-process automatic rapid three-dimensional positioning method based on single-axis magnetic survey data, and the method comprises the steps: obtaining magnetic anomaly Bz component data; two-dimensional orthogonal basis function coefficients corresponding to the magnetic anomaly Bz component data are extracted, the quadratic sum of the two-dimensional orthogonal basis function coefficients is calculated, and an energy diagram of the magnetic anomaly data is obtained; delineating a magnetic source on the energy diagram; solving energy corresponding to the two-dimensional orthogonal basis function coefficient of the delineated magnetic abnormal data in the magnetic source by adopting different heights, and finding a height corresponding to an energy extreme point, namely finding a vertical coordinate; recalculating energy according to the height of the vertical coordinate; the horizontal coordinate corresponding to the energy peak value is the horizontal position of the target. And direct inversion of the nonlinear magnetic dipole model is avoided, so that the complex inversion problem is efficiently solved.

Description

Technical Field

[0001] This application belongs to the field of magnetic exploration technology, specifically relating to a multi-target, fully automated, rapid three-dimensional positioning method based on single-axis magnetic survey data. Background Technology

[0002] Magnetic anomalies are local disturbances in the geomagnetic field caused by magnetic targets, containing information about the target's location and magnetic moment. Magnetic anomaly detection (MAD) is widely used in land and sea surveillance. As a non-line-of-sight detection technology, it possesses characteristics such as low power consumption, strong concealment, strong penetration capability, and good resistance to weather effects, thus playing an important role in target localization. Spatial localization of magnetic targets typically involves six-dimensional parameter inversion based on a nonlinear magnetic dipole model. This process usually requires time-consuming iterative calculations or redundant sensor measurements. Using complex sensor systems or increasing the number of sensors not only significantly increases the cost of the magnetometer system but also increases operational complexity and introduces additional noise interference. Currently, existing localization methods are mainly divided into three categories: direct methods, iterative optimization methods, and step-by-step methods. Direct methods suffer from linearization errors and instabilities caused by the calculation of higher-order potential field derivatives; iterative optimization methods map anomalies to dipole parameters through heuristic algorithms, but in multi-target scenarios, their computational cost is high, they are sensitive to noise, and their scalability is poor; step-by-step methods usually combine image processing techniques with the Euler method for inversion, but they inherit the inherent defects of the Euler method. Among other step-by-step methods, the typical two-dimensional orthogonal basis method has two main limitations: First, it cannot estimate the target's vertical position (z0), and it relies on a pre-defined... The method uses a fixed sliding window, but in practice, z0 is often unknown. Inaccurate presets lead to reduced method performance, while the use of a fixed sliding window also results in performance degradation, information loss, and increased computation. Multi-objective scenarios present additional challenges, as most existing methods require predefined target numbers or determination through iteration, and using a fixed threshold may fail to capture low-energy targets. Although AI-based magnetic anomaly delineation methods alleviate the limitations of multi-objective scenarios to some extent, they still face many challenges. First, due to the influence of multi-class ambiguity and low signal-to-noise ratio, existing magnetic source delineation methods on magnetic anomaly maps perform poorly in terms of accuracy and robustness. Second, existing models face significant difficulties in capturing small and morphologically diverse magnetic source features; simultaneously, in the absence of large-scale magnetic sources, model parameters often appear redundant, necessitating improvements in parameter and inference efficiency. Summary of the Invention

[0003] In view of the problems mentioned above and / or existing technologies, this invention proposes a multi-target, fully automated, rapid three-dimensional positioning method based on single-axis magnetic measurement data.

[0004] This application is implemented as follows: A multi-target, fully automated, rapid 3D localization method based on single-axis magnetic measurement data, comprising: Obtain magnetic anomaly B z Component data; Extracting magnetic anomaly B z The energy map of the magnetic anomaly data is obtained by calculating the sum of squares of the two-dimensional orthogonal basis function coefficients corresponding to the component data; Delineate the magnetic source on the energy diagram; By using different heights, the energy corresponding to the two-dimensional orthogonal basis function coefficients of the magnetic anomaly data within the delineated magnetic source is calculated, and the height corresponding to the energy extremum point is found, i.e., the vertical coordinate is found; Recalculate the energy based on the height of the vertical coordinate; The horizontal coordinate corresponding to the energy peak is the horizontal position of the target.

[0005] Furthermore, the extraction of magnetic anomaly B z The coefficients of the two-dimensional orthogonal basis functions corresponding to the component data include: The two-dimensional orthogonal basis functions are assigned values ​​using a preset height, and the two-dimensional orthogonal basis functions are truncated using half the width of a sliding window; Using assigned and window-truncated two-dimensional orthogonal basis functions and magnetic anomaly B z The component data are convolved to obtain magnetic anomaly B. z The coefficients of the two-dimensional orthogonal basis functions corresponding to the component data.

[0006] Furthermore, the sliding window half-width is dynamically updated, the dynamic update including: vertical positioning using a preset sliding window half-width; if the estimation result satisfies... Then update HW to HW is the half width of the sliding window. The coordinates are vertical.

[0007] Furthermore, the step of delineating the magnetic source on the energy map includes: performing target detection on the energy map using an improved YOLOv5s algorithm, wherein the improved YOLOv5s algorithm employs...

[0008] Furthermore, the two-dimensional orthogonal basis functions are expressed as: , Where, φ Bz1 , φ Bz2 and φ Bz3 Magnetic anomaly B z The three basis functions of the component data, w = x / z est v = y / z est , zest f is the preset value for the target vertical position z0; Bz1 f Bz2 and f Bz3 For three traditional B z Two-dimensional orthogonal basis functions of components; g Bz1 g Bz2 and g Bz3 These are the three corresponding improvements to B. z The components are two-dimensional orthogonal basis functions, where x and y represent the horizontal positions.

[0009] Furthermore, the coefficients of the two-dimensional orthogonal basis functions are expressed as: , Among them, c Bzl For B z The projection coefficients of the components in the two-dimensional orthogonal basis function space. , The coordinates of the current processing point. and The discrete integration limit is represented by the sliding half-window width.

[0010] Compared with the prior art, the advantages of this application are as follows: This invention enables fully automated multi-target 3D localization using only Z-axis magnetic measurement data, requiring a simple structure and low cost. Its features include high precision, high efficiency, and full automation. Employing convolution for 3D localization avoids direct inversion of nonlinear magnetic dipole models, thus efficiently solving complex inversion problems and possessing the potential for real-time, high-precision multi-target tracking based on distributed magnetic networks. The proposed energy-domain magnetic source delineation exhibits higher performance than traditional magnetic map delineation. High-performance magnetic source delineation is achieved using the IYOLOv5s model, which integrates CoTC3, SPD-C3, and P5 large target head removal techniques, enhancing the detection capability of small targets while maintaining lightweight and high efficiency. The proposed I2DOBF theory extends the traditional 2DOBF theory, not only achieving theoretically sound vertical localization but also significantly improving horizontal localization accuracy. Its core principles are directly applicable to total magnetic fields, gradient fields, and tensor fields, laying the foundation for wider applications of OBF decomposition. Furthermore, this application provides theoretical support for multi-target tracking using distributed sensor networks, and has the potential to achieve continuous and accurate monitoring of complex nonlinear motion of multiple targets. Attached Figure Description

[0011] Figure 1 A flowchart of the positioning method provided in the embodiments of this application; Figure 2 The embodiment provided in this application provides when z estThe surface and contour plots (a), (b), and (c) of I2DOBFs at =1m and 3m correspond to z est =1m when g Bz1 g Bz2 and g Bz3 Surface and contour maps, (d), (e), and (f) correspond to z. est =3m when g Bz1 g Bz2 and g Bz3 Surface and contour maps; Figure 3 An improved YOLOv5s-7.0 architecture is provided for embodiments of this application; Figure 4 For the detection results of different input feature maps provided in the embodiments of this application, YOLOv5s in Case 1 and Case 2 B z The results in the figure, (a) correspond to Case 1, and (b) correspond to Case 2; IYOLOv5s in Case 1 and Case 2 E c,z The results in the figure are shown. (c) corresponds to Case 1 and (d) corresponds to Case 2. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0013] This invention proposes a multi-target, fully automated, rapid 3D localization method based on uniaxial magnetic measurement data, specifically including two key stages: magnetic source delineation and target 3D localization. The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Please see Figure 1 This application proposes a multi-target, fully automated, and rapid 3D localization method based on single-axis magnetic measurement data, comprising the following steps: S1 acquires magnetic anomaly B z Component data; Z-axis magnetic measurement data and position data were acquired. A strapdown magnetic anomaly detection (MAD) device, consisting of a fluxgate sensor and an inertial navigation system (INS), was used as an example to acquire magnetic vector, attitude, and latitude / longitude data for multiple survey lines within the survey area. During data acquisition, the fluxgate sensor moved at a constant speed along the survey line and collected data continuously at a sampling frequency of 10 Hz.

[0015] The acquired magnetic vector data first undergoes sensor error correction, coordinate system transformation, misalignment error correction, and DC removal processing to obtain magnetic anomaly vector data in the geographic coordinate system. Next, the processed magnetic anomaly vector data is gridded and interpolated in two dimensions. Linear interpolation with an interpolation interval of 0.1 m is used as an example to obtain spatially uniform magnetic anomaly vector data with location information (i.e., three-axis magnetic anomaly component data). Theoretical analysis and experimental verification show that the Z-axis component performs optimally in terms of positioning accuracy; therefore, component B is selected as an example. z The components are used for 3D positioning. Other components can also be used to achieve a similar 3D positioning process.

[0016] S2 extracts magnetic anomaly B z The energy map of the magnetic anomaly data is obtained by calculating the sum of squares of the two-dimensional orthogonal basis function coefficients corresponding to the component data; An improved two-dimensional orthogonal basis function (I2DOBFs) was used to analyze magnetic anomaly B. z Convolution operation is performed on the component data to extract magnetic anomaly B. z The coefficients of the two-dimensional orthogonal basis functions corresponding to the components are then calculated. The sum of squares of these coefficients is then calculated to obtain their energy characterization. The specific form of the improved two-dimensional orthogonal basis functions is shown below: , Where, φ Bz1 , φ Bz2 and φ Bz3 Magnetic anomaly B z The three basis functions of the component data, w = x / z est v = y / z est , z est f is the preset value for the target vertical position z0; Bz1 f Bz2 and f Bz3 For three traditional B z Two-dimensional orthogonal basis functions of components; g Bz1 g Bz2 and g Bz3 These are the three corresponding improvements to B. zThe component-based two-dimensional orthogonal basis functions, where x and y represent the horizontal positions, satisfy orthogonality and normality.

[0017] The specific mathematical conditions and proof are as follows: ; ; ; in, Let Kronecker function be used.

[0018] Select the appropriate sliding window half-width (HW) and preset height (z). est (Example, the default settings are HW=2.5 m and z) est = 1 m. Based on this preset height z est The two-dimensional orthogonal basis functions are assigned values, and then the window is truncated using the half-width of the sliding window. Subsequently, the assigned and truncated two-dimensional orthogonal basis functions are compared with the magnetic anomaly B. z The component data are convolved to obtain the corresponding coefficients. These coefficients can be considered as magnetic anomaly B. z The projection values ​​of the components in the two-dimensional orthogonal basis function space. Specifically, the two-dimensional orthogonal basis functions and B are realized by the following formula. z Convolution operation of components to obtain B z Projection coefficients of the components in the two-dimensional orthogonal basis function space: , Among them, c Bzl For B z The projection coefficients of the components in I2DOBF space. , The coordinates of the current processing point. and The discrete integration limit is represented by the sliding half-window width.

[0019] Next, calculate B using the following formula. z The sum of squares of the projection coefficients of the components in the two-dimensional orthogonal basis function space is used to obtain the magnetic anomaly B. z Energy characterization of components , in, Represented as B z A quantity of energy.

[0020] S3 delineates the magnetic source on the energy diagram; To achieve precise delineation of the magnetic source, in one example of this application, target detection is performed on the energy map instead of the traditional magnetic map. An improved YOLOv5s (IYOLOv5s) is used to obtain the anchor frame of the magnetic source, thereby determining the number of targets and the data range corresponding to each target. For example... Figure 4 As shown, IYOLOv5s integrates a CoTC3 module after the SPPF module in the backbone network. The CoTC3 module introduces a sophisticated attention mechanism that fully leverages the contextual information between input keys, thereby enhancing visual representation learning and detection accuracy. Simultaneously, the large object detection head P5 (20×20×255) and its related components are removed to reduce redundant computation and improve inference efficiency. Furthermore, the SPD-C3 module is added to more effectively preserve discriminative feature information and promote deep representation learning.

[0021] YOLOv5 is a widely used single-stage detection algorithm, renowned for its efficiency and accuracy. This invention employs YOLOv5s-7.0 due to its lightweight architecture and fast inference speed. However, standard YOLOv5s is not optimal for magnetic source delineation tasks. Specifically, the extremely small and diverse shapes of magnetic sources challenge its ability to capture subtle features, while the lack of large-scale magnetic sources in the magnetic map suggests that redundant parameters can be pruned to further improve the model's compactness and inference efficiency.

[0022] To address the limitations of existing technologies, this application innovatively proposes magnetic source delineation within the energy map, rather than the traditional direct detection on the magnetic map, and proposes an improved YOLOv5s architecture (IYOLOv5s). This novel energy domain detection technique significantly improves detection performance by reducing multi-class ambiguity and enhancing noise robustness. Figure 4 As shown, in this architecture, the SPPF module in the backbone network is followed by the CoTC3 module, which introduces a sophisticated attention mechanism that fully leverages the contextual information between input keys, thereby enhancing visual representation learning and detection accuracy. Simultaneously, to reduce redundant computation and improve inference efficiency, the large target detection head P5 (20×20×255) and its related components have been removed. Here, "large" and "small" targets refer to the size of the magnetic source in the magnetogram, not their actual physical size. Furthermore, the SPD-C3 module is added to preserve discriminative feature information and promote more effective representation learning. This module is specifically designed to capture spherical features, making it particularly suitable for energy map detection, as energy distributions are typically spherical.

[0023] Magnetic source delineation mainly utilizes IYOLOv5s to detect targets on the energy map, thereby obtaining the anchor frame of the magnetic source, and thus determining the number of targets and the range of magnetic anomaly data corresponding to each target.

[0024] In multi-target scenarios, IYOLOv5s is used to automatically identify and delineate all targets, enabling automatic segmentation of target magnetic anomaly data. This process ensures that the segmented data retains information from only a single magnetic target as much as possible, thereby minimizing interference from other targets. Subsequently, a 3D localization process is sequentially performed on each target until the 3D localization of all magnetic targets is completed.

[0025] Automatically identifying and delineating all targets using IYOLOv5s requires training the IYOLOv5s system. Existing methods directly delineate magnetic sources from magnetic anomaly maps, resulting in low delineation accuracy due to the existence of multiple anomaly categories and the low signal-to-noise ratio of the input data. Multiple anomaly categories stem from the bipolar nature of dipole anomalies, as well as single positive or negative anomalies caused by specific magnetic moment directions or noise. Multi-class detection not only reduces detection accuracy but also significantly increases the complexity and workload of labeling.

[0026] To overcome these challenges, this application directly delineates magnetic sources on the energy map. Compared to magnetic anomaly data, energy data has a higher signal-to-noise ratio and can convert positive, negative, and bipolar anomalies into unipolar energies. This reduces the number of magnetic source categories, which not only facilitates labeling but also simplifies the model architecture and improves inference efficiency. Simultaneously, the higher signal-to-noise ratio and elimination of category ambiguity further improve the accuracy of target detection.

[0027] In magnetic exploration, obtaining large-scale training datasets is extremely challenging due to the high cost of measurement. To address this issue, this application uses synthetic datasets to simulate realistic magnetic anomalies. Firstly, several B-type datasets are generated based on a magnetic dipole model. z Magnetic anomaly maps, each of which can contain multiple targets, are then generated into B... z Gaussian white noise with a signal-to-noise ratio (SNR) ranging from -20 to 0 dB is added to the anomaly to simulate complex background interference. Next, the steps are as follows: The corresponding E is calculated. c,z Energy diagram. Demonstrated using default parameters HW = 2.5 m, z. est = 1 m. Both datasets share the same label set. For example, 1000 pairs of data samples are generated, B z The dataset was divided into a training set (800 images) and a validation set (200 images) in an 8:2 ratio. c,z The dataset also uses the same partitioning method to maintain pair consistency.

[0028] S4 uses different heights to find the energy corresponding to the two-dimensional orthogonal basis function coefficients of the magnetic anomaly data within the delineated magnetic source, and finds the height corresponding to the energy extremum point, that is, finds the vertical coordinate; Traditional 2DOBFs cannot be directly used to evaluate the vertical coordinate z0 because there is currently a lack of mathematical evaluation theory for the vertical coordinate z0. However, the proposed I2DOBF theory can quantify not only the height z... est The degree of closeness to the vertical coordinate z0 allows for an effective evaluation of the vertical coordinate z0 both theoretically and experimentally. The theoretical proof and evaluation process are as follows: The square of the theoretical 2DOBF coefficient is obtained by the following formula.

[0029] , The vertical coordinate z0 of the target is unknown prior to the integral calculation, therefore an estimated height z needs to be pre-set during the calculation. est When the height z est The closer the obtained coefficients are to the true vertical coordinate z0, the closer they are to the true coefficients, and the higher the horizontal positioning accuracy. When using traditional 2DOBFs for coefficient calculation, its upper bound is determined by the preset height z. est , and only if z est Accurate coefficients can only be obtained when z = 0. However, there is a lack of effective measures to quantify z. est Traditional methods are almost unable to accurately determine the proximity between z0 and z0. As shown in the following equation, , The coefficients are obtained using the improved two-dimensional orthogonal basis function (I2DOBF) of this application, which is related to z. est An irrelevant upper bound.

[0030] , If and only if , i.e. z est = z0 (where λ is a constant), the inequality holds true. Therefore, by setting different preset heights z... est Substitute I2DOBFs and calculate the value on the left side of the equation to evaluate the vertical coordinate z0: take the preset height z of the maximum value. est This refers to the vertical coordinate z0. Therefore, the improved I2DOBFs proposed in this application can be used to accurately evaluate the vertical coordinate z0, thereby achieving the vertical positioning of the target. Specifically, this is achieved by using different preset heights z0. est The values ​​are assigned to I2DOBFs, the corresponding energies are calculated, and the maximum energy value E is recorded. c,z,max Repeat this process to obtain a description of the preset height z. est With E c,z,max The curve represents the relationship. This curve exhibits convex function characteristics, and its extreme points correspond to a preset height z. est The vertical coordinate z0 of the target is used to achieve vertical positioning by finding the extreme point.

[0031] In essence, estimating the vertical coordinate z0 is an extremum search problem for a convex function. Traditional methods, such as interpolation and gradient ascent, have significant limitations. Interpolation methods have limited accuracy or high computational costs, while gradient ascent methods struggle to select an appropriate step size; an excessively large step size may hinder convergence, while an excessively small step size increases computational complexity. To achieve fast and accurate vertical positioning, this application employs a ternary search to efficiently estimate the vertical coordinate z0.

[0032] S5 recalculates the energy based on the vertical positioning height; The horizontal coordinate corresponding to the S6 energy peak is the horizontal position of the target.

[0033] When the center of the sliding window is directly above the magnetic dipole, the correlation between the orthogonal basis and the magnetic anomaly reaches its maximum, resulting in an energy peak at that location. By obtaining the horizontal coordinates of this energy peak, the target's horizontal positioning can be achieved. Specifically, the vertical coordinate z0 value from the vertical positioning evaluation is input into I2DOBFs, and the energy is recalculated. The horizontal coordinates (x0, y0) corresponding to the energy peak represent the target's horizontal position, thus completing the target's horizontal positioning and ultimately achieving 3D target localization. Based on the spatial equivalence of convolution, multi-target horizontal positioning can be achieved, with the horizontal coordinates of each energy peak corresponding to the horizontal position of its respective target.

[0034] In one embodiment, the sliding window half-width is dynamically updated, the dynamic update including: vertical positioning using a preset sliding window half-width; if the estimation result satisfies... Then update HW to HW is the half width of the sliding window. The coordinates are vertical.

[0035] As the half-width of the sliding window increases, the completeness of the orthogonal basis increases, the noise suppression capability becomes stronger, and the positioning accuracy becomes higher, but this is accompanied by an increase in computational cost. To balance accuracy and computational cost, experimental analysis shows that choosing |w|, |v| = 2 is reasonable. Considering the possibility of a large vertical coordinate z0, the half-width of the sliding window should be appropriately increased to meet the completeness requirements of the orthogonal basis. Although the true z0 is unknown prior, a preliminary estimate of the vertical coordinate z0 can be obtained using the default half-width of the sliding window, and a suitable half-width of the sliding window can be reconfigured based on this estimate before formal positioning. Specifically, vertical positioning is first performed using the default half-width of the sliding window; if the estimation result makes |HW / z0| < 2, then HW is updated to 2z0. This adaptive strategy ensures that the orthogonal basis maintains good completeness, thereby fully leveraging the performance of the method.

[0036] To verify the validity and practical value of this application, a demonstrative field experiment was conducted.

[0037] Field experiments: To verify the feasibility and scalability of this application in real-world scenarios, a field multi-target localization experiment was conducted using a self-made MAD (Magnetic Anomaly) device. This device included a three-axis fluxgate magnetometer (Bartington Mag-03), a Spectramag-6 data acquisition system, and an INS (Instrument System) integrating RTK technology. The INS achieved centimeter-level positioning accuracy through tight coupling of ground station and rover data. The magnetic target consisted of four iron tubes, each approximately 30 cm long, 10 cm in outer diameter, and 3 mm thick. Two MAD experiments were conducted (Case 1 and Case 2), and magnetic vector data containing anomaly information for these four targets were collected in each case. In both cases, the targets were randomly placed in a 20 × 20 m... 2 The specific locations within the survey area are shown in Table 1. The MAD device traveled along a serpentine path at a height of 1.25 m and a speed of 1 m / s. Both the data acquisition system and the INS operated at a sampling frequency of 10 Hz, with a spatial sampling interval of approximately 0.1 m.

[0038] After correcting for sensor and misalignment errors in the two sets of collected magnetic field vector data, they were transformed to a geographic coordinate system using Euler angles derived from INS. Subsequently, DC component removal and two-dimensional linear interpolation were performed to obtain gridded magnetic anomaly vector data. Next, I2DOBFs were used to convert the processed data into energy data, and z-axis values ​​were set. est = 1 m and HW = 2.5 m. Finally, for B z Component data and E c,z The energy data was normalized, and IYOLOv5s was used to perform target detection on the normalized Bz and Ec,z maps. The results are as follows: Figure 4 As shown, B z The signal-to-noise ratio of the image is low; three targets are vaguely visible, while another target is almost completely obscured by noise. In contrast, E c,z The signal-to-noise ratio of the image is significantly high, and all four targets are clearly visible, verifying the excellent noise suppression capability of the I2DOBF method. The detection results show that in B... z In the figure, Scheme 1 generated both false positives (FP) and false negatives (FN), detecting 4 targets with precision P = 3 / 4 and recall R = 3 / 4; Scheme 2 detected 3 targets with 1 false negative, achieving precision P = 3 / 3 and recall R = 3 / 4. In contrast, in E... c,zIn the figure, Schemes 1 and 2 successfully detected all targets with precision P = 4 / 4 and recall R = 4 / 4, and the confidence of the targets was improved. These results verify the superiority of the energy domain-based magnetic source delineation method and demonstrate the strong generalization ability of IYOLOv5s in real-world scenarios, successfully detecting 4 real targets even when the training dataset contains no more than 3 simulated targets.

[0039] Based on the magnetic anomaly B of each target defined in the energy domain by IYOLOv5s z The component data were used for 3D localization, achieving accurate 3D localization of multiple targets. The results are shown in Table 1. In both cases, almost all targets achieved near-centimeter-level localization accuracy, fully verifying the accuracy of the method proposed in this application.

[0040] Table 1 Field test parameters and location results .

[0041] The multi-target 3D localization process is fully automated, requiring no manual intervention, and has achieved satisfactory results, validating the robustness, generalization ability, and high accuracy of this application. Furthermore, this application demonstrates excellent inversion efficiency, processing each target in approximately 0.5 seconds. These advantages highlight the significant potential of this application in automated industrial inspection, particularly suitable for applications requiring high real-time performance and accuracy.

[0042] This invention proposes a multi-target, fully automated, rapid 3D localization method based on uniaxial magnetic measurement data, covering two core stages: magnetic source delineation and target 3D localization. First, magnetic anomaly B is analyzed. z The data for the components were acquired and preprocessed, and then improved two-dimensional orthogonal basis functions (I2DOBFs) were used to analyze B. z The components are processed to obtain the corresponding energy data. Next, a trained, improved YOLOv5s model is used to detect targets on the energy map, thereby determining the number of targets and the magnetic anomaly data range corresponding to each target. Subsequently, based on the magnetic anomaly data range of each target, three-dimensional localization is performed one by one. Specifically, a ternary search and adaptive sliding window update strategy is used to efficiently estimate the vertical position z0 of the target, then this z0 value is assigned and the energy is recalculated to determine the horizontal coordinates corresponding to the energy peak, thus achieving horizontal localization of the target.

[0043] Compared to existing technologies, this invention achieves fully automated multi-target 3D localization using only single-axis magnetic measurement data. It features a simple structure, low cost, high precision, high efficiency, and full automation. 3D localization is achieved through convolution operations, avoiding direct inversion of traditional nonlinear magnetic dipole models, thus efficiently solving complex inversion problems. Compared to traditional magnetic mapping methods, the proposed energy-domain magnetic source mapping exhibits superior performance. Furthermore, the improved YOLOv5s enhances the detection capability for small targets while maintaining lightweight and high efficiency. The proposed I2DOBF theory extends the traditional 2DOBF theory, achieving not only theoretically sound vertical localization but also significantly improving horizontal localization accuracy.

[0044] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-target, fully automated, rapid 3D positioning method based on single-axis magnetic measurement data, characterized in that, The method includes: Obtain magnetic anomaly B z Component data; Extracting magnetic anomaly B z The energy map of the magnetic anomaly data is obtained by calculating the sum of squares of the two-dimensional orthogonal basis function coefficients corresponding to the component data; Delineate the magnetic source on the energy diagram; By using different heights, the energy corresponding to the two-dimensional orthogonal basis function coefficients of the magnetic anomaly data within the delineated magnetic source is calculated, and the height corresponding to the energy extremum point is found, i.e., the vertical coordinate is found; Recalculate the energy based on the height of the vertical coordinate; The horizontal coordinate corresponding to the energy peak is the horizontal position of the target.

2. The method according to claim 1, characterized in that, The extraction of magnetic anomaly B z The coefficients of the two-dimensional orthogonal basis functions corresponding to the component data include: The two-dimensional orthogonal basis functions are assigned values ​​using a preset height, and the two-dimensional orthogonal basis functions are truncated using half the width of a sliding window; Using assigned and window-truncated two-dimensional orthogonal basis functions and magnetic anomaly B z The component data are convolved to obtain magnetic anomaly B. z The coefficients of the two-dimensional orthogonal basis functions corresponding to the component data.

3. The method according to claim 2, characterized in that, The sliding window half-width is dynamically updated, and the dynamic update includes: vertical positioning using a preset sliding window half-width; if the estimation result satisfies... Then update HW to HW is the half width of the sliding window. The coordinates are vertical.

4. The method according to claim 2, characterized in that, The step of delineating the magnetic source on the energy map includes: using an improved YOLOv5s algorithm to perform target detection on the energy map, wherein the improved YOLOv5s algorithm employs...

5. The method according to claim 2, characterized in that, Two-dimensional orthogonal basis functions are represented as: ， Where, φ Bz1 , φ Bz2 and φ Bz3 Magnetic anomaly B z The three basis functions of the component data, w = x / z est v = y / z est , z est The preset value for the target vertical position z0; f Bz1 f Bz2 and f Bz3 For three traditional B z Two-dimensional orthogonal basis functions of components; g Bz1 g Bz2 and g Bz3 These are the three corresponding improvements to B. z The components are two-dimensional orthogonal basis functions, where x and y represent the horizontal positions.

6. The method according to claim 5, characterized in that, The coefficients of two-dimensional orthogonal basis functions are expressed as follows: , Among them, c Bzl For B z The projection coefficients of the components in the two-dimensional orthogonal basis function space. , The coordinates of the current processing point. and The discrete integration limit is represented by the sliding half-window width.

Images