Unexploded ordnance classification and attitude calculation method based on planar array near-field detection

Abstract

The invention discloses an unexploded ordnance classification and attitude resolving method based on planar array near-field detection. The resolving capability of magnetic vector field data on unexploded ordnance types and attitudes is remarkably improved. Fusion learning is carried out on multi-source data of the planar magnetic flux gate array by using a deep neural network model, so that the problem of failure of a magnetic dipole theoretical model under a near-field detection condition is solved, and rapid and accurate calculation of the type and attitude of an unexploded ordnance is realized; by combining the finite element high-fidelity simulation data and the live-action experiment data, the unexploded ordnance detection object-driven database is established, the generation cost of the data required by the neural network model is reduced, and the training efficiency of the neural network model is improved.

Description

Technical Field

[0001] This application relates to the technical field of unexploded ordnance magnetic detection, and particularly to a method for classifying unexploded ordnance and solving its attitude based on near-field detection with a planar array. Background Technique

[0002] Unexploded ordnance left over from war conflicts and live ammunition exercises has a high explosion risk and good concealment. It can pollute the soil and underground water resources, and at the same time threaten the normal life of local residents. The ferromagnetic materials in unexploded ordnance generate magnetic anomalies under the magnetization of the geomagnetic field, providing a highly sensitive and non-active detection basis for the detection of unexploded ordnance.

[0003] Different types of unexploded ordnance have different detonation mechanisms and explosion characteristics. At the same time, the buried attitude of unexploded ordnance underground will also affect its stability and potential trigger points. Therefore, accurately identifying the type of unexploded ordnance and analyzing its buried attitude is of great significance for formulating a reasonable unexploded ordnance disposal strategy. On the one hand, vectorized and tensorized magnetic anomaly data can improve the analytical potential for the category and attitude of unexploded ordnance; on the other hand, the intensity of the magnetic anomaly signal after the magnetization of unexploded ordnance decreases rapidly with the increase of the detection distance, reducing the ability of the detection data to solve the type and attitude of unexploded ordnance.

[0004] The magnetic dipole theory model requires that the distance between the fluxgate array and the unexploded ordnance target is more than 2.5 times the target size. However, when conducting near-field detection of the unexploded ordnance target, the detection distance usually cannot meet its usage conditions. Therefore, when conducting near-field detection, the traditional magnetic method inversion method of the magnetic dipole theory model has a large error and is no longer applicable, and a new detection result calculation method needs to be established. Summary of the Invention

[0005] The purpose of the present invention is to propose a method for classifying unexploded ordnance and solving its attitude based on near-field detection with a planar array; by using a nested sub-array structure with a planar fluxgate array, high-order gradient data of magnetic anomalies are obtained; at the same time, combined with machine learning and a numerical-physics-driven training strategy, the fusion learning of the near-field magnetic vector field data of the fluxgate array and the rapid and accurate calculation of the type and attitude of the unexploded ordnance target are realized.

[0006] The technical solution adopted is as follows: A method for classifying unexploded ordnance and solving its attitude based on near-field detection with a planar array, the method for classifying unexploded ordnance and solving its attitude is specifically: Based on machine learning, an unexploded ordnance target classification model and an unexploded ordnance attitude calculation model are respectively constructed and trained. Among them, the input of the unexploded ordnance target classification model is the detection criterion including the magnetic vector field, the first-order difference of the magnetic vector field, and the second-order difference data of the magnetic vector field, and the output is the unexploded ordnance target classification result; the input of the unexploded ordnance attitude calculation model is the detection criterion and the corresponding unexploded ordnance target classification result, and the output is the unexploded ordnance attitude calculation result; Using a planar fluxgate array, collect the magnetic vector field data at the measurement point, and calculate the corresponding first-order finite difference and second-order finite difference data of the magnetic vector field to obtain the detection criterion at the measurement point; Input the detection criterion at the measurement point into the unexploded ordnance target classification model to obtain the unexploded ordnance target classification result at the measurement point; Input the detection criterion at the measurement point and the unexploded ordnance target classification result into the unexploded ordnance attitude calculation model to obtain the unexploded ordnance attitude calculation result at the measurement point.

[0007] For further optimization of the technical solution of the present invention, the planar fluxgate array is a nested planar fluxgate array composed of three square sub-array structures. The planar fluxgate array includes 8 three-axis fluxgate sensors, and the 8 three-axis fluxgate sensors are respectively numbered 1-8. The 4 three-axis fluxgate sensors numbered 1, 2, 3, and 4 form a square sub-array structure A1, the 4 three-axis fluxgate sensors numbered 3, 4, 5, and 6 form a square sub-array structure A2, and the 4 three-axis fluxgate sensors numbered 5, 6, 7, and 8 form a square sub-array structure A3.

[0008] For further optimization of the technical solution of the present invention, the first-order finite difference data of the magnetic vector field includes the first-order finite difference data of the magnetic vector field at the center of A1, the first-order finite difference data of the magnetic vector field at the center of A2, and the first-order finite difference data of the magnetic vector field at the center of A3; the specific calculation is as follows: ; represents the first-order finite difference data of the magnetic vector field at the center of the sub-array A1 calculated based on the magnetic vector field data of the fluxgate sensors numbered 1-4; represents the magnetic vector field data of the fluxgate sensor No. 1 in the x direction, represents the magnetic vector field data of the fluxgate sensor No. 1 in the y direction, represents the magnetic vector field data of the fluxgate sensor No. 1 in the z direction; represents the magnetic vector field data of the fluxgate sensor No. 2 in the x direction, represents the magnetic vector field data of the fluxgate sensor No. 2 in the y direction, represents the magnetic vector field data of the fluxgate sensor No. 2 in the z direction; Represents the magnetic vector field data of the No. 3 fluxgate sensor in the x direction, Represents the magnetic vector field data of the No. 3 fluxgate sensor in the y direction, Represents the magnetic vector field data of the No. 3 fluxgate sensor in the z direction; Represents the magnetic vector field data of the No. 4 fluxgate sensor in the x direction, Represents the magnetic vector field data of the No. 4 fluxgate sensor in the y direction, Represents the magnetic vector field data of the No. 4 fluxgate sensor in the z direction; d represents the spatial interval distance between two adjacent fluxgate sensors; ; Represents the first-order finite difference data of the magnetic vector field at the center of sub-array A2 calculated based on the magnetic vector field data of the No. 3 - 6 fluxgate sensors; Represents the magnetic vector field data of the No. 5 fluxgate sensor in the x direction, Represents the magnetic vector field data of the No. 5 fluxgate sensor in the y direction, Represents the magnetic vector field data of the No. 5 fluxgate sensor in the z direction; Represents the magnetic vector field data of the No. 6 fluxgate sensor in the x direction, Represents the magnetic vector field data of the No. 6 fluxgate sensor in the y direction, Represents the magnetic vector field data of the No. 6 fluxgate sensor in the z direction; ; Represents the first-order finite difference data of the magnetic vector field at the center of sub-array A3 calculated based on the magnetic vector field data of the No. 5 - 8 fluxgate sensors; Represents the magnetic vector field data of the No. 7 fluxgate sensor in the x direction, Represents the magnetic vector field data of the No. 7 fluxgate sensor in the y direction, Represents the magnetic vector field data of the No. 7 fluxgate sensor in the z direction; Represents the magnetic vector field data of the No. 8 fluxgate sensor in the x direction, Represents the magnetic vector field data of the No. 8 fluxgate sensor in the y direction, Represents the magnetic vector field data of the No. 8 fluxgate sensor in the z direction.

[0009] For further optimization of the technical solution of the present invention, the calculation of the second-order finite difference data of the magnetic vector field is specifically as follows: ; In the formula, represents the second-order finite difference data of the magnetic vector at the center of the planar array calculated from the sub-arrays A1, A2, and A2.

[0010] For further optimization of the technical solution of the present invention, the unexploded ordnance target classification model and the unexploded ordnance attitude calculation model are trained using the digital-physical-driven unexploded ordnance detection database. The digital-physical-driven unexploded ordnance detection database includes magnetic vector field data obtained from finite element simulations and field experiments, magnetic vector field first-order differences and second-order differences data calculated from the magnetic vector field data obtained from finite element simulations and field experiments, unexploded ordnance target classification results, and the angles α, β, and γ between the unexploded ordnance target and the three axes.

[0011] For further optimization of the technical solution of the present invention, the unexploded ordnance target classification model adopts a classification neural network.

[0012] For further optimization of the technical solution of the present invention, the unexploded ordnance attitude calculation model adopts a regression neural network.

[0013] For further optimization of the technical solution of the present invention, interpolation is used to perform interpolation processing on the magnetic vector field, the first-order difference of the magnetic vector field, and the second-order difference data of the magnetic vector field.

[0014] An electronic device includes a processor and a memory. A computer program capable of running on the processor is stored on the memory. When the processor runs the computer program, it executes the steps of any one of the above methods.

[0015] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the steps of any one of the above methods.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention's method for classifying unexploded ordnance targets and calculating attitudes based on the near field of a planar fluxgate array significantly improves the ability to calculate the type and attitude of unexploded ordnance from magnetic vector field data.

[0017] 2. The present invention uses machine learning to fuse and learn multi-source data of the planar fluxgate array, solves the problem of the failure of the magnetic dipole theoretical model under near-field detection conditions, and realizes fast and accurate calculation of the type and attitude of unexploded ordnance.

[0018] 3. The present invention combines finite element high-fidelity simulation data and real-scene experimental data to establish an unexploded ordnance detection data-driven database, reducing the generation cost of data required for machine learning and improving the training efficiency of the neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a schematic diagram of a planar fluxgate array of an unexploded ordnance classification and attitude calculation method based on planar array near-field detection in this embodiment; Figure 2 is a schematic diagram of the detection criterion input into the unexploded ordnance target classification model or the unexploded ordnance attitude calculation model in this embodiment; Figure 3 is the working principle diagram of unexploded ordnance target classification and unexploded ordnance attitude calculation in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the Figures 1-3 drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0021] An unexploded ordnance classification and attitude calculation method based on planar array near-field detection in this embodiment, the unexploded ordnance classification and attitude calculation method is specifically as follows: Based on machine learning, an unexploded ordnance target classification model and an unexploded ordnance attitude calculation model are respectively constructed and trained. Among them, the input of the unexploded ordnance target classification model is the detection criterion including magnetic vector field, first-order difference of magnetic vector field and second-order difference of magnetic vector field data, and the output is the unexploded ordnance target classification result; the input of the unexploded ordnance attitude calculation model is the detection criterion and the corresponding unexploded ordnance target classification result, and the output is the unexploded ordnance attitude calculation result; Using a planar fluxgate array, magnetic vector field data at the measurement point is collected, and the corresponding first-order finite difference and second-order finite difference data of the magnetic vector field are calculated to obtain the detection criterion at the measurement point; The detection criterion at the measurement point is input into the unexploded ordnance target classification model to obtain the unexploded ordnance target classification result at the measurement point; The detection criterion at the measurement point and the unexploded ordnance target classification result are input into the unexploded ordnance attitude calculation model to obtain the unexploded ordnance attitude calculation result at the measurement point.

[0022] As Figure 1As shown, in this embodiment, the planar fluxgate array is a nested planar fluxgate array composed of three square sub-array structures. The planar fluxgate array includes 8 triaxial fluxgate sensors, numbered 1 - 8 respectively. The 4 triaxial fluxgate sensors numbered 1, 2, 3, and 4 form a square sub-array structure A1. The 4 triaxial fluxgate sensors numbered 3, 4, 5, and 6 form a square sub-array structure A2. The 4 triaxial fluxgate sensors numbered 5, 6, 7, and 8 form a square sub-array structure A3.

[0023] In this embodiment, fluxgate sensors of different models can be used for measurement. The triaxial fluxgate sensor can simultaneously measure the vector magnetic field signals of the magnetic field on the X, Y, and Z axes in the reference coordinate system.

[0024] The first-order finite difference data of the magnetic vector field includes the first-order finite difference data of the magnetic vector field at the center of A1, the first-order finite difference data of the magnetic vector field at the center of A2, and the first-order finite difference data of the magnetic vector field at the center of A3. In this embodiment, five independent components are selected for calculation respectively, and the specific calculation is as follows: ; represents the first-order finite difference data of the magnetic vector field at the center of sub-array A1 calculated based on the magnetic vector field data of fluxgate sensors numbered 1 - 4; represents the magnetic vector field data of fluxgate sensor No. 1 in the x direction, represents the magnetic vector field data of fluxgate sensor No. 1 in the y direction, represents the magnetic vector field data of fluxgate sensor No. 1 in the z direction; represents the magnetic vector field data of fluxgate sensor No. 2 in the x direction, represents the magnetic vector field data of fluxgate sensor No. 2 in the y direction, represents the magnetic vector field data of fluxgate sensor No. 2 in the z direction; represents the magnetic vector field data of fluxgate sensor No. 3 in the x direction, represents the magnetic vector field data of fluxgate sensor No. 3 in the y direction, represents the magnetic vector field data of fluxgate sensor No. 3 in the z direction; represents the magnetic vector field data of fluxgate sensor No. 4 in the x direction, represents the magnetic vector field data of fluxgate sensor No. 4 in the y direction, represents the magnetic vector field data of fluxgate sensor No. 4 in the z direction; d represents the spatial interval distance between two adjacent fluxgate sensors; ; represents the first-order finite difference data of the magnetic vector field at the center of sub-array A2 calculated based on the magnetic vector field data of fluxgate sensors No. 3 to No. 6; represents the magnetic vector field data of the No. 5 fluxgate sensor in the x direction, represents the magnetic vector field data of the No. 5 fluxgate sensor in the y direction, represents the magnetic vector field data of the No. 5 fluxgate sensor in the z direction; represents the magnetic vector field data of the No. 6 fluxgate sensor in the x direction, represents the magnetic vector field data of the No. 6 fluxgate sensor in the y direction, represents the magnetic vector field data of the No. 6 fluxgate sensor in the z direction; ; represents the first-order finite difference data of the magnetic vector field at the center of sub-array A3 calculated based on the magnetic vector field data of fluxgate sensors No. 5 to No. 8; represents the magnetic vector field data of the No. 7 fluxgate sensor in the x direction, represents the magnetic vector field data of the No. 7 fluxgate sensor in the y direction, represents the magnetic vector field data of the No. 7 fluxgate sensor in the z direction; represents the magnetic vector field data of the No. 8 fluxgate sensor in the x direction, represents the magnetic vector field data of the No. 8 fluxgate sensor in the y direction, represents the magnetic vector field data of the No. 8 fluxgate sensor in the z direction.

[0025] The second-order finite difference data of the magnetic vector field is the second-order finite difference data of the magnetic vector at the center of the planar array (i.e., the center of sub-array A2) calculated based on the first-order finite difference data of the magnetic vector fields of A1, A2, and A3; specifically: ; In the formula, represents the second-order finite difference data of the magnetic vector at the center of the planar array calculated from sub-arrays A1, A2, and A2.

[0026] Utilize the database of unexploded ordnance (UXO) detection driven by numerical and physical data to train the UXO target classification model and the UXO attitude calculation model. The database of UXO detection driven by numerical and physical data includes magnetic vector field data obtained from finite element simulation and field experiments, the first-order difference and second-order difference data of the magnetic vector field calculated based on the magnetic vector field data obtained from finite element simulation and field experiments, the classification results of UXO targets, and the angles α, β, and γ between the UXO target and the three axes.

[0027] Model various types of UXOs, set the background magnetic field and boundary conditions, and simulate and calculate the magnetic field distribution in the set area through the finite element algorithm. Extract the finite element high-fidelity simulation data that meets 1-N-44 according to the simulation results of the typical array detection method; manufacture the UXO model according to the size ratio of 1:1, build a real experimental environment, and use a fluxgate array to detect the UXO model to obtain field experiment data.

[0028] The UXO target classification model uses a classification neural network; the UXO attitude calculation model uses a regression neural network.

[0029] As Figure 2 shown, use the interpolation method to generate detection criteria containing 1 to 44 data channels for the magnetic vector field, the first-order difference of the magnetic vector field, and the second-order difference data of the magnetic vector field according to the preset distance resolution. The 44 channels respectively correspond to the magnetic vector field data detected by 1 to 8 fluxgates in 24 channels, the first-order difference data of the magnetic vector field at the centers of sub-arrays A1, A2, and A3 in 15 channels, and the second-order difference data of the magnetic vector field at the center of the planar array in 5 channels.

[0030] As Figure 3 shown, use the 1-N-44 detection criteria as the input layer of the classification neural network, and use the classification neural network to perform calculation operations such as feature extraction on it. Select the type of UXO target with the highest probability as the type of the UXO target, where the sum of the probabilities of each type is 1, that is .

[0031] As Figure 3 shown, use the 1-N-44 detection criteria and the UXO classification target results as the input layer of the regression neural network, and use the regression neural network to perform calculation operations such as feature extraction on it. The output result is the angles α, β, and γ between the UXO target and the three axes obtained through regression calculation, etc., so as to obtain the attitude calculation of the UXO. For the directions of the three axes, please refer to Figure 1 .

[0032] An electronic device includes a processor and a memory. A computer program capable of running on the processor is stored on the memory. It is characterized in that: when the processor runs the computer program, it executes the steps of any one of the above methods.

[0033] A computer-readable storage medium has a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0034] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as a limitation on the present invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims

1. A method for unexploded bomb classification and posture calculation based on planar array near-field detection, characterized in that: The unexploded bomb classification and posture calculation method is specifically as follows: Based on machine learning, an unexploded bomb target classification model and an unexploded bomb attitude solution model are respectively constructed and trained, wherein the input of the unexploded bomb target classification model is a detection criterion including magnetic vector field, first-order difference of magnetic vector field and second-order difference of magnetic vector field data, and the output is an unexploded bomb target classification result; the input of the unexploded bomb attitude solution model is a detection criterion and the corresponding unexploded bomb target classification result, and the output is an unexploded bomb attitude solution result; The planar fluxgate array is used to collect the magnetic vector field data at the measuring point, and the corresponding first-order finite difference and second-order finite difference data of the magnetic vector field are calculated to obtain the detection criterion at the measuring point. The detection criteria at the measuring point are input into the unexploded bomb target classification model to obtain the unexploded bomb target classification result at the measuring point; The detection criteria at the measuring point and the classification results of the unexploded bomb targets are input into the unexploded bomb attitude solution model to obtain the unexploded bomb attitude solution results at the measuring point.

2. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 1 is characterized in that: The planar fluxgate array is a nested planar fluxgate array composed of three square sub-array structures. The planar fluxgate array includes 8 three-axis fluxgate sensors. The 8 three-axis fluxgate sensors are No. 1-8. The four three-axis fluxgate sensors No. 1, 2, 3, and 4 form a square sub-array structure A1, the four three-axis fluxgate sensors No. 3, 4, 5, and 6 form a square sub-array structure A2, and the four three-axis fluxgate sensors No. 5, 6, 7, and 8 form a square sub-array structure A3.

3. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 2 is characterized by: The first-order finite difference data of the magnetic vector field include the first-order finite difference data of the magnetic vector field at the center of A1, the first-order finite difference data of the magnetic vector field at the center of A2, and the first-order finite difference data of the magnetic vector field at the center of A3; the specific calculation is as follows: ； represents the first-order finite difference data of the magnetic vector field at the center of subarray A1 calculated based on the magnetic vector field data of fluxgate sensors 1 to 4; Represents the magnetic vector field data of fluxgate sensor No. 1 in the x direction, Represents the magnetic vector field data of fluxgate sensor No. 1 in the y direction, Represents the magnetic vector field data of fluxgate sensor No. 1 in the z direction; Represents the magnetic vector field data of fluxgate sensor No. 2 in the x direction, Represents the magnetic vector field data of fluxgate sensor No. 2 in the y direction, Represents the magnetic vector field data of fluxgate sensor No. 2 in the z direction; Represents the magnetic vector field data of fluxgate sensor No. 3 in the x direction, Represents the magnetic vector field data of fluxgate sensor No. 3 in the y direction, Represents the magnetic vector field data of fluxgate sensor No. 3 in the z direction; Represents the magnetic vector field data of fluxgate sensor No. 4 in the x direction, Represents the magnetic vector field data of fluxgate sensor No. 4 in the y direction, It represents the magnetic vector field data of fluxgate sensor No. 4 in the z direction; d represents the spatial distance between two adjacent fluxgate sensors; ； represents the first-order finite difference data of the magnetic vector field at the center of subarray A2 calculated based on the magnetic vector field data of fluxgate sensors 3 to 6; It represents the magnetic vector field data of fluxgate sensor No. 5 in the x direction. It represents the magnetic vector field data of fluxgate sensor No. 5 in the y direction. It represents the magnetic vector field data of fluxgate sensor No. 5 in the z direction; It represents the magnetic vector field data of fluxgate sensor No. 6 in the x direction. It represents the magnetic vector field data of fluxgate sensor No. 6 in the y direction. It represents the magnetic vector field data of fluxgate sensor No. 6 in the z direction; ； represents the first-order finite difference data of the magnetic vector field at the center of subarray A3 calculated based on the magnetic vector field data of fluxgate sensors No. 5 to 8; Represents the magnetic vector field data of fluxgate sensor No. 7 in the x direction, It represents the magnetic vector field data of fluxgate sensor No. 7 in the y direction. It represents the magnetic vector field data of fluxgate sensor No. 7 in the z direction; Represents the magnetic vector field data of fluxgate sensor No. 8 in the x direction, Represents the magnetic vector field data of fluxgate sensor No. 8 in the y direction, Represents the magnetic vector field data of fluxgate sensor No. 8 in the z direction.

4. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 1 is characterized in that: The calculation of the second-order finite difference data of the magnetic vector field is as follows: ； In the formula, It represents the second-order finite difference data of the magnetic vector at the center of the plane array calculated by sub-arrays A1, A2 and A2.

5. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 1 is characterized in that: The unexploded bomb target classification model and the unexploded bomb posture solution model are trained using the digital-physics driven unexploded bomb detection database, where the digital-physics driven unexploded bomb detection database includes magnetic vector field data obtained by finite element simulation and real-life experiments, the first-order difference of the magnetic vector field and the second-order difference of the magnetic vector field calculated based on the magnetic vector field data obtained by finite element simulation and real-life experiments, the unexploded bomb target classification results, and the angles α, β, and γ between the unexploded bomb target and the three axes.

6. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 1 is characterized in that: The unexploded bomb target classification model adopts a classification neural network.

7. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 1 is characterized in that: The unexploded bomb posture solution model uses a regression neural network.

8. The method for unexploded bomb classification and posture calculation based on planar array near-field detection according to claim 1 is characterized in that: The magnetic vector field, the first-order difference of the magnetic vector field and the second-order difference of the magnetic vector field are interpolated using the interpolation method.

9. An electronic device comprising a processor and a memory, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor runs the computer program, the steps of the method according to any one of claims 1 to 8 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images