Transient electromagnetic near-surface target feature recognition system and method

Abstract

The invention provides a transient electromagnetic near-surface target feature recognition system and method, and relates to the technical field of underground target recognition, and the method comprises the steps: arranging a measurement grid, collecting a response voltage component of each measurement point through a JPMD sensor, calculating a magnetic gradient tensor, generating a target position vector, building a global coordinate system, and carrying out the Gaussian-Newton algorithm, thereby achieving the feature recognition of a transient electromagnetic near-surface target. And determining a positioning result based on the actual detection value and the predicted value of the single dipole model. The length-width ratio of the target is obtained by decomposing the magnetic gradient tensor, the attenuation time coefficient is corrected in combination with the soil conductivity to judge the conductivity of the target, clustering analysis is carried out on the target to improve the classification accuracy, and finally, target recognition models are trained for different clusters to recognize the types of the targets in the clusters. And the real-time target is identified through the corresponding target identification model after being distributed to the cluster. According to the method, multi-measurement-point data are fused, and the positioning precision and the target recognition efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of underground target recognition, and specifically to a transient electromagnetic near-surface target feature recognition system and method. Background Art

[0002] In the field of near-surface target detection, especially the detection of underground unexploded ordnance (UXO), traditional methods mainly rely on magnetic prospecting or electromagnetic induction technology. However, these technologies often have problems such as insufficient positioning accuracy, poor anti-interference ability, and complex data processing. The existing transient electromagnetic method is vulnerable to environmental noise under complex geological conditions, resulting in a reduced target recognition rate, and mostly relies on data from a single measurement point, making it difficult to achieve high-precision three-dimensional positioning. In addition, existing systems usually lack the ability of dynamic optimization, cannot effectively adapt to the changes in the electromagnetic characteristics of underground targets, and lack further identification after detecting the target, restricting their reliability and efficiency in practical applications. Therefore, there is an urgent need for a high-precision transient electromagnetic detection system that can combine multi-measurement point data and dynamically optimize the positioning algorithm to solve the above technical bottlenecks.

[0003] In the prior art, the publication number CN105137498B discloses a system and method for detecting and identifying underground targets based on feature fusion, which includes detecting and collecting the echo signals of a ground penetrating radar; preprocessing the collected echo signals; enhancing the preprocessed echo signals; extracting the time-domain features and wavelet packet energy spectrum features of the enhanced echo signals, and performing Welch power spectrum estimation and high-order spectrum analysis on the enhanced echo signals to obtain the Welch power spectrum and high-order spectrum respectively; fusing the time-domain features, wavelet packet energy spectrum features, Welch power spectrum and high-order spectrum of the echo signals; performing target recognition on the fused four types of signal features through a wavelet neural network; and outputting the result after target recognition.

[0004] The main problems of the above solution are as follows: It involves multi-stage feature extraction and fusion, and uses a wavelet neural network for classification and recognition, resulting in excessive computational complexity and insufficient real-time performance in large-scale data scenarios, making it difficult to meet the requirements of rapid detection; and integrating the time-domain features, wavelet packet energy spectrum features, Welch power spectrum and high-order spectrum simultaneously, there is information overlap leading to feature redundancy; the ground penetrating radar is sensitive to shallow targets with high dielectric constants, and has limited detection ability for deep small targets or targets with low dielectric constants, resulting in limited application scope of the solution.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and therefore it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a transient electromagnetic near-surface target feature recognition system and method to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions: A transient electromagnetic near-surface target feature recognition system specifically includes: A data acquisition module for arranging a 3×3 measurement grid in the target area, taking the center point of each grid as a measurement point, and respectively collecting the response voltage components of each measurement point; An initial position measurement module for calculating the magnetic gradient tensor of each measurement point based on the response voltage components of each measurement point, calculating the target position vector of each measurement point based on the magnetic gradient tensor and the response voltage components, establishing a global coordinate system in the measurement grid, generating the global target position of the measurement points based on the target position vectors of all measurement points, and taking the average of the global target positions of all measurement points to generate the average global target position; A model construction module for constructing a single dipole model at each measurement point, equivalenting the underground target to a magnetic dipole with a dipole moment, and predicting the response voltage components through the single dipole model; An initial position optimization and positioning module for taking the average global target position as the initial value of the Gauss-Newton algorithm, constructing a residual matrix reflecting the difference between the actual response voltage and the response voltage predicted by the single dipole model, taking the negative derivative of the residual matrix to generate a Jacobian matrix, starting from the initial value for iteration to update the average global target position until the maximum number of iterations is reached, and outputting the average global target position updated last time as the positioning result; A target feature extraction module for obtaining and decomposing the magnetic gradient tensor in the area where the positioning result is located, generating the aspect ratio of the target, synchronously measuring the soil conductivity, correcting the decay time coefficient of the transient electromagnetic response through the soil conductivity to generate the actual decay time coefficient, and reflecting the conductivity of the target through the actual decay time coefficient; The target clustering module is used to collect the known target aspect ratios and actual attenuation time coefficients that appear near the surface in historical data, perform Min-Max normalization on the aspect ratios and actual attenuation time coefficients, generate feature vectors for each known target, calculate the Euclidean distances between all pairs of known target feature vectors, take the 30% quantile of the Euclidean distances as the neighborhood radius, and set the minimum number of samples for clustering. Based on the DBSCAN clustering algorithm, perform unsupervised clustering analysis on the feature vectors of each known target. In the clustering, the feature vector of each known target is equivalent to a sample point. Calculate the mean of the sample points in each cluster as the cluster center, generate the Mahalanobis distance from each sample point in the cluster to the cluster center, set the Mahalanobis distance threshold. When the Mahalanobis distance from a sample point to the cluster center exceeds the Mahalanobis distance threshold, remove the sample point from the cluster. For each group of removed sample points, recalculate the cluster center and calculate the Mahalanobis distance from other sample points to the cluster center. Repeat the above steps of removing sample points until there are no sample points in the cluster whose Mahalanobis distance from the cluster center exceeds the Mahalanobis distance threshold. The target recognition module is used to construct a deep learning network. Respectively take the known target feature vectors in different clusters as inputs and the known target categories to which they belong as labels, train a target recognition model for each cluster. Calculate the feature vectors of the real-time located target according to the above steps, select the cluster center with the closest Euclidean distance to the feature vector of the real-time located target, assign the feature vector of the real-time located target to the cluster corresponding to the cluster center, and input the feature vector of the real-time located target into the target recognition model of the corresponding cluster to identify the category of the real-time located target.

[0008] Furthermore, a JPMD sensor including a single-layer transmitting coil and five three-axis receiving coils is used for data collection. The spacing of the measurement grid is 25 cm, and 15 response voltage components are collected at each measurement point.

[0009] Furthermore, in the JPMD sensor, taking the center of the JPMD sensor as the origin of the coordinate system, there is a three-axis receiving coil at the origin position of the coordinate system. The other four three-axis receiving coils are respectively distributed in the four positive directions of the front, back, left, and right of the origin of the coordinate system. Taking the direction from the origin of the coordinate system to the positive right as the positive x-axis direction, taking the direction from the origin of the coordinate system to the positive back as the positive y-axis direction, and perpendicular to the x-axis and y-axis and passing through the origin of the coordinate system and pointing upward as the positive z-axis direction, establish a measurement point coordinate system. Set the three-axis receiving coil at the origin position of the coordinate system as R3, and the coils in the positive and negative directions of the x-axis are respectively set as R2 and R4, and the coils in the positive and negative directions of the y-axis are respectively set as R1 and R5.

[0010] Furthermore, the principle for calculating the target position vector of each measurement point is as follows: The magnetic gradient tensor of the measurement point is: ; Among them, represents the magnetic gradient tensor at the th measurement point, represents the index of the measurement point, and , represents the rate of change of the magnetic field in the x - direction in the x - direction at the th measurement point, represents the rate of change of the magnetic field in the y - direction in the x - direction at the th measurement point, represents the rate of change of the magnetic field in the z - direction in the x - direction at the th measurement point, represents the rate of change of the magnetic field in the x - direction in the y - direction at the th measurement point, represents the rate of change of the magnetic field in the y - direction in the y - direction at the th measurement point, represents the rate of change of the magnetic field in the z - direction in the y - direction at the th measurement point, represents the rate of change of the magnetic field in the x - direction in the z - direction at the th measurement point, represents the rate of change of the magnetic field in the y - direction in the z - direction at the th measurement point, represents the rate of change of the magnetic field in the z - direction in the z - direction at the th measurement point; ; ; ; ; ; ; Among them, represents the x - axis response voltage component of R4 at the th measurement point, represents the x - axis response voltage component of R2 at the th measurement point, , represents the y - axis response voltage component of R4 at the th measurement point, represents the y - axis response voltage component of R2 at the th measurement point, represents the z - axis response voltage component of R2 at the th measurement point, represents the y - axis response voltage component of R1 at the th measurement point, represents the The x-axis response voltage component of R5 among the measurement points, representing the z-axis response voltage component of R1 among the measurement points, representing the z-axis response voltage component of R5 among the measurement points; The formula for calculating the target position vector of each measurement point is: ; wherein, representing the target position vector of the measurement point, representing the response voltage component of the measurement point.

[0011] Furthermore, the principle for generating the average global target position is: Taking the center point of the 3×3 measurement grid as the global origin, on the plane where the measurement grid is located, there is a measurement point in each of the eight positive directions of up, down, left, right, upper left, lower left, upper right, and lower right of the global origin. Taking the direction from the global origin to the due right as the positive direction of the global x-axis, the direction from the global origin to the due up as the positive direction of the global y-axis, and the direction passing through the global origin, perpendicular to the global x-axis and the global y-axis, and pointing outside the grid plane as the positive direction of the global z-axis, a global coordinate system of the target area is constructed; For each measurement point, it has a known global coordinate and a target position vector, then the global target position of a single measurement point is: ; wherein, representing the global target position of the measurement point, representing the global coordinate of the measurement point; The average global target position is: ; wherein, representing the average global target position.

[0012] Furthermore, the principle for updating the target position is: The residual matrix is: ; wherein, representing the residual matrix, representing the response voltage of the measurement point, representing the response voltage predicted by the single dipole model; The Jacobian matrix is: ; Among them, represents the Jacobian matrix; The iterative equation for the target position is: ; ; Among them, represents the average global target position at the -th iteration, represents the average global target position at the -th iteration, represents the index of the number of iterations, and , represents the maximum number of iterations. When , , represents the correction amount for the average global target position at the -th iteration.

[0013] Furthermore, the principle for generating the target aspect ratio is: The magnetic gradient tensor in the area where the positioning result is located is: ; Among them, represents the magnetic gradient tensor in the area where the positioning result is located, , , respectively represent the magnetic field components of the x, y, and z axes; The formula for decomposing into three eigenvalues is: Solve the equation to obtain three eigenvalues , and , where represents the identity matrix; The aspect ratio is: ; Among them, represents the target aspect ratio.

[0014] Furthermore, the principle for judging the conductivity of the target based on the corrected decay time coefficient is: The formula for correcting the decay time coefficient is: ; Among them, represents the actual decay time coefficient, represents the decay time coefficient before correction, represents the empirical coefficient, represents the soil conductivity; Is proportional to the target conductivity.

[0015] Furthermore, the principle for performing unsupervised clustering analysis is as follows: The formula for calculating the Euclidean distance between all pairs of known target feature vectors is: ; Wherein, , respectively represent the o-th and p-th known target feature vectors, represents , the Euclidean distance between, and , represents the total number of known target feature vectors, represents the normalized aspect ratio of the o-th known target feature vector, represents the normalized aspect ratio of the p-th known target feature vector, represents the normalized actual decay time coefficient of the o-th known target feature vector, represents the normalized actual decay time coefficient of the p-th known target feature vector; Take the neighborhood radius of clustering as the 30% quantile of the Euclidean distances of all sample points, and the minimum number of sample points for clustering is , which means taking the ceiling of the square root of the total number of known target feature vectors; randomly select a sample point, count the other sample points within its neighborhood radius, if the number exceeds , then regard the selected sample point as the clustering center, and assign all sample points to the cluster where the clustering center with the closest Euclidean distance to itself is located, regarded as completing the unsupervised clustering analysis; The formula for calculating the clustering center in each cluster is: ; Wherein, represents the clustering center of the -th cluster, represents the index of the cluster, represents the -th number of sample points in the cluster, represents the index of the sample point in the cluster, and , represents the -th number of sample points in the cluster, represents the -th target feature vector corresponding to the sample point; The formula for generating the Mahalanobis distance from each feature vector in the cluster to the clustering center is: ; ; wherein, represents the Mahalanobis distance from the -th sample point in the -th cluster to the cluster center, represents the within-class covariance matrix of the -th cluster; When , the sample point is removed from the cluster, represents the Mahalanobis distance threshold.

[0016] The present invention also provides a transient electromagnetic near-surface target feature recognition method, which is executed by the above transient electromagnetic near-surface target feature recognition system. The specific steps include: Step 1: Arrange a 3×3 measurement grid in the target area, take the center point of each grid as a measurement point, and collect the response voltage components of each measurement point respectively; Step 2: Calculate the magnetic gradient tensor of the measurement point based on the response voltage components of each measurement point, calculate the target position vector of each measurement point based on the magnetic gradient tensor and the response voltage components, establish a global coordinate system in the measurement grid, generate the global target position of the measurement point based on the target position vectors of all measurement points, and take the average value of the global target positions of all measurement points to generate the average global target position; Step 3: Construct a single dipole model at each measurement point, equivalent the underground target to a magnetic dipole with a dipole moment, and predict the response voltage through the single dipole model; Step 4: Take the average global target position as the initial value of the Gauss-Newton algorithm, construct a residual matrix reflecting the difference between the actual response voltage and the predicted response voltage of the single dipole model, take the negative derivative of the residual matrix to generate the Jacobian matrix, start iterating from the initial value to update the average global target position until the maximum number of iterations is reached, and output the last updated average global target position as the positioning result; Step 5: Obtain and decompose the magnetic gradient tensor in the area where the positioning result is located to generate the target aspect ratio, synchronously measure the soil conductivity, correct the decay time coefficient of the transient electromagnetic response through the soil conductivity to generate the actual decay time coefficient, and reflect the conductivity of the target through the actual decay time coefficient; Step 6: Collect the known target aspect ratios and actual attenuation time coefficients that appear near the surface in historical data. Perform Min-Max normalization on the aspect ratios and actual attenuation time coefficients to generate feature vectors for each known target. Calculate the Euclidean distances between all pairs of known target feature vectors. Take the 30% quantile of the Euclidean distances as the neighborhood radius and set the minimum number of samples for clustering. Conduct unsupervised clustering analysis on the feature vectors of each known target based on the DBSCAN clustering algorithm. In the clustering, the feature vector of each known target is equivalent to a sample point. Calculate the mean value of the sample points in each cluster as the cluster center, generate the Mahalanobis distance from each sample point in the cluster to the cluster center, and set the Mahalanobis distance threshold. When the Mahalanobis distance from a sample point to the cluster center exceeds the Mahalanobis distance threshold, remove the sample point from the cluster. For each group of removed sample points, recalculate the cluster center and calculate the Mahalanobis distance from other sample points to the cluster center. Repeat the above steps of removing sample points until there are no sample points in the cluster whose Mahalanobis distance from the cluster center exceeds the Mahalanobis distance threshold; Step 7: Construct a deep learning network. Respectively use the known target feature vectors in different clusters as inputs and the known target category to which the target belongs as labels. Train a target recognition model for each cluster. Calculate the feature vector of the real-time located target according to the above steps, select the cluster center with the closest Euclidean distance to the feature vector of the real-time located target, assign the feature vector of the real-time located target to the cluster corresponding to the cluster center, and input the feature vector of the real-time located target into the target recognition model of the corresponding cluster to identify the category of the real-time located target.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can more accurately deduce the target position vector by calculating the magnetic gradient tensor of each measurement point. This positioning method based on the magnetic field gradient can effectively reduce the influence of environmental noise and measurement errors compared with the traditional single measurement point response data, improve the accuracy of target positioning, and realize the effective fusion of data at each measurement point. This method can form a more consistent global positioning result, which helps to eliminate local measurement errors and improve the credibility of the overall data; The present invention can also effectively simulate the electromagnetic response of small near-surface targets by constructing a single dipole model for measurement points, with a wide range of applications. Compared with complex multi-dipole models, the single dipole model can still provide relatively accurate predictions under conditions of a small amount of data, reducing computational complexity; by constructing a residual matrix that reflects the difference between actual data and model prediction results, the deficiencies of the model can be clearly identified, enabling real-time and continuous optimization, reducing computational complexity while ensuring accuracy and improving operational efficiency. The present invention can also generate the aspect ratio of a target and extract the geometric features of the target by acquiring and decomposing the magnetic gradient tensor, providing more comprehensive target information. By correcting the decay time coefficient of the transient electromagnetic response through soil conductivity, the true electrical characteristics of the target can be accurately reflected, reducing the influence of environmental factors on target recognition and improving the accuracy of judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the system module of an embodiment of the present invention; Figure 2 It is a broken line graph of the soil conductivity correction decay time coefficient of an embodiment of the present invention; Figure 3 It is a schematic diagram of the method flow of an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] 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 in conjunction with specific embodiments.

[0020] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those of ordinary skill in the field to which the present invention pertains. The "first", "second", and similar terms used in the present invention do not denote any order, quantity, or importance, but are only used to distinguish different components. The terms "including" or "comprising" and the like mean that the elements or items appearing before this term cover the elements or items listed after this term and their equivalents, without excluding other elements or items. The terms "connected" or "coupled" and the like are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0021] Embodiment: Please refer to Figures 1 to 2 , the present invention provides a technical solution: A transient electromagnetic near-surface target feature recognition system specifically includes: The data acquisition module is used to arrange a 3×3 measurement grid in the target area, take the center point of each grid as a measurement point, and collect the response voltage components of each measurement point respectively. In this embodiment, a JPMD sensor including a single-layer transmitting coil and five three-axis receiving coils is used for data acquisition. The spacing of the measurement grid is 25 cm, and 15 response voltage components are collected at each measurement point.

[0022] The diameter of the transmitting coil of the JPMD sensor is 50 cm, the height is 7 cm, and the number of turns is 30. The height of the five receiving coils is 6 cm, and the distance between the receiving coil at the center of the transmitting coil and the other receiving coils is 20 cm. In the JPMD sensor, taking the center of the JPMD sensor as the origin of the coordinate system, there is a three-axis receiving coil at the origin position of the coordinate system. The other four three-axis receiving coils are respectively distributed in the four positive directions of the front, back, left, and right of the origin of the coordinate system. Taking the direction from the origin of the coordinate system to the positive right as the positive x-axis direction, taking the direction from the origin of the coordinate system to the positive back as the positive y-axis direction, and perpendicular to the x-axis and y-axis and passing through the origin of the coordinate system and pointing upward as the positive z-axis direction, a measurement point coordinate system is established. The three-axis receiving coil at the origin position of the coordinate system is set as R3, the coils in the positive and negative directions of the x-axis are respectively set as R2 and R4, and the coils in the positive and negative directions of the y-axis are respectively set as R1 and R5.

[0023] The initial position measurement module is used to calculate the magnetic gradient tensor of the measurement point based on the response voltage components of each measurement point, calculate the target position vector of each measurement point based on the magnetic gradient tensor and the response voltage components, establish a global coordinate system in the measurement grid, generate the global target position of the measurement point based on the target position vectors of all measurement points, and take the average of the global target positions of all measurement points to generate the average global target position. In this embodiment, the principle for calculating the target position vector of each measurement point is as follows: The magnetic gradient tensor of the measurement point is: ; Among them, represents the magnetic gradient tensor of the th measurement point, represents the index of the measurement point, and , represents the change rate of the magnetic field in the x direction in the x direction of the th measurement point, represents the change rate of the magnetic field in the y direction in the x direction of the th measurement point, represents the change rate of the magnetic field in the z direction in the x direction of the th measurement point, represents the The rate of change of the magnetic field in the x - direction at a measurement point in the y - direction, denotes the rate of change of the magnetic field in the y - direction at a measurement point in the y - direction, denotes the rate of change of the magnetic field in the z - direction at a measurement point in the y - direction, denotes the rate of change of the magnetic field in the x - direction at a measurement point in the z - direction, denotes the rate of change of the magnetic field in the y - direction at a measurement point in the z - direction, denotes the rate of change of the magnetic field in the z - direction at a measurement point in the z - direction; The magnetic gradient tensor is a 3×3 matrix that reflects the rate of change of the magnetic field in the x, y, and z directions. The calculation of the magnetic gradient tensor is based on the response voltages measured by multiple receiving coils. Each receiving coil measures the response voltages in the x, y, and z directions; the rate of change of the magnetic field in each direction is calculated by the differential approximation method; ; ; ; ; ; ; where, denotes the x - axis response voltage component of R4 at the denotes the x - axis response voltage component of R2 at the , denotes the y - axis response voltage component of R4 at the denotes the y - axis response voltage component of R2 at the denotes the z - axis response voltage component of R2 at the denotes the y - axis response voltage component of R1 at the denotes the x - axis response voltage component of R5 at the denotes the z - axis response voltage component of R1 at the denotes the z - axis response voltage component of R5 at the In a uniform grid, the rate of change of the magnetic field is calculated based on the first-order central difference formula. represents the rate of change of the magnetic field in the x-direction at the measurement point in the x-direction. To calculate this rate of change, two symmetric points need to be selected on the x-axis. The difference between the measured values of the response voltage components between the symmetric points is divided by the spacing, and the rate of change of the magnetic field can be approximately obtained. The distances from R1, R2, R4, and R5 to the origin of the coordinate system are the same, all being d. R4 is in the positive x-axis direction, and the distance to the origin is d. R2 is in the negative x-axis direction, and the distance to the origin is d. Therefore, the magnetic field components and directly reflect the change in the magnitude of the magnetic field in the x-axis direction. Therefore, the calculation of the difference formula is . Since the magnetic field is a divergence-free field and its divergence is zero, , , does not need to be measured independently and can be directly calculated from and . In the low-frequency electromagnetic field, the magnetic gradient tensor is symmetric, that is, , , ; when calculating , the calculation formula of is directly used, that is, the rate of change of the magnetic field in the x-direction in the y-direction. Similarly, is calculated.

[0024] The formula for calculating the target position vector of each measurement point is: ; where represents the target position vector of the th measurement point, represents the response voltage component of the th measurement point.

[0025] The formula for calculating the target position vector is based on the magnetic dipole model. The magnetic field generated by the target is approximately a dipole field. Through the linear relationship between the magnetic gradient tensor and the response voltage, the position vector of the target relative to the measurement point is inversely calculated, that is, . The underground target is approximated as a point magnetic dipole, and the magnetic field it generates satisfies in space, where represents the magnetic field generated by the target, m represents the magnetic dipole moment of the target, represents the position vector from the target to the sensor, represents the unit direction vector, represents the vacuum permeability. In the far field, that is, when the target distance is much greater than the dipole size, then , in transient electromagnetic method, the response voltage of the receiving coil is proportional to the magnetic field change rate. , the magnetic moment m is proportional to the external excitation magnetic field and the polarizability of the target. The external excitation magnetic field is the magnetic field generated by the transmitting coil. , taking the inverse on both sides, then , the proportionality coefficient is about -3, then ; the above formula is based on the observed electromagnetic response to reverse infer the target position.

[0026] The principle for generating the average global target position is as follows: Taking the center point of the 3×3 measurement grid as the global origin, on the plane where the measurement grid is located, there is a measurement point in each of the eight positive directions of up, down, left, right, upper left, lower left, upper right, and lower right of the global origin. Taking the direction pointing directly to the right from the global origin as the positive direction of the global x-axis, the direction pointing directly upward from the global origin as the positive direction of the global y-axis, and taking the line passing through the global origin, perpendicular to the global x-axis and the global y-axis, and pointing outside the grid plane as the positive direction of the global z-axis, to construct the global coordinate system of the target area; For each measurement point, there is a known global coordinate and a target position vector. Then the global target position of a single measurement point is: ; Among them, represents the global target position of the th measurement point, represents the global coordinate of the th measurement point; The target position vector represents the offset of the target relative to the sensor position of the current measurement point. Then is the coordinate of the measurement point in the global coordinate system, which reflects the position of the sensor in the global space. The position of the target in the global coordinate system = the global position of the sensor + the offset of the target relative to the sensor. Keeping the axial directions of the local coordinate system of the sensor and the global coordinate system completely consistent, so and can be directly added. The essence of is the vector pointing from the sensor position to the target. The essence of adding the two is to convert the target offset in the local coordinate system to the global coordinate system. In the global coordinate system, the absolute position of the target is the sensor position plus this vector. By taking the average of for all measurement points, the single-point measurement error can be suppressed and the positioning robustness can be improved.

[0027] The average global target position is: ; Among them, Indicates the average global target position.

[0028] A model construction module, which is used to construct a single-dipole model at each measurement point, equivalent the underground target to a magnetic dipole with a dipole moment, and predict the response voltage component through the single-dipole model; In this embodiment, the technical feature of predicting the response voltage through the single-dipole model has been disclosed in the paper "Fast Localization of Underground Targets by Magnetic Gradient Tensor and Gaussian-Newton Algorithm With a Portable Transient Electromagnetic System Li Jie Wang etc.", and will not be elaborated here.

[0029] An initial position optimization and positioning module, which is used to take the average global target position as the initial value of the Gaussian-Newton algorithm, construct a residual matrix reflecting the difference between the actual response voltage and the response voltage predicted by the single-dipole model, take the negative derivative of the residual matrix to generate a Jacobian matrix, start iterating from the initial value to update the target position until the maximum number of iterations is reached, and output the target position updated last time as the positioning result; In this embodiment, the principle based on which the target position is updated is: The residual matrix is: ; Among them, represents the residual matrix, represents the response voltage at the measurement point, represents the response voltage predicted by the single-dipole model; The core of the residual matrix is to quantify the difference between the actual observed data and the model-predicted data. During the target positioning process, there is a deviation between the response voltage predicted by the single-dipole model and the response voltage actually measured by the sensor. The residual matrix gradually adjusts the estimated value of the target position by calculating this deviation to change the response voltage actually measured by the sensor until the estimated value of the target position is close to the true position. represents the response voltage data actually measured by the sensor, which is a multi-dimensional vector containing multiple measurement points and coil components. is the theoretical voltage calculated through the single-dipole model. The smaller the residual, the closer the actual voltage is to the theoretical voltage, and the closer the currently calculated average global target position is to the actual positioning result. The voltage predicted by the single-dipole model depends on the average global target position , when changes, the predicted magnetic field distribution and response voltage change, resulting in a change in the residual; Denotes the response voltage, which is a matrix composed of the response voltage components at the measurement points, and is a matrix composed of the response voltage components predicted by the single dipole model; the matrix of the response voltage components at the i-th measurement point is composed of the response voltage components of the five receiving coils in the x, y, and z directions, , where, Denotes the matrix of the response voltage components at the i-th measurement point, to Respectively denote the response voltage components of R1 to R5 in the x-axis direction, to Respectively denote the response voltage components of R1 to R5 in the y-axis direction, to Respectively denote the response voltage components of R1 to R5 in the z-axis direction. Combine the matrices of all measurement points to generate , which is also in the same form.

[0030] The Jacobian matrix is: ; where, Denotes the Jacobian matrix; The Jacobian matrix is used to describe the sensitivity of the residual to the target position, that is, how the residual changes with the change of the target position. The Jacobian matrix is the negative partial derivative matrix of the residual matrix with respect to the target position. The negative sign indicates updating the position along the direction of decreasing residual. Each row in the Jacobian matrix corresponds to the residual of a measurement point, and each column corresponds to the partial derivative of the target position.

[0031] The iterative equation for the target position is: ; ; where, Denotes the average global target position at the -th iteration, Denotes the average global target position at the -th iteration, Denotes the index of the iteration number, and , Denotes the maximum number of iterations. When , , Denotes the correction amount for the average global target position at the -th iteration.

[0032] The target feature extraction module is used to obtain and decompose the magnetic gradient tensor in the area where the positioning result is located, generate the target aspect ratio, synchronously measure the soil conductivity, correct the decay time coefficient of the transient electromagnetic response through the soil conductivity to generate the actual decay time coefficient, and reflect the conductivity of the target through the actual decay time coefficient; In this embodiment, the principle for generating the target aspect ratio is as follows: The magnetic gradient tensor in the area where the positioning result is located is: ; Among them, represents the magnetic gradient tensor in the area where the positioning result is located, , , respectively represent the magnetic field components of the x, y, and z axes; For The formula for decomposing it to obtain three eigenvalues is: Based on the eigenvalue decomposition of the matrix, solve the equation , and obtain three eigenvalues , and , where represents the identity matrix; The aspect ratio is: ; Among them, represents the target aspect ratio.

[0033] The eigenvalue decomposition of the magnetic gradient tensor essentially decouples the distribution of the magnetic field gradient in space to three orthogonal principal axes. Each eigenvalue corresponds to a direction in which the magnetic field gradient changes most significantly; the three eigenvalues reflect the geometric and physical characteristics of the target magnetic source. The largest eigenvalue represents the rate of change of the magnetic field gradient in the direction of the longest principal axis of the target, which is positively correlated with the length or extension direction of the target. The middle eigenvalue reflects the rate of change of the magnetic field gradient of the target in the direction of the secondary principal axis and is used to reflect the symmetry of the target. The smallest eigenvalue represents the rate of change of the magnetic field gradient in the direction of the shortest principal axis of the target. Through and , the shape of the target can be roughly reflected. If the target is long and strip-shaped, then is significantly larger than the other two eigenvalues. If the target is flat, then will be significantly smaller than the other two eigenvalues. Through to reflect the target aspect ratio, which quantifies the anisotropy of the target. When , the target is long and strip-shaped, meeting the shape characteristics of objects such as cables and unexploded ordnance. When , the target is close to a sphere or cube, meeting the shape characteristics of objects such as landmines and metal blocks.

[0034] The principle for determining the conductivity level of the target based on the corrected decay time coefficient is as follows: The formula for correcting the decay time coefficient is: ; Wherein, represents the actual decay time coefficient, represents the decay time coefficient before correction, represents the empirical coefficient, represents the soil conductivity; is directly proportional to the target conductivity.

[0035] In transient electromagnetic response, the decay time coefficient of the signal not only depends on the conductivity of the target itself but is also affected by the soil conductivity of the surrounding area. High-conductivity soil will absorb some electromagnetic energy due to the eddy current effect, resulting in the observed decay time coefficient being smaller than the true value. Therefore, it is necessary to dynamically correct the soil conductivity. And in most surface environments, the influence of soil conductivity on the decay time coefficient is linear. Through the linear correction model corrects , reflecting the positive influence of soil conductivity on the correction amount of the decay time coefficient, that is, the higher the soil conductivity, the greater the correction amount of the decay time coefficient to balance the influence of soil conductivity; the empirical coefficient quantifies the correction intensity of soil conductivity on the decay time coefficient. Measure the decay time coefficient of the target under different soil conductivity conditions, and determine the optimal value of k through regression analysis; the decay time coefficient represents the time required for the transient electromagnetic signal to decay from an initial value to a certain ratio, reflecting the speed of the eddy current dissipation ability. The larger the decay time coefficient, the longer the decay time, the slower the eddy current decay, and the better the target conductivity. On the contrary, the shorter the decay time, the faster the eddy current decay, and the worse the target conductivity; for targets with high conductivity, such as copper, aluminum, etc., their eddy current decay is slow, is larger. For targets with low conductivity, such as rocks, etc., the eddy current decay is slow, is smaller. And targets with large size or high aspect ratio may exhibit a longer decay time due to eddy current path delay. The corrected , eliminating the influence of soil conductivity, more directly reflects the true conductivity of the target, improves the accuracy of target recognition, and avoids misjudgment caused by environmental changes. The correction of soil conductivity is shown in Table 1; Table 1. ; A target clustering module is used to collect the known target aspect ratios and actual attenuation time coefficients that appear near the surface in historical data, perform Min-Max normalization on the aspect ratios and actual attenuation time coefficients to generate feature vectors for each known target, calculate the Euclidean distances between all pairs of known target feature vectors, take the 30th percentile of the Euclidean distances as the neighborhood radius, and set the minimum number of samples for clustering. Based on the DBSCAN clustering algorithm, unsupervised clustering analysis is performed on the feature vectors of each known target. In the clustering, the feature vector of each known target is equivalent to a sample point. The mean value of the sample points is calculated as the cluster center in each cluster, and the Mahalanobis distance from each sample point in the cluster to the cluster center is generated. A Mahalanobis distance threshold is set. When the Mahalanobis distance from a sample point to the cluster center exceeds the Mahalanobis distance threshold, the sample point is removed from the cluster. For each group of removed sample points, the cluster center is recalculated, and the Mahalanobis distances from other sample points to the cluster center are calculated. The above steps of removing sample points are repeated until there are no sample points in the cluster whose Mahalanobis distance from the cluster center exceeds the Mahalanobis distance threshold. In this embodiment, the principle for performing unsupervised clustering analysis is as follows: The formula for calculating the Euclidean distances between all pairs of known target feature vectors is: ; Where and represent the o-th and p-th known target feature vectors respectively, represents the Euclidean distance between and , represents the total number of known target feature vectors, represents the normalized aspect ratio of the o-th known target feature vector, represents the normalized aspect ratio of the p-th known target feature vector, represents the normalized actual attenuation time coefficient of the o-th known target feature vector, represents the normalized actual attenuation time coefficient of the p-th known target feature vector; The neighborhood radius of the clustering is taken as the 30th percentile of the Euclidean distances of all sample points, and the minimum number of sample points for clustering is , which means taking the ceiling of the square root of the total number of known target feature vectors; randomly select a sample point, count the other sample points whose distances from it are within the neighborhood radius. If the number exceeds , then the selected sample point is regarded as the cluster center, and all sample points are assigned to the cluster where the cluster center with the closest Euclidean distance to itself is located, which is regarded as the completion of unsupervised clustering analysis; The formula for calculating the cluster center in each cluster is: ; Among them, represents the cluster center of the th cluster, represents the index of the cluster, represents the number of sample points in the th cluster, represents the index of the sample point in the cluster, and , represents the number of sample points in the th cluster, represents the target feature vector corresponding to the th sample point; The formula for generating the Mahalanobis distance from each feature vector in the cluster to the cluster center is: ; ; Among them, represents the Mahalanobis distance from the th sample point in the th cluster to the cluster center, represents the within-class covariance matrix of the th cluster; When , remove this sample point from the cluster, represents the Mahalanobis distance threshold.

[0036] The above unsupervised clustering analysis is to ensure that the grouping of targets that may appear near the surface is more accurate based on the features of known targets, group targets with similar electromagnetic features, so as to identify different target types that may exist near the surface. In the absence of prior labels, the natural distribution pattern in the data is discovered through clustering, avoiding the subjectivity of manual classification. After generating the clusters, small-sample analysis is carried out, and the deviation degree of each sample point in the cluster from the cluster center is evaluated through the Mahalanobis distance. Those with too large deviations are removed, and multiple iterations are performed. Each time a batch of sample points is removed, the cluster center is recalculated until there are no sample points in the cluster whose deviation from the cluster center exceeds the Mahalanobis clustering threshold, ensuring that each cluster contains only targets with consistent features; compared with the Euclidean distance, the Mahalanobis distance takes into account the correlation between features, that is, the aspect ratio of an object may affect its conductivity. The aspect ratio reflects the geometric characteristics of the target, and the decay time coefficient reflects the conductivity. Combining the two can distinguish targets with similar shapes but different materials. Clustering analysis makes a comprehensive judgment through multi-dimensional features, avoiding the limitations of a single feature, and ensuring that the generated clusters can truly reflect that targets with similar feature vectors can be grouped into the same category.

[0037] The target recognition module is used to construct a deep learning network. The known target feature vectors in different clusters are respectively used as inputs, and the categories to which the known targets belong are used as labels. For each cluster, a target recognition model is trained. The feature vectors of the targets located in real time are calculated according to the above steps, the cluster center with the closest Euclidean distance to the feature vectors of the targets located in real time is selected, the feature vectors of the targets located in real time are assigned to the cluster corresponding to this cluster center, and the feature vectors of the targets located in real time are input into the target recognition models of the corresponding clusters to identify the categories of the targets located in real time.

[0038] In this embodiment, the structure of the deep learning network for each cluster is as follows: Input layer: It contains 1 node and is used to input the target feature vectors; First hidden layer: It contains 64 neurons, and ReLU is used as the activation function; Second hidden layer: It contains 32 neurons, and ReLU is used as the activation function; Output layer: It contains 1 node and is used to output the target categories.

[0039] Please refer to Figure 3 , the present invention also provides a transient electromagnetic near-surface target feature recognition method, which is executed by the above transient electromagnetic near-surface target feature recognition system. The specific steps include: Step 1: Arrange a 3×3 measurement grid in the target area. The center point of each grid is used as a measurement point, and the response voltage components of each measurement point are respectively collected; Step 2: Calculate the magnetic gradient tensor of the measurement point based on the response voltage components of each measurement point, calculate the target position vector of each measurement point based on the magnetic gradient tensor and the response voltage components, establish a global coordinate system in the measurement grid, generate the global target position of the measurement points based on the target position vectors of all measurement points, and take the average of the global target positions of all measurement points to generate the average global target position; Step 3: Construct a single-dipole model at each measurement point, equivalent the underground target to a magnetic dipole with a dipole moment, and predict the response voltage components through the single-dipole model; Step 4: Use the average global target position as the initial value of the Gauss-Newton algorithm, construct a residual matrix reflecting the difference between the actual response voltage and the response voltage predicted by the single-dipole model, calculate the Jacobian matrix by taking the negative derivative of the residual matrix, start iterating from the initial value to update the average global target position until the maximum number of iterations is reached, and output the average global target position updated last time as the positioning result; Step 5: Obtain and decompose the magnetic gradient tensor in the region where the positioning result is located, generate the target aspect ratio, synchronously measure the soil conductivity, correct the decay time coefficient of the transient electromagnetic response through the soil conductivity to generate the actual decay time coefficient, and reflect the conductivity of the target through the actual decay time coefficient; Step 6: Collect the known target aspect ratios and actual decay time coefficients that appear near the surface in the historical data, perform Min-Max normalization on the aspect ratio and the actual decay time coefficient to generate the feature vectors of each known target, calculate the Euclidean distances between all pairs of known target feature vectors, take the 30% quantile of the Euclidean distances as the neighborhood radius, and set the minimum number of samples for clustering. Based on the DBSCAN clustering algorithm, perform unsupervised clustering analysis on the feature vectors of each known target. In the clustering, the feature vector of each known target is equivalent to a sample point. Calculate the mean value of the sample points in each cluster as the cluster center, generate the Mahalanobis distance from each sample point in the cluster to the cluster center, set the Mahalanobis distance threshold. When the Mahalanobis distance from a sample point to the cluster center exceeds the Mahalanobis distance threshold, remove this sample point from the cluster. For each group of removed sample points, recalculate the cluster center and calculate the Mahalanobis distance from other sample points to the cluster center. Repeat the above steps of removing sample points until there are no sample points in the cluster whose Mahalanobis distance from the cluster center exceeds the Mahalanobis distance threshold; Step 7: Construct a deep learning network. Respectively use the known target feature vectors in different clusters as inputs and the category to which the known target belongs as labels. Train the target recognition model for each cluster. Calculate the feature vector of the real-time positioned target according to the above steps, select the cluster center with the closest Euclidean distance to the feature vector of the real-time positioned target, assign the feature vector of the real-time positioned target to the cluster corresponding to this cluster center, and input the feature vector of the real-time positioned target into the target recognition model of the corresponding cluster to identify the category of the real-time positioned target.

[0040] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0041] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0042] The unit described as a separation component may or may not be physically separated. The component shown as a unit may or may not be a physical unit, and it may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0043] As described above, the specific implementation manners of the present application are only described, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A transient electromagnetic near-surface target feature recognition system, characterized in that, Specifically, it includes: A data acquisition module, which is used to arrange a 3×3 measurement grid in the target area, take the center point of each grid as a measurement point, and respectively collect the response voltage components of each measurement point; An initial position measurement module, which is used to calculate the magnetic gradient tensor of the measurement point based on the response voltage components of each measurement point, calculate the target position vector of each measurement point based on the magnetic gradient tensor and the response voltage components, establish a global coordinate system in the measurement grid, generate the global target position of the measurement points based on the target position vectors of all measurement points, and take the average value of the global target positions of all measurement points to generate the average global target position; A model construction module, which is used to construct a single dipole model at each measurement point, equivalent the underground target to a magnetic dipole with a dipole moment, and predict the response voltage components through the single dipole model; An initial position optimization and positioning module, which is used to take the average global target position as the initial value of the Gauss-Newton algorithm, construct a residual matrix reflecting the difference between the actual response voltage and the response voltage predicted by the single dipole model, take the negative derivative of the residual matrix to generate the Jacobian matrix, start iteration from the initial value to update the average global target position until the maximum number of iterations is reached, and output the average global target position updated for the last time as the positioning result; A target feature extraction module, which is used to obtain and decompose the magnetic gradient tensor in the area where the positioning result is located, generate the target aspect ratio, synchronously measure the soil conductivity, correct the decay time coefficient of the transient electromagnetic response through the soil conductivity to generate the actual decay time coefficient, and reflect the conductivity of the target through the actual decay time coefficient; A target clustering module, which is used to collect the known target aspect ratios and actual decay time coefficients that appear near the surface in the historical data, perform Min-Max normalization on the aspect ratio and the actual decay time coefficient to generate the feature vector of each known target, calculate the Euclidean distance between every two feature vectors of all known targets, take the 30% quantile of the Euclidean distance as the neighborhood radius, and set the minimum number of samples for clustering. Based on the DBSCAN clustering algorithm, perform unsupervised clustering analysis on the feature vectors of each known target. In the clustering, the feature vector of each known target is equivalent to a sample point. Calculate the mean value of the sample points in each cluster as the cluster center, generate the Mahalanobis distance from each sample point in the cluster to the cluster center, set the Mahalanobis distance threshold. When the Mahalanobis distance from the sample point to the cluster center exceeds the Mahalanobis distance threshold, remove the sample point from the cluster. For each group of removed sample points, recalculate the cluster center and calculate the Mahalanobis distance from other sample points to the cluster center. Repeat the above steps of removing sample points until there are no sample points in the cluster whose Mahalanobis distance from the cluster center exceeds the Mahalanobis distance threshold; The target recognition module is used to construct a deep learning network. Respectively, the known target feature vectors in different clusters are used as inputs, and the types to which the known targets belong are used as labels. For each cluster, a target recognition model is trained. Calculate the feature vector of the real-time located target according to the above steps, select the cluster center with the closest Euclidean distance to the feature vector of the real-time located target, assign the feature vector of the real-time located target to the cluster corresponding to this cluster center, and input the feature vector of the real-time located target into the target recognition model of the corresponding cluster to identify the type of the real-time located target.

2. The transient electromagnetic near-surface target feature recognition system according to claim 1, wherein: In the data acquisition module, a JPMD sensor including a single-layer transmitting coil and five three-axis receiving coils is used for data acquisition. The spacing of the measurement grid is 25 cm, and 15 response voltage components are collected at each measurement point.

3. A transient electromagnetic near-surface target feature recognition system according to claim 2, characterized in that: In the JPMD sensor, taking the center of the JPMD sensor as the origin of the coordinate system, there is a three-axis receiving coil at the origin position of the coordinate system. The other four three-axis receiving coils are respectively distributed in the four positive directions of the front, back, left, and right of the origin of the coordinate system. Taking the direction from the origin of the coordinate system to the positive right as the positive x-axis direction, taking the direction from the origin of the coordinate system to the positive back as the positive y-axis direction, perpendicular to the x-axis and y-axis and passing through the origin of the coordinate system and pointing upward as the positive z-axis direction, a measurement point coordinate system is established. The three-axis receiving coil at the origin position of the coordinate system is set as R3, the coils in the positive and negative directions of the x-axis are respectively set as R2 and R4, and the coils in the positive and negative directions of the y-axis are respectively set as R1 and R5.

4. A transient electromagnetic near-surface target feature recognition system according to claim 3, characterized in that: In the initial position measurement module, the principle for calculating the target position vector of each measurement point is: The magnetic gradient tensor of the measurement point is: ； Among them, represents the magnetic gradient tensor of the th measurement point, represents the index of the measurement point, and , represents the rate of change of the magnetic field in the x - direction in the x - direction of the th measurement point, represents the rate of change of the magnetic field in the y - direction in the x - direction of the th measurement point, represents the rate of change of the magnetic field in the z - direction in the x - direction of the th measurement point, represents the rate of change of the magnetic field in the x - direction in the y - direction of the th measurement point, represents the rate of change of the magnetic field in the y - direction in the y - direction of the th measurement point, represents the rate of change of the magnetic field in the z - direction in the y - direction of the th measurement point, represents the rate of change of the magnetic field in the x - direction in the z - direction of the th measurement point, represents the rate of change of the magnetic field in the y - direction in the z - direction of the th measurement point, represents the rate of change of the magnetic field in the z - direction in the z - direction of the th measurement point; ； ； ； ； ； ； Among them, represents the x-axis response voltage component of R4 at the th measurement point, represents the x-axis response voltage component of R2 at the th measurement point, represents the distance from R1 to the origin of the coordinate system, represents the y-axis response voltage component of R4 at the th measurement point, represents the y-axis response voltage component of R2 at the th measurement point, represents the z-axis response voltage component of R2 at the th measurement point, represents the y-axis response voltage component of R1 at the th measurement point, represents the x-axis response voltage component of R5 at the th measurement point, represents the z-axis response voltage component of R1 at the th measurement point, represents the z-axis response voltage component of R5 at the th measurement point; The formula for calculating the target position vector of each measurement point is: ； Among them, represents the target position vector of the th measurement point, represents the response voltage component of the th measurement point.

5. A transient electromagnetic near-surface target feature recognition system according to claim 4, wherein: The principle for generating the average global target position is: Taking the center point of the 3×3 measurement grid as the global origin, on the plane where the measurement grid is located, there is a measurement point in each of the eight positive directions of up, down, left, right, upper left, lower left, upper right, and lower right of the global origin. Taking the direction from the global origin to the positive right as the global x-axis positive direction, the direction from the global origin to the positive up as the global y-axis positive direction, and passing through the global origin, perpendicular to the global x-axis and global y-axis, and pointing outside the grid plane as the global z-axis positive direction, a global coordinate system of the target area is constructed; For each measurement point, there is a known global coordinate and a target position vector, then the global target position of a single measurement point is: ； Among them, represents the global target position of the th measurement point, represents the global coordinates of the th measurement point; The average global target position is: ； Among them, represents the average global target position.

6. The transient electromagnetic near-surface target feature recognition system according to claim 1, characterized in that: In the initial position optimization and positioning module, the principle for updating the target position is: The residual matrix is: ； Among them, represents the residual matrix, represents the response voltage of the measurement point, represents the response voltage predicted by the single-dipole model; The Jacobian matrix is: ； Among them, represents the Jacobian matrix; The iterative equation of the target position is: ； ； Among them, represents the average global target position at the -th iteration, represents the average global target position at the -th iteration, represents the index of the number of iterations, and , represents the maximum number of iterations. When , , represents the correction amount for the average global target position at the -th iteration.

7. A transient electromagnetic near-surface target feature recognition system according to claim 1, characterized in that: In the target feature extraction module, the principle for generating the aspect ratio of the target is: The magnetic gradient tensor in the area where the positioning result is located is: ； Among them, represents the magnetic gradient tensor of the region where the positioning result is located, , , respectively represent the magnetic field components of the x, y, and z axes; Pair The formula for decomposing into three eigenvalues is as follows: Solve the equation , and obtain three eigenvalues , and , where represents the identity matrix; The aspect ratio is: ； Among them, represents the target aspect ratio.

8. The transient electromagnetic near-surface target feature recognition system according to claim 1, wherein: In the target feature extraction module, the principle for judging the conductivity level of the target based on the corrected decay time coefficient is: The formula for correcting the decay time coefficient is: ； Among them, represents the actual decay time coefficient, represents the decay time coefficient before correction, represents the empirical coefficient, represents the soil conductivity; Is directly proportional to the target conductivity.

9. The transient electromagnetic near-surface target feature recognition system according to claim 1, wherein: In the target clustering module, the principle for performing unsupervised clustering analysis is: The formula for calculating the Euclidean distance between all pairs of known target feature vectors is as follows: ； Among them, , respectively represent the o-th and p-th known target feature vectors, represents , the Euclidean distance between, and , represents the total number of known target feature vectors, represents the normalized aspect ratio of the o-th known target feature vector, represents the normalized aspect ratio of the p-th known target feature vector, represents the normalized actual decay time coefficient of the o-th known target feature vector, represents the normalized actual decay time coefficient of the p-th known target feature vector; Take the neighborhood radius of clustering as the 30th percentile of the Euclidean distances of all sample points, and the minimum number of samples for clustering is , which means taking the ceiling of the square root of the total number of known target feature vectors; randomly select a sample point, and count other sample points whose distances from it are within the neighborhood radius. If the number exceeds , then regard the selected sample point as the current clustering center, and assign all sample points to the cluster where the current clustering center closest to their own Euclidean distance is located, which is regarded as completing the unsupervised clustering analysis; The formula for calculating the cluster center in each cluster is as follows: ； Among them, represents the cluster center of the th cluster, represents the index of the cluster, represents the number of sample points in the th cluster, represents the index of the sample point in the cluster, and , represents the number of sample points in the th cluster, represents the target feature vector corresponding to the th sample point; The formula for generating the Mahalanobis distance from each feature vector in the cluster to the cluster center is as follows: ； ； Among them, represents the th Mahalanobis distance from the th sample point in the th cluster to the cluster center, and represents the within-class covariance matrix of the th cluster; When this sample point is removed from the cluster represents the Mahalanobis distance threshold.

10. A method for identifying transient electromagnetic near-surface target characteristics, characterized in that: The method is executed by the transient electromagnetic near-surface target feature recognition system described in any one of claims 1-9, and the specific steps include: Step 1: Arrange a 3×3 measurement grid in the target area, take the center point of each grid as a measurement point, and collect the response voltage components of each measurement point respectively; Step 2: Calculate the magnetic gradient tensor of the measurement point based on the response voltage components of each measurement point, calculate the target position vector of each measurement point based on the magnetic gradient tensor and the response voltage components, establish a global coordinate system in the measurement grid, generate the global target position of the measurement point based on the target position vectors of all measurement points, and take the average of the global target positions of all measurement points to generate the average global target position; Step 3: Construct a single dipole model at each measurement point, equivalent the underground target to a magnetic dipole with a dipole moment, and predict the response voltage components through the single dipole model; Step 4: Take the average global target position as the initial value of the Gauss-Newton algorithm, construct a residual matrix reflecting the difference between the actual response voltage and the response voltage predicted by the single dipole model, take the negative derivative of the residual matrix to generate the Jacobian matrix, start iterating from the initial value to update the average global target position until the maximum number of iterations is reached, and output the average global target position updated last time as the positioning result; Step 5: Obtain and decompose the magnetic gradient tensor in the area where the positioning result is located, generate the target aspect ratio, synchronously measure the soil conductivity, correct the decay time coefficient of the transient electromagnetic response through the soil conductivity to generate the actual decay time coefficient, and reflect the conductivity of the target through the actual decay time coefficient; Step 6: Collect the known target aspect ratio and actual decay time coefficient that appear near the surface in the historical data, perform Min-Max normalization on the aspect ratio and the actual decay time coefficient to generate the feature vector of each known target, calculate the Euclidean distance between all pairs of known target feature vectors, take the 30% quantile of the Euclidean distance as the neighborhood radius, and set the minimum number of samples for clustering. Perform unsupervised clustering analysis on the feature vectors of each known target based on the DBSCAN clustering algorithm. Each feature vector of a known target in the cluster is equivalent to a sample point. Calculate the sample point mean as the cluster center in each cluster, generate the Mahalanobis distance from each sample point in the cluster to the cluster center, set the Mahalanobis distance threshold. When the Mahalanobis distance from the sample point to the cluster center exceeds the Mahalanobis distance threshold, remove the sample point from the cluster. For each group of removed sample points, recalculate the cluster center and calculate the Mahalanobis distance from other sample points to the cluster center. Repeat the above steps of removing sample points until there are no sample points in the cluster whose Mahalanobis distance from the cluster center exceeds the Mahalanobis distance threshold; Step 7: Construct a deep learning network. Respectively take the known target feature vectors in different clusters as inputs, and the known target categories as labels. Train a target recognition model for each cluster. Calculate the feature vectors of the real-time located target according to the above steps. Select the cluster center with the closest Euclidean distance to the feature vector of the real-time located target, assign the feature vector of the real-time located target to the cluster corresponding to this cluster center, and input the feature vector of the real-time located target into the target recognition model of the corresponding cluster to identify the category of the real-time located target.

Images