Magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion

Abstract

The invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion, and belongs to the technical field of magnetic target detection, and the method comprises the steps: controlling a pre-carrying platform to circularly move according to a preset program set matched with an initial placement graph of a true target and a false target, starting a magnetic sensor to collect the data of a surrounding magnetic field in real time, performing multi-scale decomposition on the original magnetic anomaly data under each preset program by adopting a wavelet transform and sparse representation theory, extracting feature information under different scales, constructing a multi-feature fusion deep learning model, and performing multi-sequence feature extraction and multi-dimensional feature fusion on the feature information under each preset program to obtain a multi-feature fusion deep learning model; and performing result placement analysis on all the magnetic anomaly recognition results based on the initial placement graph to obtain a final recognition result, and outputting the final recognition result. And the false alarm rate of the magnetic anomaly target in the identification process is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic target detection, and in particular to a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. Background Art

[0002] Magnetic target detection is based on the identification and positioning of the geomagnetic field disturbance anomalies caused by the magnetic target itself. The detection platform is equipped with a magnetometer to collect magnetic field data, integrates intelligent recognition modules and related algorithms, detects and identifies magnetic anomalies and locates magnetic field changes, distinguishes magnetic anomalies, and then determines them as magnetic targets.

[0003] Magnetic anomaly curves are used in magnetic target exploration to show the deviation of the geomagnetic field strength measured along a specific survey line or path relative to the normal geomagnetic field strength. This graphical representation can intuitively reveal the location, shape, occurrence and possible distribution information of magnetic objects. Magnetic anomaly curves usually contain the following key elements: Horizontal coordinate: represents the geographical location or distance of the measurement point along the survey line. It can be a straight-line distance, longitude and latitude, or other location indicators in a geographic reference system.

[0004] The vertical axis represents the magnetic anomaly value, which is the measured geomagnetic field intensity minus the theoretical geomagnetic field intensity (normal field) in the area. The unit is usually nanotesla (nT). A positive value indicates an increase in the magnetic field, and a negative value indicates a decrease.

[0005] Curve shape: The shape of the magnetic anomaly curve can reflect the characteristics of the magnetic object. For example, simple magnetic bodies such as cylinders, plates or spheres will produce specific patterns on the curve, such as peaks, depressions or complex double peak structures.

[0006] like Figure 1 The figure shows a partial screenshot of the magnetic anomaly curve under ideal conditions, with a good signal-to-noise ratio and a close-range test of the magnetometer. In aerial surveys, the magnetometer is far away from the target, and due to the carrier itself, the dynamic noise is relatively large. The target signal is submerged in the large background of the geological body and is difficult to identify. Based on modern signal processing methods, the total field data collected is subjected to magnetic compensation, denoising, geomagnetic field correction, filtering, etc. to obtain a magnetic anomaly curve. Due to various environmental interferences, platform dynamic noise and data processing methods, there is often a lot of false target information, which has a great impact on the accuracy of magnetic target recognition, making the false alarm rate of magnetic anomaly targets too high during the recognition process.

[0007] Therefore, the present invention proposes a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. Summary of the invention

[0008] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion to solve the above-mentioned technical problems.

[0009] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion, including: Step 1: After placing the true target and the false target, control the pre-mounted platform to move in a loop according to a preset program set that matches the initial placement diagram of the true target and the false target. Meanwhile, during the movement of the pre-mounted platform, start the magnetic sensor to collect the surrounding magnetic field data in real time, and store the original magnetic field data based on each preset program respectively, where the preset program is related to the pitch, roll, turning, acceleration, descent, lift, and uniform speed states of the pre-mounted platform, and the state combination order under different preset programs is different; Step 2: Adopt wavelet transform and sparse representation theory to perform multi-scale decomposition on the original magnetic anomaly data under each preset program, and extract the feature information at different scales; Step 3: Construct a multi-feature fusion deep learning model, and perform multi-sequence feature extraction and multi-dimensional feature fusion on the feature information under each preset program respectively to obtain the magnetic anomaly recognition result; Step 4: Perform result placement analysis on all magnetic anomaly recognition results based on the initial placement diagram to obtain the final recognition result and output it.

[0010] Preferably, before controlling the movement of the pre-mounted platform, it further includes: Perform continuous movement on the involved motion states to construct a first motion sequence; Perform discontinuous movement on the involved motion states to construct a second motion sequence; Based on the first motion sequence and the second motion sequence, form a cyclic motion; Configure corresponding preset programs for each motion sequence to obtain a preset program set based on the cyclic motion; Among them, each motion state has a pre-set motion trajectory program, and the motion trajectory program is related to the initial placement diagram.

[0011] Preferably, performing continuous movement on the involved motion states to construct a first motion sequence includes: Based on the standard motion curves under different set motion states, and obtaining the environmental noise interference coefficient under each motion state according to the motion speed of each motion point in the standard motion curve; Sort all the environmental noise interference coefficients from large to small to obtain a first sub-sequence; Obtain the first vector of the starting point and the second vector of the ending point of the standard motion curve under different set motion states, and separately calculate a third vector from the first vector and the second vector, where the third vector includes: a vector direction and a vector magnitude; Sort the vector magnitudes under all motion states from largest to smallest to obtain a second sub-sequence; Sort the vector directions under all motion states clockwise with the y-axis as the reference line to obtain a third sub-sequence; Take all starting points as the first origin and place them overlappingly, globally calculate a fourth vector from all the first vectors, assign a first coefficient to each first vector according to the angular difference and length difference between the fourth vector and each first vector, and sort all the first coefficients from largest to smallest to obtain a fourth sub-sequence; Take all ending points as the second origin and place them overlappingly, globally calculate a fifth vector from all the second vectors, assign a second coefficient to each second vector according to the angular difference and length difference between the fifth vector and each second vector, and sort all the second coefficients from largest to smallest to obtain a fifth sub-sequence; Based on the first sub-sequence, the second sub-sequence, the third sub-sequence, the fourth sub-sequence, and the fifth sub-sequence, form a first motion sequence.

[0012] Preferably, the characteristic information includes: time-domain characteristics, image characteristics, and frequency-domain characteristics.

[0013] Preferably, construct a multi-feature fusion deep learning model, including: Extract features from historical magnetic anomaly data, divide them into historical time-domain features and historical frequency-domain features, use a recurrent neural network to model the features of historical magnetic anomaly data at different scales and perform convolutional backbone network processing to obtain first feature domain data; Perform two-dimensional convolutional backbone neural network processing on the image characteristics of the time-frequency images of historical magnetic anomaly data to obtain second feature domain data; Based on the interpolation method, align the feature domain data under the same historical magnetic anomaly data in the time and space dimensions, and map them to a unified feature space, and combine the attention mechanism to enhance the attention to important regions and important features to achieve data fusion; Perform neural network target recognition on the fused data, add an adaptive optimization algorithm, introduce an adaptive weight allocation mechanism and a dynamic learning rate adjustment strategy, and automatically optimize the network parameters according to the difficulty of the data and the performance of the model during the training process until the set standard is met to obtain a multi-feature fusion deep learning model.

[0014] Preferably, before aligning the feature domain data under the same historical magnetic anomaly data in the time and space dimensions based on the interpolation method, it further includes: Sort the first feature domain data by time and lock the missing time points; If the number of missing time points is 0, at this time, keep the first feature domain data unchanged; If the number of missing time points is multiple, at this time, lock the first data, the second nearest first data, and the third nearest first data that have a time distance from each missing time point; Count the first number of missing time points between the first data and the second data, and the second number of missing time points between the second data and the third data. At the same time, count the first missing continuous frequency in the data sorted by time; Based on the average values of the first data, the second data, and the third data, calculate the data corresponding to the missing time points and perform one interpolation.

[0015] Preferably, before analyzing the result placement of all magnetic anomaly recognition results based on the initial placement map, it includes: Capture the actual motion process parameter set and the standard motion process parameter set of the pre-carried platform under different preset programs based on each motion state, and determine the motion difference; Obtain the first difference between the second vector of the first motion state and the first vector of the second motion state, and the second difference between the first vector of the second motion state and the second vector in two adjacent motion states under each preset program; Based on the motion difference, the first difference, and the second difference in the same motion state, construct an instantaneous function corresponding to the motion state, and set a stuttering coefficient for the corresponding motion combination state based on the instantaneous function, where the motion combination state includes two different motion states; Assign the stuttering coefficient to each motion state of the corresponding motion combination state; Compare the assignment result with the state combination order under each preset program, and supplement the analysis of the magnetic anomaly recognition result under the corresponding preset program.

[0016] Preferably, analyzing the result placement of all magnetic anomaly recognition results based on the initial placement map includes: Lock the positions of the true target and the false target in the initial placement map, and place the magnetic anomaly recognition results under each sequential motion one-to-one corresponding to the corresponding locked positions to obtain an initial recognition matrix, and standardize the initial recognition matrix according to the relative position relationship; According to the covered line segments of the initial placement map in each motion state under each preset program, and at the same time, lock the positions of the true target and the false target in the initial placement map; According to the locked target under the connecting line segment of two adjacent covering line segments and combined with the jamming coefficient under the corresponding motion combination state, the standardized result of the locked position of the corresponding locked target is adjusted, and the average value of each column vector is calculated to obtain the final recognition result.

[0017] Compared with the prior art, the present invention has the following beneficial effects: By allowing the pre-mounted platform to move and collect data according to different preset programs, magnetic field data in multiple perspectives and multiple motion states can be obtained, enriching the data dimension and providing a more comprehensive data basis for the subsequent accurate identification of magnetic anomaly targets. The multi-feature fusion deep learning model can fully mine the features of different aspects of magnetic anomaly data. Multi-sequence feature extraction and multi-dimensional feature fusion make the model's understanding of magnetic anomaly data more comprehensive and in-depth, improving the accuracy and reliability of magnetic anomaly target identification.

[0018] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0019] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 is a schematic diagram of a magnetic anomaly curve under ideal conditions in an embodiment of the present invention; Figure 2 It is a flow chart of a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion in an embodiment of the present invention; Figure 3 This is a structural diagram of missing continuous frequencies in an embodiment of the present invention; Figure 4 This is a structural diagram of a multi-feature fusion deep learning model in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0022] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion, such as Figure 2 As shown, including: Step 1: After arranging the real target and the false target, control the pre-mounted platform to move in a loop according to a preset program set that matches the initial layout diagram of the real target and the false target. Meanwhile, during the movement of the pre-mounted platform, start the magnetic sensor to collect the surrounding magnetic field data in real time, and store the original magnetic field data based on each preset program respectively. Among them, the preset program is related to the pitch, roll, turning, acceleration, descent, lift, and uniform motion states of the pre-mounted platform, and the state combination sequences under different preset programs are different; Step 2: Adopt wavelet transform and sparse representation theory to perform multi-scale decomposition on the original magnetic anomaly data under each preset program, and extract the feature information at different scales; Step 3: Construct a multi-feature fusion deep learning model, and perform multi-sequence feature extraction and multi-dimensional feature fusion on the feature information under each preset program respectively to obtain the magnetic anomaly recognition result; Step 4: Perform result placement analysis on all magnetic anomaly recognition results based on the initial layout diagram to obtain the final recognition result and output it.

[0023] In this embodiment, the pre-mounted platform selects a carrier with multiple motion modes. The mounted platform is equipped with a high-precision attitude control and motion control system, and the platform motion is controlled based on a preset program set that matches the initial layout diagram of the real target and the false target. The preset program is related to the pitch (rotating up and down around the horizontal axis), roll (rolling left and right around the longitudinal axis), turning (rotating around the vertical axis), acceleration, descent, and lift states of the platform, and the state combination sequences in different programs are different. For example, for an aerial detection scenario, program 1 can be set as the platform first flying horizontally at a certain speed (without pitch and roll), then performing a small-angle pitch-up, and then turning; program 2 can be set as first rolling a certain angle, and then accelerating and descending, etc. This is to enable the magnetic sensor to collect magnetic field data from different angles and motion states to ensure the accuracy of the acquired data.

[0024] In this embodiment, the magnetic sensors are distributed at different positions on the mounted platform to comprehensively collect the magnetic field information of the surrounding space. Such as Hall sensors, giant magnetoresistance sensors, fluxgate sensors, etc. For example, high-precision fluxgate sensors are used to collect data at a frequency of 100 times per second. The collected data contains the magnetic field information of the real target, the false target, and the surrounding environment, and the original magnetic field data based on each preset program is stored respectively for subsequent classification and analysis.

[0025] In this embodiment, a special target object placement device is set in the area where the mounted platform moves. This device can flexibly arrange various types of false targets and real target objects according to experimental requirements. The false targets can be non-magnetic or weakly magnetic objects of different shapes and materials as interference objects; the real targets are target objects with specific magnetic characteristics, such as ferromagnetic metal blocks, magnetic weapon equipment, etc.

[0026] In this embodiment, after the true target and the false target are placed, since their positions are determined, a placement result diagram, that is, an initial placement diagram, will be obtained.

[0027] In this embodiment, during the operation of the control platform, due to some special conditions during the movement, such as pitching, rolling, turning, accelerating, descending, and lifting, the sudden change of the state will affect the magnetic field to a certain extent. When the carrying platform performs movements such as accelerating, decelerating, and turning, its internal components such as motors and circuits will generate electromagnetic signals. Taking an unmanned aerial vehicle as an example, electromagnetic noise will be generated when the motor rotates at high speed and the electronic speed controller works, and these noises will be superimposed on the magnetic field signal of the magnetic target, causing the measurement result to deviate from the true value. The change of the motion state, especially during the processes of acceleration, lifting, and descending, the carrying platform will generate mechanical vibrations of different degrees. The vibration will cause the magnetometer and magnetic sensor installed on the platform to displace or shake, changing the measurement direction and position of the sensor, thereby affecting the accuracy of the measurement result. For example, on a vehicle carrying platform, the bumpy vibration when passing through an uneven road surface will interfere with the stable measurement of the magnetic target magnetic field by the sensor. Attitude changes such as pitching and rolling will change the relative position and angle between the sensor and the magnetic target. When the carrying platform performs pitching motion, the magnetic field component of the magnetic target measured by the sensor will change. If accurate attitude compensation is not performed, measurement errors will occur. It should be noted that the preset program for each motion state is preset. For example, in the acceleration state, it advances at an acceleration of 0.1 square meters per meter in the horizontal direction, and the advancing time is 2 seconds.

[0028] Therefore, it is necessary to determine the cyclic motion through the combination design of states. For example, there are 6 combination designs. At this time, controlling the movement of the carrying platform in sequence according to combination 1, combination 2, combination 3, combination 4, combination 5, and combination 6 is a cyclic motion until the carrying platform finishes executing combination 6.

[0029] In this embodiment, for a certain segment of original magnetic anomaly data collected by program A in step 1, assume it is a discrete signal sequence with a length of 1000. After 3-layer decomposition using the Daubechies-4 wavelet basis, 3 groups of high-frequency detail coefficients and 1 group of low-frequency approximation coefficients are obtained. Then, using the orthogonal matching pursuit algorithm, for example, 100 key coefficients are selected from these coefficients. These key coefficients represent the main feature information of this segment of magnetic anomaly data at different scales, realizing the multi-scale feature extraction of the data. Through wavelet transform and sparse representation theory, the feature information at different scales is extracted from the original magnetic anomaly data, removing redundant information and highlighting key features, enabling the subsequent model to process data more efficiently and improving the accuracy of magnetic anomaly target recognition.

[0030] In this embodiment, for step 2, it specifically includes: Normalize the original magnetic anomaly data to ensure the comparability of features at different scales and eliminate the influence of dimension.

[0031] Apply techniques such as low-pass filters to remove high-frequency noise and retain the essential features of the signal.

[0032] Use wavelet transform to decompose the magnetic data into components at different scales. Each scale corresponds to a different frequency bandwidth, and features at different time scales can be extracted.

[0033] Use empirical mode decomposition (EMD) to adaptively decompose the signal into a series of intrinsic mode functions (IMFs). Each IMF represents a fluctuation mode at a certain time scale.

[0034] Based on the subsequent processing of EMD, combine Hilbert transform to analyze the frequency characteristics of each IMF and further refine the multi-scale analysis.

[0035] Calculate statistics such as mean, variance, maximum value, minimum value, kurtosis, and skewness at different scales to extract the central tendency, dispersion degree, and distribution form of the magnetic data.

[0036] Through the spectral analysis of Fourier transform and wavelet transform, draw a time-frequency diagram, distinguish the energy with different colors, generate an RGB image, and extract the corresponding distribution features.

[0037] In this embodiment, as Figure 4 shown is a multi-feature fusion deep learning model. Design a new deep neural network architecture, combine convolutional neural network (CNN) and recurrent neural network (RNN). Use CNN to capture local spatial features and RNN to capture the temporal or spatial correlation of magnetic data sequences, improving the model's ability to understand complex target features. Use an attention-based feature fusion method to fuse time-domain features, frequency-domain features, and image features, and perform neural network recognition on the fused features to output the recognition result.

[0038] In this embodiment, during the result placement analysis process, it is necessary to perform proportional adjustment of the magnetic data according to the positional relationship to facilitate the comprehensive processing of the magnetic data at the same position and ensure the accuracy of the measurement and obtained results.

[0039] The beneficial effects of the above technical solution are as follows: By allowing the pre-mounted platform to move according to different preset programs and collect data, magnetic field data in multiple perspectives and multiple motion states can be obtained, enriching the data dimension and providing a more comprehensive data basis for accurately identifying magnetic anomaly targets in the subsequent process. The multi-feature fusion deep learning model can fully exploit the features of magnetic anomaly data in different aspects. The multi-sequence feature extraction and multi-dimensional feature fusion enable the model to understand the magnetic anomaly data more comprehensively and deeply, improving the accuracy and reliability of magnetic anomaly target recognition and reducing the false alarm rate during the recognition process of magnetic anomaly targets.

[0040] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. Before controlling the movement of the pre-mounted platform, it further includes: Performing continuous movement on the involved motion states to construct a first motion sequence; Performing discontinuous movement on the involved motion states to construct a second motion sequence; Based on the first motion sequence and the second motion sequence, a cyclic motion is formed; Configuring corresponding preset programs for each motion sequence to obtain a preset program set based on the cyclic motion; Wherein, there is a preset motion trajectory program for each motion state, and the motion trajectory program is related to the initial placement diagram.

[0041] In this embodiment, the first motion sequence includes 5 sequences. Discontinuous movement refers to controlling the pre-mounted platform to operate separately according to each motion state as a most basic reference.

[0042] In this embodiment, continuous movement means that after the involved motion states run for a set time interval respectively, they immediately change to another state. It should be noted that there are only the mentioned motion states.

[0043] The beneficial effects of the above technical solution are as follows: By forming a cyclic motion through continuous movement and discontinuous movement, and through program configuration, reasonable control of the pre-mounted platform is achieved.

[0044] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. Performing continuous movement on the involved motion states to construct a first motion sequence includes: Based on the standard motion curves under different set motion states, and obtaining the environmental noise interference coefficient for each motion state according to the motion speed of each motion point in the standard motion curve; Sorting all the environmental noise interference coefficients from large to small to obtain a first sub-sequence; Obtain the first vector of the starting point and the second vector of the ending point of the standard motion curve under different set motion states, and perform separate calculations on the first vector and the second vector to obtain a third vector, where the third vector includes: vector direction and vector magnitude; Sort the vector magnitudes under all motion states from largest to smallest to obtain a second sub-order; Sort the vector directions under all motion states clockwise with the y-axis as the reference line to obtain a third sub-order; Take all starting points as the first origin and place them overlappingly, perform a global calculation on all first vectors to obtain a fourth vector, assign a first coefficient to each first vector according to the angular difference and length difference between the fourth vector and each first vector, and sort all first coefficients from largest to smallest to obtain a fourth sub-order; Take all ending points as the second origin and place them overlappingly, perform a global calculation on all second vectors to obtain a fifth vector, assign a second coefficient to each second vector according to the angular difference and length difference between the fifth vector and each second vector, and sort all second coefficients from largest to smallest to obtain a fifth sub-order; Based on the first sub-order, second sub-order, third sub-order, fourth sub-order, and fifth sub-order, form a first motion order.

[0045] Preferably, the characteristic information includes: time-domain characteristics, image characteristics, and frequency-domain characteristics.

[0046] In this embodiment, the calculation of the environmental noise interference coefficient is as follows: ; Among them, max represents the maximum symbol; Zg represents the environmental noise interference coefficient under the corresponding motion state; N1 represents the number of motion points involved in the corresponding standard motion curve; 、 、 respectively represent proportional constants, which are preset; represents the derivative of the velocity vi with respect to time ti at the i1-th motion point; represents the motion curvature at the i1-th motion point i; fi represents the motion friction at the i1-th motion point; In this embodiment, 、 、 are preset to adapt to different motion states, that is, the proportional constants under different motion states are different. For example, in the accelerating motion state: = , , It should be noted that the standard motion curves under different motion states are set, and the programs corresponding to these curves are also set, in order to control the motion of the carrying platform according to the preset program.

[0047] In this embodiment, under the actual motion state, several factors such as speed change, motion curvature, and motion friction may all generate environmental noise interference, but usually the noise generated by a certain one of these factors dominates. For example, when the pre-carrying platform starts or brakes quickly, the noise generated by the speed change rate may be much greater than the noise generated by other factors; while when the platform performs large-curvature motions such as sharp turns, the noise generated by the motion curvature may be the most prominent. By taking the maximum value, it is possible to accurately capture the factor that contributes the most to the environmental noise interference under a specific motion state, and use it as a key indicator representing the noise interference level of the entire motion state, avoiding being interfered by other relatively minor factors in the judgment of the overall noise level. If the method of taking the maximum value is not adopted, but the three factors are simply added together or other complex operations are performed, on the one hand, it will increase the computational complexity, and on the other hand, it may not clearly and explicitly reflect the main source of interference. The operation of taking the maximum value is relatively simple and direct, and can quickly determine which factor generates the key noise under the current motion state. Moreover, the obtained maximum value can intuitively reflect the upper limit level of the environmental noise interference, and use a clear numerical value to represent the severity of the environmental noise interference in this motion state, which is convenient for subsequent comparison and analysis. For example, when comparing the noise interference situations under different motion programs, by comparing the maximum values, it can be quickly judged which motion program is more severely affected by noise. In actual situations, the contributions of different factors to the noise interference may vary greatly, and the magnitudes of these factors may also be different. If the strategy of taking the maximum value is not adopted, but operations such as averaging them are performed, it will cover up the truly dominant factor, resulting in the calculated environmental noise interference coefficient not being able to truly reflect the actual interference situation. Taking the maximum value can ensure that in various complex situations, the most critical interference factor can be accurately captured, making the calculated environmental noise interference coefficient more effective and representative.

[0048] For example, the code for the acceleration state is as follows: / / Initial state of the carrying platform let x = 50; let y = 50; let vx = 0; let vy = 0; const ax = 0.1; / / Horizontal acceleration const ay = 0.1; / / Vertical acceleration.

[0049] In this embodiment, the standard motion curves under different motion states are pre-set and can be directly used. They are stored in the state-curve comparison table, which contains the standard motion curves under different motion states. That is, at this motion state, the velocity directions and magnitudes executed at each time point are different, but they are all known and pre-set. For example, at the first time point in the acceleration state, the velocity direction is the positive 90° direction and the velocity magnitude is 5 cm / s. Therefore, the first vector and the second vector can be directly determined. In the vector calculation process, through the vector integration situation designed in the technical solution, the third vector, the fourth vector, and the fifth vector are obtained. It should be noted that the first vector and the second vector are the motion velocities and directions of the first moving point and the last moving point in the corresponding curve. The motion velocity corresponds to the vector magnitude, and the motion direction corresponds to the vector direction. And the third vector is obtained by vector addition (vector integration) of the first vector and the second vector respectively based on the origin placement in the coordinate system.

[0050] In this embodiment, clockwise sorting means sorting the clockwise angles from large to small by calculating the clockwise angles between each vector direction and the y-axis, and then the third sub-sequence can be obtained.

[0051] In this embodiment, global calculation is to perform vector calculation on all the first vectors related to the first origin after overlapping placement, and then the fourth vector can be obtained.

[0052] In this embodiment, the angle difference is the included angle between the fourth vector and the first vector; the length difference is the difference in the velocity magnitudes between the fourth vector and the first vector.

[0053] In this embodiment, , and the calculation method of the second coefficient is similar to that of the first coefficient, which will not be elaborated here.

[0054] The beneficial effects of the above technical solution are: by analyzing the interference coefficient, the magnitudes and directions of the vectors corresponding to the first point and the last point, the states in different situations are sorted to obtain the first motion sequence in the continuous state.

[0055] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion, which constructs a multi-feature fusion deep learning model, including: Feature extraction is performed on historical magnetic anomaly data, which is divided into historical time-domain features and historical frequency-domain features. A recurrent neural network is used to model the features of historical magnetic anomaly data at different scales and perform convolutional backbone network processing to obtain the first feature domain data; The image features of the time-frequency image of historical magnetic anomaly data are processed by a two-dimensional convolutional backbone neural network to obtain the second feature domain data; Based on the interpolation method, the feature domain data under the same historical magnetic anomaly data are aligned in the time and space dimensions and mapped to a unified feature space. Combining the attention mechanism enhances the attention to important regions and important features to achieve data fusion; Perform neural network target recognition on the fused data. Add an adaptive optimization algorithm, introduce an adaptive weight allocation mechanism and a dynamic learning rate adjustment strategy, and automatically optimize the network parameters according to the difficulty of the data and the model performance during the training process until the set standard is met to obtain a multi-feature fusion deep learning model.

[0056] In this embodiment, the historical magnetic anomaly data are the magnetic field data collected by devices such as magnetic sensors in a specific time period and specific region in the past, which are different from the normal magnetic field situation. These differences may be caused by the presence of magnetic target objects (such as ferromagnetic ores, magnetic metal products, etc.).

[0057] In this embodiment, the historical time-domain features are the features extracted from the historical magnetic anomaly data that reflect the variation law of the magnetic field characteristics over time. For example, the change rate of the magnetic field intensity at different times, the periodic change situation, etc. A certain period of historical magnetic anomaly data records the magnetic field intensity values per second.

[0058] In this embodiment, the historical frequency-domain features are the features presented about the distribution of magnetic field energy on different frequency components after converting the historical magnetic anomaly data from the time domain to the frequency domain. It can reveal information such as the signal intensity and dominant frequency of different frequencies in the magnetic field change. Perform Fourier transform on the above historical magnetic anomaly data recorded per second to obtain its representation in the frequency domain. It is found that the magnetic field energy is relatively concentrated near the 5Hz frequency in the data. The characteristic of the concentrated energy near 5Hz and the energy distribution at other frequencies are the historical frequency-domain features.

[0059] In this embodiment, the first feature domain data is the data representation in a specific feature space obtained after the historical magnetic anomaly data are processed by a recurrent neural network modeling and a convolutional backbone network. It contains the time-domain and frequency-domain related features of the magnetic anomaly data after processing and abstraction. First, the RNN is used to capture the time correlation, and then the convolutional backbone network is used to extract local features to obtain a new set of data vectors. This set of data vectors constitutes the first feature domain data, which is no longer the original magnetic anomaly data, but a representation that can better reflect the data features.

[0060] In this embodiment, the time-frequency image is an image obtained by converting the magnetic anomaly data in the time domain to the time-frequency plane through a specific transformation (such as short-time Fourier transform, wavelet transform, etc.). The abscissa of the image represents time, the ordinate represents frequency, and the pixel value of the image reflects information such as the magnetic field energy or amplitude at the corresponding time and frequency. The short-time Fourier transform is performed on the continuously collected magnetic anomaly data over a period of time, and the magnetic field data at each time point is decomposed into different frequencies. Then, according to the energy magnitude of each frequency component, it is plotted on a two-dimensional plane with different grayscales or colors to form the time-frequency image.

[0061] In this embodiment, the second feature domain data is the data representation in a specific feature space obtained after processing the time-frequency image of historical magnetic anomaly data through a two-dimensional convolutional backbone neural network. It mainly reflects the feature information of the magnetic anomaly data on the time-frequency image. After extracting features from the time-frequency image of the magnetic anomaly data through the two-dimensional convolutional backbone neural network, a new set of data feature vectors is obtained. These vectors form the second feature domain data, which is a further abstraction and representation of the time-frequency image features.

[0062] In this embodiment, the Adam optimization algorithm combines the ideas of momentum and adaptive learning rate. When training a multi-feature fusion deep learning model, the Adam algorithm automatically adjusts its learning rate according to the gradient change of each parameter during the training process. For parameters with large gradient changes, a smaller learning rate is used, and for parameters with small gradient changes, a larger learning rate is used, thereby accelerating the model convergence speed.

[0063] In this embodiment, during the training of the magnetic anomaly target recognition model, for the connection weights corresponding to the key features that can accurately distinguish magnetic targets and non-magnetic targets, the adaptive weight allocation mechanism increases their weight values to make the model pay more attention to these features, while reducing the weights corresponding to the features that contribute less to target recognition.

[0064] The beneficial effects of the above technical solutions are as follows: The adaptive weight allocation mechanism increases their weight values to make the model pay more attention to these features, while reducing the weights corresponding to the features that contribute less to target recognition. When training a multi-feature fusion deep learning model, a larger learning rate is set at the beginning to allow the model to quickly update parameters. As the training progresses, if it is found that the loss function value decreases slowly or fluctuates, the learning rate is reduced to enable the model to perform more detailed parameter adjustment in the region closer to the optimal solution.

[0065] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. Before aligning the feature domain data under the same historical magnetic anomaly data in the time and space dimensions based on the interpolation method, it further includes: Sort the data in the first feature domain by time and lock the missing time points; If the number of missing time points is 0, at this time, keep the data in the first feature domain unchanged; If the number of missing time points is more than one, at this time, lock the first data, the second nearest first data, and the third nearest first data that have a time distance from each missing time point; Count the first number of missing time points between the first data and the second data, and the second number of missing time points between the second data and the third data. At the same time, count the first missing consecutive frequency in the data sorted by time; Based on the average values of the first data, the second data, and the third data, calculate the data corresponding to the missing time points and perform one interpolation.

[0066] ; Among them, d1, d2, and d3 respectively represent the values of the first data, the second data, and the third data; e1 represents a set threshold, and the value is 0.1; represents the first number; represents the second number; n represents the total number of values in the data sorted by time; Lp represents the first missing consecutive frequency; pN represents the total number of missing time points involved; It should be noted that as long as the missing time points are continuous and the continuous quantity is greater than or equal to 2, at this time, it is regarded as the consecutive frequency. For example, Figure 3 As shown, there are 10 time points, and the circles represent the non-missing time points, and the squares represent the missing time points. At this time, 6 and 7, 9 and 10 are both continuously missing, the value of Lp is 2, and the value of pN is 5.

[0067] In this embodiment, the exponential decay term can dynamically adjust the calculation result according to the relevant parameters of the missing time points. While ensuring a certain smoothness, it can more accurately estimate the missing data, reduce the interference of improper handling of missing data on subsequent analysis, and improve the overall quality of the data. It effectively supplements the missing part in the first feature domain data and makes the data more complete in the time dimension. Complete data is crucial for subsequent analysis and modeling based on this feature domain data (such as inputting into a deep learning model for magnetic anomaly recognition), etc. It can improve the quality of the model input data and then improve the accuracy and reliability of the analysis results.

[0068] In this embodiment, under normal circumstances, there will be corresponding feature values at each time point. Therefore, by sorting the data in the first feature domain (feature values) in time order, it is possible to directly determine whether there are missing time points.

[0069] In this embodiment, the first data, the second data, and the third data involved all refer to the corresponding feature values.

[0070] The beneficial effects of the above technical solution are as follows: By sorting the data by time to directly lock the missing time points, and then based on three data, combined with the quantity and the frequency of consecutive missing data, effective and reasonable interpolation is achieved to ensure the integrity of the data and provide a basis for subsequent analysis.

[0071] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. Before analyzing the results of all magnetic anomaly recognition results based on the initial placement map, it includes: Capturing the actual instantaneous process parameter sets and standard instantaneous process parameter sets of the pre-carried platform under different preset programs based on each motion state to determine the motion differences; Obtaining the first difference between the second vector of the first motion state and the first vector of the second motion state and the second difference between the first vector of the second motion state and the second vector in two adjacent motion states under each preset program; Based on the motion differences, the first difference, and the second difference in the same motion state, constructing an instantaneous function corresponding to the motion state, and setting a lag coefficient for the corresponding motion combination state based on the instantaneous function, where the motion combination state includes two different motion states; Assigning the lag coefficient to each motion state of the corresponding motion combination state; Comparing the assignment results with the state combination order under each preset program, and supplementing the analysis of the magnetic anomaly recognition results under the corresponding preset program.

[0072] In this embodiment, the first difference and the second difference refer to the differences in speed and direction on the vector.

[0073] In this embodiment, the actual motion process parameter set refers to the operation parameters of the specified motion state at the initial operation moment, including: actual speed, actual direction. During actual flight, the actual speed of the unmanned aerial vehicle measured by the sensor is 4.8 m / s, and there is a deviation from the standard direction. Then, the motion difference in speed magnitude is 5 - 4.8 = 0.2 m / s, and the motion difference in speed direction is .

[0074] In this embodiment, the instantaneous function f(x) = g1 (motion differences between two states) + g2 (first difference) + g3 (second difference).

[0075] The lag coefficient = u1×g1 (motion differences between two states) + u2×g2 (first difference) + u3×g3 (second difference), where u1, u2, and u3 respectively represent weights, and the values are 0.5, 0.3, and 0.2.

[0076] In this embodiment, g1 (the motion difference in two states), g2 (the first difference), and g3 (the second difference) are obtained by matching from a dual-state-difference comparison table. The comparison table contains the motion differences, the first difference, the second difference, and their corresponding difference coefficients under different combined motion states, all of which are stored in advance and can be directly retrieved. By calculating the motion difference, vector difference, and setting the lag coefficient, it is found that there are relatively large lag coefficients in the motion state combinations under some preset programs. By comparing the magnetic anomaly recognition results in the cases of not considering the lag coefficient and considering the lag coefficient, it is found that after considering the lag coefficient, some misjudgment situations caused by abnormal platform motion can be identified. When the lag coefficient is not considered, the misjudgment rate of magnetic anomaly recognition is 15%; after considering the lag coefficient for supplementary analysis, the misjudgment rate is reduced to 10%.

[0077] The beneficial effects of the above technical solution are as follows: By determining the motion differences, the first difference, and the second difference of two different motion states, a sequential function is constructed to obtain the lag coefficient, the lag coefficient is set and a control analysis is carried out, and supplementary analysis and correction are carried out on the magnetic anomaly recognition results, thereby improving the accuracy and reliability of magnetic anomaly recognition and reducing the misjudgment risk caused by unstable platform motion.

[0078] The present invention provides a magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion. The result placement analysis is carried out on all magnetic anomaly recognition results based on the initial placement diagram, including: Locking the positions of the true target and the false target in the initial placement diagram, and placing the magnetic anomaly recognition results under each sequential motion in one-to-one correspondence with the corresponding locked positions to obtain an initial recognition matrix, and standardizing the initial recognition matrix according to the relative position relationship; According to the covered line segments of the initial placement diagram under each motion state of each preset program, at the same time, locking the positions of the true target and the false target in the initial placement diagram; According to the locked targets under the connecting line segments of two adjacent covered line segments, and combining the lag coefficient in the corresponding motion combination state, adjusting the standardized results of the locked positions of the corresponding locked targets, and calculating the average value of each column vector to obtain the final recognition result.

[0079] In this embodiment, in the initial placement diagram, the true target and the false target both have clear position coordinates. When the pre-mounted platform completes a sequential motion according to the preset program, a set of magnetic anomaly recognition results (judging which positions are true targets and which are false targets) will be obtained. Corresponding these recognition results with the locked positions of the targets in the initial placement diagram one by one, and placing the recognition results of each position (such as using 1 to represent the true target and 0 to represent the false target) at the corresponding positions of a matrix to form an initial recognition matrix.

[0080] Normalize the initial recognition matrix according to the relative position relationship. For example, arrange the rows and columns of the matrix in the coordinate order of the targets in the initial placement diagram to ensure that the position correspondence of the elements in the matrix is consistent with the actual positions of the targets in space. At the same time, perform operations such as normalization on the values in the matrix to make different recognition results comparable. For example, if the recognition results use values of different intensities to represent the possibility of the targets, these values can be normalized to the interval [0,1].

[0081] In this embodiment, it is assumed that there are 4 targets in the initial placement diagram, arranged in a 2×2 rectangle, with coordinates (1,1), (1,2), (2,1), and (2,2) respectively. After the pre-mounted platform completes one movement, the recognition result is that the position (1,1) is a true target (marked as 1), the position (1,2) is a false target (marked as 0), the position (2,1) is a false target (marked as 0), and the position (2,2) is a true target (marked as 1). Then the initial recognition matrix is When performing the normalization process, if the values are in other ranges, they can be normalized to [0,1] through linear transformation and other methods. Here, the matrix elements are already in the appropriate range, and only the position correspondence needs to be ensured accurately.

[0082] In this embodiment, under each preset program, each motion state of the pre-mounted platform will move within the corresponding spatial area of the initial placement diagram. According to the motion trajectory of the platform and the detection range of the sensor, the covered line segments of the initial placement diagram for each motion state are determined. For example, if the platform moves in a straight line and the sensor detection range is d distance on both sides, then the covered line segment is the line segment area within d range on both sides of the projection of the platform motion trajectory on the plane of the initial placement diagram. Still taking the initial placement diagram of the above 4 targets as an example, assume that a motion state of the pre-mounted platform under a certain preset program is a straight-line motion from left to right, and the sensor detection range is 0.5 unit distance on both sides. The projection of the platform motion trajectory on the plane of the initial placement diagram is a straight line in the x-axis direction, passing through the area from x = 0.5 to x = 2.5. Then the covered line segment is the line segment area on the x-axis from 0.5 to 2.5 and in the y-axis direction from -0.5 to 0.5. At the same time, lock the coordinates of the 4 targets (1,1), (1,2), (2,1), and (2,2).

[0083] Adjust the standardized result according to the target locked under the connecting line segment of two adjacent coverage line segments, in combination with the jitter coefficient in the corresponding motion combination state. If the jitter coefficient is large, it indicates that the platform motion is unstable in this motion combination state, which may affect the recognition result. At this time, correct the standardized result of the locked target in this area. For example, appropriately reduce the credibility value recognized as a true target (such as adjusting the value originally recognized as a true target from 0.8 to 0.6), and calculate the average value of each column vector of the initial recognition matrix. Each column vector represents the recognition results in different motion states or programs at the same column position. By calculating the average value, multiple results can be integrated to obtain a more accurate and reliable final recognition result. For example, if a column vector is [0.6, 0.4, 0.8], the calculated average value is (0.6 + 0.4 + 0.8) / 3 = 0.6. Determine whether the position is a true target or a false target based on this average value (such as setting a threshold of 0.5, and judging as a true target if it is greater than 0.5).

[0084] Suppose there is a locked target under the connecting line segment of two adjacent coverage line segments, and its standardized result in the initial recognition matrix is recognized as a true target (the value is 0.8), but the jitter coefficient of the corresponding motion combination state at this position is large. Adjust its value to 0.6. The initial recognition matrix is , calculate the average value of the column vectors. The average value of the first column is (0.6 + 0.8) / 2 = 0.7, and the average value of the second column is (0.4 + 0.2) / 2 = 0.3. Judging according to the threshold of 0.5, the corresponding position in the first column is finally recognized as a true target, and the corresponding position in the second column is finally recognized as a false target.

[0085] In this embodiment, by generating the initial recognition matrix, determining the coverage line segments, adjusting the result in combination with the jitter coefficient, and calculating the final recognition result, etc., compared with the recognition results without using these processing methods. It is found that after adopting this method, the recognition accuracy is significantly improved. For example, the accuracy rate is 70% before processing, and it is increased to 85% after processing.

[0086] The beneficial effects of the above technical solution are: it can systematically integrate the magnetic anomaly recognition result and the target position information, considering the influence of the jitter coefficient of the platform motion state on the recognition result. Through standardization processing and comprehensive calculation, it effectively improves the accuracy and reliability of magnetic anomaly target recognition, and reduces the situations of misjudgment and missed judgment.

[0087] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion, characterized in that Including: Step 1: After arranging the real target and the false target, control the pre-mounted platform to perform cyclic motion according to a preset program set that matches the initial arrangement diagram of the real target and the false target. Meanwhile, during the motion of the pre-mounted platform, start the magnetic sensor to collect the surrounding magnetic field data in real time, and store the original magnetic field data based on each preset program respectively. Among them, the preset program is related to the pitch, roll, turning, acceleration, descent, lift, and uniform motion states of the pre-mounted platform, and the state combination sequences under different preset programs are different; Step 2: Adopt wavelet transform and sparse representation theory to perform multi-scale decomposition on the original magnetic anomaly data under each preset program, and extract the feature information at different scales; Step 3: Construct a multi-feature fusion deep learning model, and perform multi-sequence feature extraction and multi-dimensional feature fusion on the feature information under each preset program respectively to obtain the magnetic anomaly recognition result; Step 4: Perform result placement analysis on all magnetic anomaly recognition results based on the initial arrangement diagram to obtain the final recognition result and output it.

2. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 1, characterized in that Before controlling the motion of the pre-mounted platform, it further includes: Performing continuous motion on the involved motion states to construct a first motion sequence; Performing discontinuous motion on the involved motion states to construct a second motion sequence; Based on the first motion sequence and the second motion sequence, form a cyclic motion; Configuring corresponding preset programs for each motion sequence to obtain a preset program set based on the cyclic motion; Among them, each motion state has a pre-set motion trajectory program, and the motion trajectory program is related to the initial arrangement diagram.

3. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 1, wherein Performing continuous motion on the involved motion states to construct a first motion sequence, including: Based on the standard motion curves under different set motion states, and obtaining the environmental noise interference coefficient under each motion state according to the motion speed of each motion point in the standard motion curve; Sorting all the environmental noise interference coefficients from large to small to obtain a first sub-sequence; Obtaining the first vector of the starting point and the second vector of the ending point of the standard motion curves under different set motion states, and separately calculating the third vector for the first vector and the second vector, where the third vector includes: vector direction and vector magnitude; Sorting the vector magnitudes under all motion states from large to small to obtain a second sub-sequence; Sorting the vector directions under all motion states clockwise with the y-axis as the reference line to obtain a third sub-sequence; Overlapping and placing all the starting points as the first origin, globally calculating the fourth vector for all the first vectors, assigning a first coefficient to each first vector according to the angle difference and length difference between the fourth vector and each first vector, and sorting all the first coefficients from large to small to obtain a fourth sub-sequence; Overlapping and placing all the ending points as the second origin, globally calculating the fifth vector for all the second vectors, assigning a second coefficient to each second vector according to the angle difference and length difference between the fifth vector and each second vector, and sorting all the second coefficients from large to small to obtain a fifth sub-sequence; Based on the first sub - order, the second sub - order, the third sub - order, the fourth sub - order, and the fifth sub - order, a first motion order is formed.

4. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 1, characterized in that, The feature information includes: time - domain features, image features, and frequency - domain features.

5. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 1, characterized in that, Constructing a multi - feature fusion deep - learning model includes: Performing feature extraction on historical magnetic anomaly data, which is divided into historical time - domain features and historical frequency - domain features. Using a recurrent neural network to model the features of historical magnetic anomaly data at different scales and processing them with a convolutional backbone network to obtain first - feature - domain data; Performing two - dimensional convolutional backbone neural network processing on the image features of the time - frequency images of historical magnetic anomaly data to obtain second - feature - domain data; Based on the interpolation method, align the feature - domain data under the same historical magnetic anomaly data in the time and space dimensions, and map them to a unified feature space. Combine the attention mechanism to enhance the attention to important regions and important features to achieve data fusion; Performing neural network target recognition on the fused data, adding an adaptive optimization algorithm, introducing an adaptive weight allocation mechanism and a dynamic learning rate adjustment strategy, and automatically optimizing the network parameters according to the difficulty of the data and the performance of the model during the training process until the set standard is met to obtain a multi - feature fusion deep - learning model.

6. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 5, characterized in that Before aligning the feature - domain data under the same historical magnetic anomaly data in the time and space dimensions based on the interpolation method, it also includes: Performing time sorting on the first - feature - domain data to lock the missing time points; If the number of missing time points is 0, at this time, keep the first - feature - domain data unchanged; If the number of missing time points is multiple, at this time, lock the first data, the second - nearest first data, and the third - nearest first data that have a time distance from each missing time point; Statistical the first number of missing time points between the first data and the second data, the second number of missing time points between the second data and the third data. At the same time, statistical the first missing continuous frequency in the data after time sorting; Based on the average values of the first data, the second data, and the third data, calculate the data corresponding to the missing time points and perform one - time interpolation.

7. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 1, wherein Before analyzing the result placement of all magnetic anomaly recognition results based on the initial placement map, it includes: Capturing the actual motion process parameter set and the standard motion process parameter set of the pre - carried platform under different preset programs based on each motion state to determine the motion difference; Obtaining the first difference between the second vector of the first motion state and the first vector of the second motion state and the second difference between the first vector of the second motion state and the second vector in two adjacent motion states under each preset program; Based on the motion difference, the first difference, and the second difference under the same motion state, constructing an instantaneous function corresponding to the motion state, and setting a lag coefficient for the corresponding motion combination state based on the instantaneous function, where the motion combination state includes two different motion states; Assigning the lag coefficient to each motion state of the corresponding motion combination state; Comparing the assignment result with the state combination order under each preset program and supplementing the analysis of the magnetic anomaly recognition results under the corresponding preset program.

8. The magnetic anomaly recognition method based on multi-sequence feature extraction and multi-dimensional feature fusion according to claim 7, wherein, Based on the initial placement diagram, the result placement analysis is carried out for all magnetic anomaly recognition results, including: Lock the positions of the true targets and false targets in the initial placement diagram, and place the magnetic anomaly recognition results under each sequential movement in one-to-one correspondence with the corresponding locked positions to obtain an initial recognition matrix, and standardize the initial recognition matrix according to the relative position relationship; According to the covered line segments of the initial placement diagram in each movement state under each preset program, at the same time, lock the positions of the true targets and false targets in the initial placement diagram; According to the locked targets under the connecting line segments of two adjacent covered line segments, and combined with the jamming coefficient in the corresponding movement combination state, adjust the standardized results of the locked positions of the corresponding locked targets, and calculate the average value of each column vector to obtain the final recognition result.

Images