Detection and identification method based on optical-magnetic fusion

Abstract

The invention designs a detection and identification method based on optical-magnetic fusion, and belongs to the technical field of target detection and identification. Optical imaging equipment and a magnetic measurement load are carried through a motion platform, optical image data and magnetic measurement data are collected in a task area, enhancement processing and target detection are carried out on the optical image data, and earth surface target information is extracted. Background magnetic field elimination and noise suppression are carried out on the magnetic measurement data, magnetic anomaly data are extracted, and a magnetic anomaly distribution diagram is generated. And superposing the magnetic anomaly distribution diagram with the optical image, carrying out spatial matching and target classification, carrying out magnetic field modeling on visible ferromagnetic targets, and stripping contribution from total magnetic anomaly to enhance signals of hidden targets. And performing target identification through the classification model, and outputting a target type, a position contour and magnetic characteristic parameters. The method integrates optical and magnetic anomaly detection technologies, improves the sensitivity and resolution of target detection in a complex environment, and is suitable for the fields of military reconnaissance, geological exploration, security monitoring and the like.

Description

Technical Field

[0001] This invention belongs to the field of target detection and recognition technology, specifically relating to a detection and recognition method based on optical-magnetic fusion. Background Technology

[0002] In the field of target detection and identification, magnetic anomaly detection technology and optical imaging technology are two types of technologies with important application value. However, when used alone, they both have certain limitations and it is difficult to achieve both high-sensitivity detection and adaptability to complex environments.

[0003] Magnetic anomaly detection technology detects magnetic anomalies by measuring the local distortion of the geomagnetic field caused by a target. It offers advantages such as all-weather operation and compatibility with various media (non-magnetic media: soil, forests, underwater, etc.). However, it suffers from limitations including limited detection range, susceptibility to background magnetic interference, and difficulty in complex environments. Typically, when the detection distance is 2-3 times greater than the target size, the target's magnetic anomaly can be equivalent to a magnetic dipole, and the magnetic anomaly signal attenuates with the cube of the distance. This leads to the signal from deep targets being easily overwhelmed by signals from shallow targets, and the similarity in magnetic anomaly morphology among different types of targets at the same elevation. These issues restrict the widespread application of magnetic anomaly detection technology in complex environments.

[0004] Optical imaging technologies (such as visible light, infrared, or laser imaging) achieve high-resolution visualization by capturing reflected or radiated light signals from a target. They are suitable for surface feature identification, but are susceptible to lighting conditions, weather (such as fog, rain, etc.) and obstructions, and cannot detect non-visible or hidden targets (such as underground or underwater objects).

[0005] In summary, traditional single-technology methods are insufficient to meet the application requirements of target detection and recognition in complex environments. There is an urgent need for a fusion technology solution to address issues such as sensitivity, resolution, and anti-interference in target detection and recognition under complex environments. Summary of the Invention

[0006] To address the problems existing in the background technology, the present invention provides a detection and identification method based on optical-magnetic fusion, comprising the following steps:

[0007] S1. Data Acquisition: Using a motion platform equipped with optical imaging equipment and a magnetic measurement payload, data is collected within the mission area according to the planned survey line to acquire optical image data and magnetic measurement data, wherein the magnetic measurement data includes measurement position data and measurement height data; then, the optical image data and the magnetic measurement data are spatially aligned using a spatial registration algorithm to generate a spatially aligned multimodal dataset;

[0008] S2. Optical data processing: The optical image data is subjected to image enhancement processing to generate an enhanced optical image; then, a target detection algorithm is used to extract the surface targets in the enhanced optical image, and the type and location information of each surface target are marked to generate an optical target detection result.

[0009] S3. Magnetic Anomaly Data Processing: Background magnetic field elimination and noise suppression are performed on the magnetic measurement data to extract magnetic anomaly data; then, the location and contour information of the magnetic anomaly target are obtained using a target localization algorithm; and finally, the magnetic anomaly data is converted into regular grid data through grid interpolation to generate a magnetic anomaly distribution map.

[0010] S4. Optical and magnetic data fusion processing: The magnetic anomaly distribution map is superimposed on the enhanced optical image; then, the optical target detection results are spatially matched with the positions of the magnetic anomaly targets, and the targets are initially classified according to the matching results; then, magnetic field modeling is performed on the visible ferromagnetic targets with large magnetic anomaly intensity, and the magnetic anomaly distribution generated by them in the task area is calculated; then, the magnetic anomaly distribution generated by the visible ferromagnetic targets is subtracted from the total magnetic anomaly data in the task area to obtain the remaining magnetic anomaly data, thereby enhancing the magnetic anomaly signal of the concealed targets;

[0011] S5. Target Recognition and Output: Input the optical-magnetic fusion data into the classification model for target recognition, and output the target type, target location contour, and magnetic characteristic parameters.

[0012] Furthermore, step S1 includes the following sub-steps:

[0013] S11. Survey Line Planning: Based on the mission area, flight safety requirements, and resolution requirements, set the survey line height and survey line interval parameters, and plan the flight survey lines of the motion platform;

[0014] S12, Data Acquisition: Control the motion platform to fly along the planned flight survey line, and the optical imaging device and the magnetic measurement load acquire data synchronously or in time-sharing manner to obtain optical image data and magnetic measurement data containing timestamps;

[0015] S13. Spatial Registration: A registration method based on feature point matching is used to spatially align the optical image data with the magnetic measurement data. Specifically, it includes: extracting feature points from the optical image, establishing a correspondence between magnetic measurement points and optical image pixels based on the position data in the magnetic measurement data, achieving spatial registration of the two types of data through affine transformation or perspective transformation, and generating a spatially aligned multimodal dataset.

[0016] Furthermore, step S2 includes the following sub-steps:

[0017] S21. Image preprocessing: Denoising the optical image data to eliminate random noise introduced during image acquisition;

[0018] S22. Image Enhancement: The contrast of the denoised optical image is enhanced by using an adaptive histogram equalization method. The image is divided into several sub-regions, and histogram equalization is performed on each sub-region to generate the enhanced optical image.

[0019] S23. Target detection: Using a deep learning target detection model or edge detection algorithm, target detection is performed on the enhanced optical image to identify surface targets such as buildings, vehicles, bridges, and roads.

[0020] S24. Target Marking: Number each detected surface target, record its type name, center position coordinates and outline bounding box, and generate optical target detection results.

[0021] Furthermore, step S3 includes the following sub-steps:

[0022] S31. Background magnetic field elimination: Based on the sampling rate of the magnetic measurement data and the flight speed of the motion platform, a detrending method is used to eliminate the influence of the Earth's background magnetic field and extract local magnetic anomaly signals.

[0023] S32. Noise suppression: The local magnetic anomaly signal is filtered using wavelet transform, Kalman filtering, or adaptive filtering methods to eliminate environmental magnetic noise and obtain magnetic anomaly data.

[0024] S33. Target localization: The magnetic anomaly data is processed using the analytical signal method, Euler inversion method, or matched filtering method to obtain the horizontal position coordinates and depth information of the magnetic anomaly target;

[0025] S34. Grid Interpolation: Based on the density of the magnetic survey data and the type of survey line, select Kriging interpolation, inverse distance weighted interpolation, or spline interpolation to convert irregularly distributed magnetic anomaly data into regular grid data and generate a magnetic anomaly distribution map.

[0026] Furthermore, step S4 involves preliminary classification of the targets based on the matching results, including the following sub-steps:

[0027] S41. Image overlay: The magnetic anomaly distribution map and the enhanced optical image are overlaid and displayed according to the same spatial coordinate system to generate an optical-magnetic overlay image;

[0028] S42. Spatial matching: Traverse each surface target in the optical target detection results and detect whether there is a magnetic anomaly signal at its location. At the same time, traverse each magnetic anomaly region in the magnetic anomaly distribution map and detect whether there is an optically visible target at its location.

[0029] S43. Preliminary target classification: Based on the spatial matching results, targets are classified into three categories: camouflaged targets, i.e., targets visible in optical images but without magnetic anomaly response; visible ferromagnetic targets, i.e., targets visible in both optical images and magnetic anomaly distribution maps; and concealed targets, i.e. targets invisible in optical images but with magnetic anomaly response.

[0030] Furthermore, in step S4, magnetic field modeling is performed on visible ferromagnetic targets with large magnetic anomalies, and the distribution of magnetic anomalies generated by them within the mission area is calculated. This includes the following sub-steps:

[0031] S44. Interference target screening: Select targets with magnetic anomaly intensity greater than a preset threshold from the visible ferromagnetic targets as magnetic interference targets;

[0032] S45. Magnetic anomaly data extraction: Extract magnetic anomaly data within the contour region of each magnetic interference target;

[0033] S46. Magnetic field modeling: A magnetic field model is established for each of the magnetic interference targets using the multi-magnetic dipole equivalent method. The position, magnetic moment magnitude and direction parameters of each magnetic dipole are determined by least squares fitting.

[0034] S47. Magnetic Anomaly Calculation: Based on the established magnetic field model, calculate the magnetic anomaly value generated by each magnetic interference target at each grid point in the entire mission area, and sum up the magnetic anomaly values ​​generated by all magnetic interference targets to obtain the total magnetic anomaly distribution of the magnetic interference targets.

[0035] S48. Magnetic Anomaly Stripping: Subtract the total magnetic anomaly distribution of the magnetic interference target from the total magnetic anomaly data of the mission area to obtain the remaining magnetic anomaly data, thereby enhancing the visualization effect of magnetic anomaly signals of weak magnetic targets and concealed targets.

[0036] Furthermore, in step S46, a magnetic field model is established for the magnetic interference target using the multi-magnetic dipole equivalent method. The specific implementation process is as follows:

[0037] S461. Initialize magnetic dipole parameters: Determine the number of equivalent magnetic dipoles based on the outline size of the magnetic interference target. Distribute evenly within the target contour area The initial position of each magnetic dipole;

[0038] S462. Constructing a forward model of the magnetic field: Calculate the theoretical value of the magnetic anomaly generated by each magnetic dipole at the measurement point according to the magnetic dipole magnetic field formula. The total field strength of the magnetic anomaly generated by the magnetic dipole at any point in space is calculated by the following formula:

[0039] ;

[0040] in, The magnetic anomaly (nT) generated by the magnetic dipole at the observation point; Let be the vacuum permeability, with a value of . (H / m); is the magnetic moment of the magnetic dipole (A·m²). The distance (m) from the observation point to the magnetic dipole. The angle (rad) between the direction of the observation point and the direction of the magnetic moment. Pi;

[0041] S463. Parameter Inversion: Taking the minimum sum of squared residuals between measured magnetic anomaly data and theoretical calculations as the objective function, the position coordinates and magnetic moment parameters of each magnetic dipole are iteratively optimized using the least squares method until the residuals converge to the preset accuracy, thus completing the establishment of the magnetic field model.

[0042] Furthermore, step S5 includes the following sub-steps:

[0043] S51. Feature Extraction: Extract target features from the optical-magnetic fusion data. The target features include optical features and magnetic features. The optical features include target shape, texture and color information. The magnetic features include magnetic anomaly amplitude, magnetic anomaly range and magnetic anomaly gradient information.

[0044] S52. Classification and Recognition: The extracted target features are input into a pre-trained classification model, which is a convolutional neural network or a support vector machine, and the target type label and confidence score are output.

[0045] S53. Result Output: For targets with confidence scores higher than a preset threshold, output their target type, location contour information, and magnetic characteristic parameters. The target type includes underground fortifications, underground pipelines, camouflaged vehicles, and surface buildings. The magnetic characteristic parameters include equivalent magnetic moment and burial depth information.

[0046] Furthermore, in step S22, an adaptive histogram equalization method is used to enhance the image contrast. The specific implementation process is as follows:

[0047] S221. Image segmentation: The denoised optical image is segmented into... There are three equal-sized subregions, among which This represents the number of blocks in the horizontal direction. This represents the number of blocks in the vertical direction;

[0048] S222, Local Histogram Statistics: Calculate the grayscale histogram for each sub-region and the number of pixels at each grayscale level;

[0049] S223. Gray-scale mapping transformation: Transform the gray-scale values ​​of pixels within each sub-region using the following formula:

[0050] ;

[0051] in, This is the grayscale value of the output pixel; This represents the total number of gray levels in the image; for an 8-bit image, the value is 256. Indicates to From 0 to Summation; For the sub-region, the gray level is The number of pixels (in units); This represents the total number of pixels within the sub-region. The gray level of the input pixel, with a value ranging from 0 to... ;

[0052] S224. Bilinear Interpolation: The bilinear interpolation method is used to smooth the pixels at the boundaries of adjacent sub-regions, eliminating the discontinuity of the block boundaries and generating an enhanced optical image.

[0053] Furthermore, in step S33, the analytical signal method is used to locate the target in the magnetic anomaly data. The specific implementation process is as follows:

[0054] S331. Calculate the magnetic anomaly gradient: Calculate the magnetic anomaly gradient in the magnetic anomaly data respectively. direction, direction and The partial derivatives in the direction are used to obtain the three-component gradient data of the magnetic anomaly;

[0055] S332. Calculate the analytical signal amplitude: Calculate the analytical signal amplitude at each point based on the three-component gradient data. The calculation formula is as follows:

[0056] ;

[0057] in, For point The amplitude of the analytical signal at the location (nT / m); This represents the square root operation; Magnetic anomaly exist Partial derivative in direction (nT / m); Magnetic anomaly exist Partial derivative in direction (nT / m); Magnetic anomaly exist Partial derivative in direction (nT / m); The total magnetic anomaly field strength (nT); , , These are the east, north, and vertical coordinates (m), respectively.

[0058] S333, Extreme point extraction: Search for local maxima in the amplitude map of the analytical signal. The horizontal position of the maxima is the horizontal position of the magnetic anomaly target.

[0059] S334. Depth estimation: Estimate the burial depth of the magnetic anomaly target based on the half-amplitude width of the analytical signal amplitude, and complete the three-dimensional positioning of the magnetic anomaly target.

[0060] The beneficial effects achieved by this invention are as follows:

[0061] This invention presents a detection and identification method based on optical-magnetic fusion. Through the coordinated operation of optical imaging and magnetic anomaly detection, it achieves multi-dimensional perception of targets in complex environments. Optical data processing acquires visual information of surface targets, while magnetic anomaly data processing acquires magnetic response information of ferromagnetic targets. Optical-magnetic fusion processing comprehensively analyzes these two types of information, using optical detection results to assist in the interpretation of magnetic anomaly data. By modeling the magnetic field and stripping magnetic anomalies from known ferromagnetic targets, the detection effect on weakly magnetic and concealed targets is enhanced. This method fully leverages the high-resolution advantage of optical imaging and the cross-media advantage of magnetic anomaly detection, overcoming the limitations of single-technology applications in complex environments. It can be applied to target detection and identification tasks in fields such as military reconnaissance, geological exploration, and security monitoring. Attached Figure Description

[0062] Figure 1 This is a flowchart of a detection and identification method based on optical-magnetic fusion.

[0063] Figure 2 This is a schematic diagram of multimodal data acquisition according to the present invention. Detailed Implementation

[0064] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] This invention provides a detection and identification method based on optical-magnetic fusion. By integrating optical imaging technology with magnetic anomaly detection technology, it fully utilizes the complementarity of the two techniques to achieve high-sensitivity, high-resolution detection and identification of targets in complex environments. Optical imaging technology has the advantage of high-resolution visualization and is suitable for identifying surface targets, but it is easily affected by lighting conditions, weather, and obstructions, and cannot detect concealed targets such as underground or underwater targets. Magnetic anomaly detection technology has the advantages of all-weather and cross-media detection, and can penetrate non-magnetic media such as soil, forests, and water to detect ferromagnetic targets, but it has problems such as limited detection distance and susceptibility to background magnetic interference. This invention integrates the two techniques, using optical detection results to assist in magnetic anomaly data processing. By modeling the magnetic field of optically visible ferromagnetic targets and removing their contribution from the total magnetic anomaly data, the detection effect on weak magnetic targets and concealed targets is enhanced.

[0066] The implementation process includes multimodal data acquisition, using a mobile platform such as a drone equipped with optical imaging equipment such as visible light cameras, infrared thermal imagers, and magnetic measurement payloads such as high-precision magnetometers to collect data along planned survey lines within the mission area. Survey line planning must comprehensively consider the mission area range, flight safety requirements, and resolution needs, setting survey line height and interval parameters. Optical image data and magnetic measurement data, containing position and altitude information, are synchronized using a time synchronization unit to ensure temporal consistency, and spatial alignment is performed based on registration methods such as affine transformation or perspective transformation using feature point matching to generate a multimodal dataset. Optical data processing involves preprocessing the optical image data, including denoising such as Gaussian filtering or median filtering and image enhancement. An adaptive histogram equalization method is used to segment the image into sub-regions, perform histogram equalization on each sub-region, and eliminate boundary discontinuities through bilinear interpolation to generate enhanced optical images. Deep learning object detection models such as YOLO, Faster R-CNN, or edge detection algorithms are used to identify surface targets such as buildings, vehicles, and bridges, and to label the target type, center location, and outline bounding box, generating optical object detection results. Magnetic anomaly data processing involves background magnetic field elimination and noise suppression of magnetic measurement data to extract magnetic anomaly data. Target localization is performed using analytical signal methods, Euler inversion methods, or matched filtering methods to obtain the horizontal position and depth information of magnetic anomaly targets. Irregular magnetic anomaly data is converted into regular grid data through gridded interpolation methods such as Kriging interpolation or inverse distance weighted interpolation to generate a magnetic anomaly distribution map. Optical-magnetic data fusion processing involves overlaying the magnetic anomaly distribution map with the enhanced optical image to generate an optical-magnetic superimposed image. Spatial matching is performed by traversing the optical target detection results and the magnetic anomaly distribution map. Based on the matching results, targets are initially classified into: camouflaged targets that are optically visible but have no magnetic anomaly, visible ferromagnetic targets that have both optical and magnetic anomalies, and concealed targets that are optically invisible but have magnetic anomalies. For visible ferromagnetic targets with strong magnetic anomaly intensity, a magnetic field model is established using the multi-magnetic dipole equivalent method. Magnetic dipole parameters are determined through least-squares inversion, and their magnetic anomaly distribution within the mission area is calculated. This distribution is then separated from the total magnetic anomaly data to obtain residual magnetic anomaly data, enhancing the signals of weak magnetic targets and concealed targets. Target recognition and output involves extracting optical features such as shape, texture, and color from optical-magnetic fusion data, and magnetic features such as magnetic anomaly amplitude, range, and gradient. These features are then input into a pre-trained classification model, such as a convolutional neural network or support vector machine, for target recognition. The output includes target type (e.g., underground fortifications, underground pipelines, camouflaged vehicles, surface buildings), location contour, and magnetic property parameters (e.g., equivalent magnetic moment, burial depth). Results with confidence scores above a threshold are visualized or automatically processed for decision output.

[0067] Reference Figure 1The specific implementation steps of the optical-magnetic fusion-based detection and identification method of this invention include S1 to S5. Step S1 is the data acquisition step. In this step, an optical imaging device and a magnetic measurement payload are mounted on a motion platform to collect data according to the planned survey line within the mission area, acquiring optical image data and magnetic measurement data. The magnetic measurement data includes measurement location data and measurement altitude data. Then, a spatial registration algorithm is used to spatially align the optical image data and the magnetic measurement data to generate a spatially aligned multimodal dataset. The motion platform can be in the form of a UAV, manned aircraft, or ground mobile platform, etc. The optical imaging device mounted on it includes a visible light camera, an infrared thermal imager, or a multispectral camera, etc., used to acquire optical images of the mission area. The magnetic measurement payload includes a high-precision magnetometer and its matching positioning and attitude determination system, used to measure the total intensity or vector components of the geomagnetic field, and simultaneously record the geographical coordinates and flight altitude of the measurement points. The optical imaging device and the magnetic measurement payload work together through a time synchronization unit to ensure the temporal consistency of the two types of data. The synchronous acquisition of multimodal data lays the foundation for subsequent data fusion processing, enabling optical information and magnetic field information to have a spatial and temporal correspondence.

[0068] Step S1 further includes sub-steps S11, S12, and S13. Sub-step S11 is the survey line planning step. Based on the mission area, flight safety requirements, and resolution requirements, the survey line height and spacing parameters are set to plan the flight survey lines for the motion platform. The selection of the survey line height needs to comprehensively consider both flight safety and detection resolution. The lower the survey line height, the higher the resolution of the acquired optical images and the stronger the response of the magnetic measurement data to shallow targets, but the flight safety risk increases accordingly. The setting of the survey line spacing depends on the target detection coverage requirements and data density needs. The smaller the survey line spacing, the more complete the data coverage, but the acquisition time and data volume increase accordingly. In practical applications, appropriate survey line height and spacing parameters can be selected according to the terrain characteristics and target characteristics of the mission area. (Refer to...) Figure 2 The multimodal data acquisition method involves setting parameters such as the distance between survey lines on a moving platform, and using optical or magnetic measurement payloads to collect data along the planned survey lines. In the diagram, AF represents potential concealed, surface, or underground magnetic anomaly targets within the mission area. The types of scenarios and targets include, but are not limited to, those that may exist on the surface or underground. Figure 2 The diagram shows the following: A represents a possible concealed target (vehicle under a tree); B represents a possible camouflaged target; C and E represent possible underground targets (underground fortifications, pipelines, etc.); and D and F represent possible exposed ground targets (vehicles, capacitor-equipped buildings, etc.).

[0069] Sub-step S12 is the data acquisition step. The motion platform is controlled to fly along the planned flight path. The optical imaging equipment and the magnetic measurement payload acquire data synchronously or in a time-sharing manner, obtaining optical image data and magnetic measurement data containing timestamps. In synchronous acquisition mode, the optical imaging equipment and the magnetic measurement payload work simultaneously, and both types of data have the same acquisition time. In time-sharing acquisition mode, optical data can be acquired first to obtain overall image information of the mission area, and then a magnetic measurement route can be planned based on key areas in the optical image to perform detailed magnetic field measurements. During data acquisition, the motion platform's navigation system records the platform's position and attitude information in real time, providing a reference for subsequent data spatial registration.

[0070] Sub-step S13 is the spatial registration step, which uses a feature point matching registration method to spatially align the optical image data with the magnetic measurement data. This step first extracts feature points from the optical image, which can be obtained using feature extraction algorithms such as SIFT, SURF, or ORB. Then, it establishes a correspondence between the magnetic measurement points and the pixels in the optical image based on the positional data in the magnetic measurement data. Spatial registration of the two types of data is achieved through affine transformation or perspective transformation. Affine transformation can handle geometric deformations such as translation, rotation, and scaling, while perspective transformation can further handle geometric distortions caused by changes in viewing angle. After spatial registration, a spatially aligned multimodal dataset is generated, establishing a one-to-one correspondence between each pixel in the optical image and the corresponding magnetic measurement data, providing a unified spatial reference framework for subsequent optical-magnetic fusion processing.

[0071] Step S2 is the optical data processing step. Image enhancement processing is performed on the optical image data to generate an enhanced optical image. Then, a target detection algorithm is used to extract surface targets from the enhanced optical image, and the type and location information of each surface target are labeled to generate optical target detection results. The purpose of optical data processing is to extract surface target information within the task area from the original optical image, including the target's type, location, and contour features. Image enhancement processing can improve the discriminability of target features, providing high-quality input images for subsequent target detection. The target detection algorithm is used to automatically identify and locate targets in the image, reducing the workload of manual interpretation and improving the efficiency and accuracy of target detection.

[0072] Step S2 further includes sub-steps S21, S22, S23, and S24. Sub-step S21 is an image preprocessing step, which denoises the optical image data to eliminate random noise introduced during image acquisition. Image noise mainly originates from thermal noise and electronic noise of the imaging sensor, as well as interference during signal transmission. Denoising can be achieved using methods such as Gaussian filtering, median filtering, or bilateral filtering. Gaussian filtering is suitable for removing Gaussian white noise, median filtering has a good suppression effect on salt-and-pepper noise, and bilateral filtering can preserve image edge features while removing noise. The denoised image has a better signal-to-noise ratio, which is beneficial for subsequent image enhancement and target detection.

[0073] Sub-step S22 is the image enhancement step, which uses an adaptive histogram equalization method to enhance the contrast of the denoised optical image. The image is divided into several sub-regions, and histogram equalization is performed on each sub-region separately to generate the enhanced optical image. Adaptive histogram equalization, also known as contrast-limited adaptive histogram equalization, avoids the problem of over-enhancement or noise amplification in local image areas compared to global histogram equalization. This method first divides the image into several equally sized sub-regions, then independently calculates the gray-level histogram of each sub-region and performs equalization transformation. Finally, interpolation is used to eliminate discontinuities at the boundaries of adjacent sub-regions. The adaptive histogram equalization method has good enhancement effects on low-contrast images, visible light images in low-light environments, or infrared images.

[0074] Sub-step S22 further includes sub-steps S221, S222, S223, and S224. Sub-step S221 is an image segmentation step, which divides the denoised optical image into... There are three equal-sized subregions, among which This represents the number of blocks in the horizontal direction. This represents the number of blocks in the vertical direction. Choosing the number of blocks requires a trade-off between local enhancement and computational efficiency; more blocks result in stronger local adaptability, but also increase computational cost. In practical applications, appropriate block parameters can be selected based on the image size and contrast distribution characteristics.

[0075] Sub-step S222 is a local histogram statistics step, which calculates the gray-level histogram for each sub-region and the number of pixels at each gray level. The gray-level histogram reflects the distribution of pixel gray values ​​within the sub-region and is the basic data for histogram equalization transformation.

[0076] Sub-step S223 is the grayscale mapping transformation step, which transforms the pixel grayscale values ​​in each sub-region using the following transformation formula:

[0077] ;

[0078] in, This is the grayscale value of the output pixel; This represents the total number of gray levels in the image; for an 8-bit image, the value is 256. Indicates to From 0 to Summation; For the sub-region, the gray level is The number of pixels; This represents the total number of pixels within the sub-region. The gray level of the input pixel, with a value ranging from 0 to... This transformation formula is based on the principle of cumulative distribution function, which maps the gray-level distribution of the original image to an approximately uniform distribution, thereby achieving the effect of contrast enhancement.

[0079] Sub-step S224 is a bilinear interpolation step, which uses bilinear interpolation to smooth pixels at the boundaries of adjacent sub-regions, eliminating discontinuities at block boundaries and generating an enhanced optical image. Because each sub-region undergoes independent histogram equalization, differences in transformation mapping between adjacent sub-regions may occur, leading to noticeable block effects at the boundaries. Bilinear interpolation uses a weighted average of the transformation results from the four sub-regions surrounding the pixel, with the weights inversely proportional to the distance from the pixel to the center of each sub-region, resulting in smoother and more natural boundary transitions.

[0080] Sub-step S23 is the object detection step. It utilizes a deep learning object detection model or edge detection algorithm to detect objects in the enhanced optical image, identifying surface targets such as buildings, vehicles, bridges, and roads. Deep learning object detection models can employ classic models such as the YOLO series, Faster R-CNN, and SSD. These models, trained on large-scale datasets, possess strong object recognition and localization capabilities. The YOLO model employs an end-to-end single-stage detection framework, transforming object detection into a regression problem, offering the advantage of high detection speed. Faster R-CNN uses a region proposal network to generate candidate boxes, then classifies and refines their positions, achieving high detection accuracy. Edge detection algorithms, such as the Canny operator and the Sobel operator, can be used to detect object contours in images, suitable for scenarios involving geometrically regular objects.

[0081] Sub-step S24 is the target labeling step, which assigns a number to each detected surface target, records its type name, center location coordinates, and outline bounding box, and generates optical target detection results. The target number uses a unique identifier for easy subsequent data management and correlation analysis. The type name describes the target's category attribute, such as building, vehicle, bridge, etc. The center location coordinates record the target's geographical location in a unified spatial coordinate system. The outline bounding box uses a rectangle to represent the target's spatial extent, including the coordinates of the top-left corner and parameters such as width and height. The optical target detection results are stored in structured data format, providing input for subsequent optical-magnetic fusion processing.

[0082] Step S3 is the magnetic anomaly data processing step. Background magnetic field elimination and noise suppression are performed on the magnetic survey data to extract the magnetic anomaly data. Then, a target localization algorithm is used to obtain the location and contour information of the magnetic anomaly target. Finally, grid interpolation is used to convert the magnetic anomaly data into regular grid data, generating a magnetic anomaly distribution map. The magnetic survey data contains the Earth's background magnetic field, the magnetic anomaly signal generated by the target, and various environmental magnetic noises. The purpose of background magnetic field elimination and noise suppression is to separate the magnetic anomaly signal generated by the target from the magnetic survey data. The amplitude and spatial distribution characteristics of the magnetic anomaly signal are related to factors such as the target's magnetic properties, geometric dimensions, burial depth, and attitude. Through the analysis and processing of the magnetic anomaly data, the target's location information and magnetic characteristic parameters can be obtained.

[0083] Step S3 further includes sub-steps S31, S32, S33, and S34. Sub-step S31 is the background magnetic field elimination step. Based on the sampling rate of the magnetic measurement data and the flight speed of the motion platform, a detrending method is used to eliminate the influence of the Earth's background magnetic field and extract local magnetic anomaly signals. The Earth's background magnetic field is mainly generated by the current in the Earth's core and can be approximated as a slowly changing low-frequency component within the mission area. The detrending method removes the low-frequency trend component in the magnetic measurement data through polynomial fitting or high-pass filtering, retaining the high-frequency magnetic anomaly signal. The polynomial fitting method performs low-order polynomial fitting on the survey data, and the fitting result is used as the background magnetic field estimate and subtracted from the original data. The high-pass filtering method designs an appropriate cutoff frequency to filter out low-frequency background components. The effectiveness of background magnetic field elimination directly affects the accuracy of subsequent magnetic anomaly analysis.

[0084] Sub-step S32 is the noise suppression step, which uses wavelet transform, Kalman filtering, or adaptive filtering to filter the local magnetic anomaly signal, eliminate environmental magnetic noise, and acquire magnetic anomaly data. Environmental magnetic noise mainly includes magnetic interference from the moving platform itself, external electromagnetic interference, and sensor measurement noise. Wavelet transform can decompose the signal into multiple scales, separating signal and noise components at different scales, and has good time-frequency localization characteristics. Kalman filtering, based on a state-space model, performs optimal estimation of the magnetic anomaly signal and is suitable for dynamic measurement environments. Adaptive filtering can automatically adjust filtering parameters according to noise characteristics, and has a good suppression effect on non-stationary noise. In practical applications, a suitable filtering method can be selected based on the spectral characteristics and statistical features of the noise.

[0085] Sub-step S33 is the target localization step. It processes the magnetic anomaly data using the analytical signal method, Euler inversion method, or matched filtering method to obtain the horizontal coordinates and depth information of the magnetic anomaly target. The analytical signal method constructs an analytical signal using the magnetic anomaly and its gradient data. The maximum amplitude point of the analytical signal corresponds to the horizontal position of the target, and the half-width of the amplitude is related to the target depth. The Euler inversion method is based on the Euler homogeneous equation, obtaining the location and burial depth information of the magnetic anomaly source by solving the equation, without needing to pre-assume the magnetization direction. The matched filtering method designs matched filters for specific types of targets, detects the target through correlation analysis, and estimates its position. Different target localization methods have their own applicable conditions and accuracy characteristics; the appropriate method can be selected based on the quality of the magnetic anomaly data and the characteristics of the target.

[0086] Sub-step S33, which further employs the analytical signal method for target localization, includes sub-steps S331, S332, S333, and S334. Sub-step S331 is the step of calculating the magnetic anomaly gradient, which calculates the magnetic anomaly data at... direction, direction and The partial derivatives in the directional direction yield the three-component gradient data of the magnetic anomaly. The magnetic anomaly gradient reflects the spatial rate of change of the magnetic field and can be calculated using the finite difference method or the frequency domain differentiation method. The finite difference method approximates the partial derivatives by using the difference between magnetic anomaly values ​​at adjacent measurement points, and is suitable for situations with small data sampling intervals. The frequency domain differentiation method transforms the magnetic anomaly data to the frequency domain, calculates the gradient using the differential properties of the frequency domain, and then transforms it back to the spatial domain, effectively suppressing the influence of high-frequency noise. The calculation of the vertical gradient requires the use of potential field extension techniques, extending the magnetic anomaly data upwards or downwards before performing the difference calculation.

[0087] Sub-step S332 is the step of calculating the analytical signal amplitude. It calculates the analytical signal amplitude at each point based on the three-component gradient data. The calculation formula is as follows:

[0088] ;

[0089] in, For point The amplitude of the analyzed signal at that location; This represents the square root operation; Magnetic anomaly exist Partial derivatives in direction; Magnetic anomaly exist Partial derivatives in direction; Magnetic anomaly exist Partial derivatives in direction; This represents the total field strength of the magnetic anomaly. , , These represent the east, north, and vertical coordinates, respectively. The amplitude of the analytical signal is independent of the magnetization direction, and the location of its maximum point corresponds to the horizontal position of the magnetic anomaly source, eliminating the influence of geomagnetic tilt and deflection on target positioning.

[0090] Sub-step S333 is the extreme point extraction step, which searches for local maxima in the analytical signal amplitude map. The horizontal position of the maxima is the horizontal position of the magnetic anomaly target. The search for local maxima can use a sliding window method, comparing the amplitude of the center point with that of surrounding points within the window. If the amplitude of the center point is greater than that of all adjacent points, it is marked as a local maximum. An amplitude threshold can be set to filter out low-amplitude noise peaks, retaining the physically meaningful target response.

[0091] Sub-step S334 is the depth estimation step, which estimates the burial depth of the magnetic anomaly target based on the half-amplitude width of the analytical signal, thus completing the three-dimensional localization of the magnetic anomaly target. For approximate point sources or compact targets, the analytical signal amplitude curve exhibits a bell-shaped distribution, and its half-amplitude width has an approximately linear relationship with the target burial depth. By measuring the horizontal distance corresponding to the drop in analytical signal amplitude to half of its peak value on the profile passing through the extreme point, the burial depth of the target can be estimated. This method is simple and practical, providing depth reference information for subsequent target classification and magnetic field modeling.

[0092] Sub-step S34 is the gridded interpolation step. Based on the density of the magnetic survey data and the survey line type, Kriging interpolation, inverse distance weighted interpolation, or spline interpolation is selected to convert the irregularly distributed magnetic anomaly data into regular grid data, generating a magnetic anomaly distribution map. Actual magnetic survey data is usually distributed along the survey line, with sparse data in the direction perpendicular to the survey line, making it unsuitable for direct image display and spatial analysis. Gridded interpolation interpolates scattered data onto regular grid nodes, making the data spatially uniform. Kriging interpolation, based on geostatistical principles, uses a variogram to describe the spatial correlation of the data, possessing the characteristic of optimal unbiased estimation. Inverse distance weighted interpolation assumes that the value of the interpolation point is inversely proportional to the weighted average of the distances to surrounding known points, making calculation simple and efficient. Spline interpolation uses piecewise polynomial functions for interpolation, obtaining smooth surfaces, suitable for smooth reconstruction of magnetic anomaly fields. The gridded magnetic anomaly data, displayed as an image, is the magnetic anomaly distribution map, which can intuitively reflect the spatial distribution characteristics of the magnetic anomaly.

[0093] Step S4 is the optical-magnetic data fusion processing step. The magnetic anomaly distribution map is superimposed onto the enhanced optical image. Then, the optical target detection results are spatially matched with the positions of the magnetic anomaly targets. Based on the matching results, the targets are initially classified. Next, magnetic field modeling is performed on visible ferromagnetic targets with high magnetic anomaly intensity, calculating the magnetic anomaly distribution they generate within the task area. Then, the magnetic anomaly distribution generated by visible ferromagnetic targets is subtracted from the total magnetic anomaly data in the task area to obtain the remaining magnetic anomaly data, thus enhancing the magnetic anomaly signal of concealed targets. Optical-magnetic fusion processing is the core step of this invention. By comprehensively analyzing optical detection information and magnetic anomaly detection information, it enables the differentiation and identification of different types of targets. Simultaneously, by utilizing the position and contour information of ferromagnetic targets detected optically, their magnetic anomaly contributions are modeled and stripped, reducing the interference of known targets on the magnetic anomaly data, thereby enhancing the detection sensitivity for deep and concealed targets.

[0094] The process of initially classifying the target based on the matching results in step S4 includes sub-steps S41, S42 and S43.

[0095] Sub-step S41 is the image overlay step, which overlays the magnetic anomaly distribution map and the enhanced optical image using the same spatial coordinate system to generate an optical-magnetic overlay image. Image overlay requires that the two types of data have the same spatial resolution and coordinate reference system, which can be achieved by resampling to adjust the two types of images to the same pixel size. The overlay display method can employ transparency overlay, pseudo-color overlay, or split-screen comparison display, allowing operators to intuitively observe the spatial correspondence between optical features and magnetic anomaly responses.

[0096] Sub-step S42 is the spatial matching step. It iterates through each surface target in the optical target detection results, detecting whether a magnetic anomaly signal exists at its location. Simultaneously, it iterates through each magnetic anomaly region in the magnetic anomaly distribution map, detecting whether an optically visible target exists at its location. Spatial matching performs correlation analysis based on the target's location information, setting a matching distance threshold to determine whether the optical target and the magnetic anomaly response correspond to the same physical entity. If a magnetic anomaly extreme value exists within the optical target's outline, the match is considered successful. For isolated magnetic anomaly regions outside the optical target's outline, no corresponding optical magnetic anomaly target is identified.

[0097] Sub-step S43 is the preliminary target classification step. Based on the spatial matching results, targets are divided into three categories. Category 1 is camouflaged targets, which are visible in the optical image but show no magnetic anomaly response. These targets may be made of non-ferromagnetic materials or have their magnetic properties concealed through camouflage. Category 2 is visible ferromagnetic targets, which are visible in both the optical image and the magnetic anomaly distribution map. These targets have both optically visible features and produce a significant magnetic anomaly response, typically ferromagnetic facilities or vehicles exposed on the ground. Category 3 is concealed targets, which are not visible in the optical image but exhibit magnetic anomaly response. These targets may be located underground, underwater, or covered by obstructions, making them impossible to observe directly through optical means, but their magnetic anomaly signals can be detected by magnetic measurement payloads. This preliminary target classification provides a framework for subsequent refined identification.

[0098] Step S4 involves modeling the magnetic field of a visible ferromagnetic target with a large magnetic anomaly intensity and calculating the distribution of the magnetic anomaly generated within the mission area. This process includes sub-steps S44, S45, S46, S47, and S48.

[0099] Sub-step S44 is the interference target screening step, which selects targets with magnetic anomaly strength greater than a preset threshold from visible ferromagnetic targets as magnetic interference targets. Magnetic interference targets are usually large or strongly magnetic ferromagnetic facilities, which generate high-amplitude magnetic anomaly signals with a wide spatial coverage, potentially masking or interfering with the response of deep, weakly magnetic targets. The preset threshold can be set according to the background level of magnetic anomalies in the mission area and the target detection requirements, selecting targets that have a significant impact on the overall magnetic anomaly distribution for focused processing.

[0100] Sub-step S45 is the magnetic anomaly data extraction step, which extracts magnetic anomaly data within the contour region of each magnetically disturbed target. Based on the target contour bounding box obtained by optical detection, the data of the corresponding region is extracted from the magnetic anomaly distribution map as the magnetic anomaly measurement value of the target. The extracted magnetic anomaly data will be used for subsequent magnetic field modeling. The data range should include the magnetic anomaly response generated by the target body, while avoiding the introduction of interference signals from adjacent targets.

[0101] Sub-step S46 is the magnetic field modeling step. A magnetic field model is established for each magnetic interference target using the multi-magnetic dipole equivalent method. The position, magnetic moment magnitude, and direction parameters of each magnetic dipole are determined through least-squares fitting. When the detection distance is greater than 2 to 3 times the target size, the magnetic anomaly generated by the target can be equivalently represented as a magnetic dipole field. For larger targets, a combination of multiple magnetic dipoles can be used to equivalently represent their magnetic field distribution. The multi-magnetic dipole equivalent method simplifies the complex target magnetic field distribution to a superposition of several discrete magnetic dipoles, facilitating analytical calculations and parameter inversion.

[0102] Sub-step S46 further includes sub-steps S461, S462, and S463. Sub-step S461 is the magnetic dipole parameter initialization step, which determines the number of equivalent magnetic dipoles based on the contour size of the magnetic interference target. Distribute evenly within the target contour area The initial positions of the magnetic dipoles. The selection of the number of magnetic dipoles needs to comprehensively consider the target's geometry and modeling accuracy requirements. Too few dipoles may not accurately represent the target's magnetic field distribution, while too many dipoles increase the complexity and instability of the inversion calculation. The initial positions provide a starting point for subsequent parameter inversion; proper initialization can accelerate the inversion convergence speed.

[0103] Sub-step S462 is the step of constructing the forward model of the magnetic field. It calculates the theoretical value of the magnetic anomaly generated by each magnetic dipole at the measurement point based on the magnetic dipole magnetic field formula. The total field strength of the magnetic anomaly generated by the magnetic dipole at any point in space is calculated using the following formula:

[0104] ;

[0105] in, This represents the magnetic anomaly value generated by the magnetic dipole at the observation point; Let be the vacuum permeability, with a value of . H / m; The magnetic moment of a magnetic dipole; The distance from the observation point to the magnetic dipole; The angle between the direction of the observation point and the direction of the magnetic moment; π is the mathematical constant pi. This formula describes the spatial attenuation of the magnetic field of a magnetic dipole with distance. The intensity of the magnetic anomaly signal is inversely proportional to the cube of the distance, directly proportional to the magnitude of the magnetic moment, and related to the observation direction. For multiple magnetic dipoles, the magnetic anomaly values ​​generated by each dipole at the measurement point are linearly superimposed to obtain the total theoretical magnetic anomaly value.

[0106] Sub-step S463 is the parameter inversion step. With the objective function of minimizing the sum of squared residuals between the measured magnetic anomaly data and the theoretically calculated values, the least squares method is used to iteratively optimize the position coordinates and magnetic moment parameters of each magnetic dipole until the residuals converge to a preset accuracy, thus completing the establishment of the magnetic field model. Least squares inversion minimizes the fitting error between the theoretically calculated values ​​and the measured data by adjusting the model parameters. Numerical optimization methods such as gradient descent, Newton's method, or the Levenberg-Marquardt algorithm can be used during the iterative optimization process. The preset accuracy serves as the termination condition for the iteration; when the residual change is less than this accuracy threshold, the inversion is considered converged, and the optimal model parameters are output. The established magnetic field model contains the position coordinates, magnetic moment magnitude, and direction information of each magnetic dipole, and can be used to calculate the magnetic anomaly value generated by the target at any location.

[0107] Sub-step S47 is the magnetic anomaly calculation step. Based on the established magnetic field model, the magnetic anomaly value generated by each magnetic interference target at each grid point within the entire mission area is calculated. The magnetic anomaly values ​​generated by all magnetic interference targets are summed to obtain the total magnetic anomaly distribution of the magnetic interference targets. Using the magnetic dipole magnetic field formula in sub-step S462, the coordinates of each grid point within the mission area are substituted into the calculation to obtain the magnetic anomaly value generated by each magnetic interference target at that point. Since the magnetic field satisfies the superposition principle, the magnetic anomalies generated by multiple targets can be directly algebraically summed. The calculation results are stored in the form of grid data, reflecting the magnetic field distribution generated by known magnetic interference targets throughout the entire mission area.

[0108] Sub-step S48 is the magnetic anomaly stripping step, which subtracts the total magnetic anomaly distribution of magnetically interfering targets from the total magnetic anomaly data of the mission area to obtain residual magnetic anomaly data, enhancing the visualization of magnetic anomaly signals of weakly magnetic targets and concealed targets. Magnetic anomaly stripping is based on the principle of field source separation, highlighting the response of unknown sources by removing the contributions of known sources. In the stripped residual magnetic anomaly data, the magnetic anomaly contribution of optically visible ferromagnetic targets is effectively suppressed, while the responses of previously obscured deep targets and weakly magnetic targets become prominent. Compared to the original magnetic anomaly distribution map, the residual magnetic anomaly distribution map has higher detection sensitivity for concealed targets, which is beneficial for discovering underground fortifications, buried pipelines, and other concealed facilities.

[0109] Step S5 is the target recognition and output step. The data after optical-magnetic fusion is input into the classification model for target recognition, and the output includes the target type, target location contour, and magnetic characteristic parameters. Target recognition is the final step in the detection process, comprehensively utilizing optical and magnetic features to classify targets and estimate their attributes. The classification model learns the feature distribution of known samples to determine the category and assess the confidence level of newly detected targets. The output results are presented in structured information form, providing data support for subsequent decision-making applications.

[0110] Step S5 further includes sub-steps S51, S52, and S53. Sub-step S51 is the feature extraction step, which extracts target features from the optical-magnetic fusion data. Target features include optical features and magnetic features. Optical features include target shape, texture, and color information. Shape features describe the geometric contour of the target and can be quantified using methods such as moment features and Fourier descriptors. Texture features reflect the roughness and regularity of the target surface and can be extracted using methods such as gray-level co-occurrence matrix and local binary mode. Color features describe the spectral response of the target and can be represented using methods such as color histogram and color moments. Magnetic features include magnetic anomaly amplitude, magnetic anomaly range, and magnetic anomaly gradient information. Magnetic anomaly amplitude reflects the overall magnetic strength of the target, magnetic anomaly range reflects the spatial scale of the target's influence, and magnetic anomaly gradient reflects the spatial variation characteristics of the magnetic field. Optical and magnetic features constitute a multi-dimensional feature vector, which serves as the input to the classification model.

[0111] Sub-step S52 is the classification and recognition step. The extracted target features are input into a pre-trained classification model, which can be a convolutional neural network (CNN) or a support vector machine (SVM). The model outputs the target's type label and confidence score. CNNs automatically learn hierarchical representations of features through multiple convolutional and pooling operations, possessing powerful nonlinear modeling capabilities and suitable for classifying complex patterns. SVMs achieve class separation by constructing an optimal classification hyperplane, exhibiting good generalization performance under small sample conditions. During the training phase, the classification model uses labeled sample data to learn the mapping relationship between features and categories. In the application phase, it predicts the category of newly detected targets and provides a confidence assessment. The confidence score reflects the reliability of the classification result, providing a reference for decision-making.

[0112] Sub-step S53 is the result output step. For targets with confidence scores higher than a preset threshold, it outputs their target type, location contour information, and magnetic characteristic parameters. Target types include typical categories such as underground fortifications, underground pipelines, camouflaged vehicles, and surface buildings, determined by category labels from a classification model. Location contour information includes the target's center coordinates, boundary range, and depth estimate, obtained by combining optical detection and magnetic anomaly localization results. Magnetic characteristic parameters include equivalent magnetic moment and burial depth information, determined by magnetic field modeling and inversion processes. The output results can be displayed visually in a geographic information system with target location and attributes labeled, or output in a structured data format for use by automated decision-making systems. The confidence threshold is used to filter low-confidence identification results to avoid false alarms interfering with normal decision-making.

[0113] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A detection and identification method based on optical-magnetic fusion, characterized in that, Includes the following steps: S1. Data Acquisition: Using a motion platform equipped with optical imaging equipment and a magnetic measurement payload, data is collected within the mission area according to the planned survey line to acquire optical image data and magnetic measurement data, wherein the magnetic measurement data includes measurement position data and measurement height data; then, the optical image data and the magnetic measurement data are spatially aligned using a spatial registration algorithm to generate a spatially aligned multimodal dataset; S2. Optical data processing: Perform image enhancement processing on the optical image data to generate an enhanced optical image; Then, the target detection algorithm is used to extract the surface targets in the enhanced optical image, and the type and location information of each surface target are marked to generate optical target detection results; S3. Magnetic Anomaly Data Processing: Background magnetic field elimination and noise suppression are performed on the magnetic measurement data to extract magnetic anomaly data; then, the location and contour information of the magnetic anomaly target are obtained using a target localization algorithm; and finally, the magnetic anomaly data is converted into regular grid data through grid interpolation to generate a magnetic anomaly distribution map. S4. Optical and magnetic data fusion processing: The magnetic anomaly distribution map is superimposed on the enhanced optical image; then, the optical target detection results are spatially matched with the positions of the magnetic anomaly targets, and the targets are initially classified according to the matching results; then, magnetic field modeling is performed on the visible ferromagnetic targets with large magnetic anomaly intensity, and the magnetic anomaly distribution generated by them in the task area is calculated; then, the magnetic anomaly distribution generated by the visible ferromagnetic targets is subtracted from the total magnetic anomaly data in the task area to obtain the remaining magnetic anomaly data, thereby enhancing the magnetic anomaly signal of the concealed targets; S5. Target Recognition and Output: Input the optical-magnetic fusion data into the classification model for target recognition, and output the target type, target location contour, and magnetic characteristic parameters.

2. The detection and identification method based on optical-magnetic fusion according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11. Survey Line Planning: Based on the mission area, flight safety requirements, and resolution requirements, set the survey line height and survey line interval parameters, and plan the flight survey lines of the motion platform; S12, Data Acquisition: Control the motion platform to fly along the planned flight survey line, and the optical imaging device and the magnetic measurement load acquire data synchronously or in time-sharing manner to obtain optical image data and magnetic measurement data containing timestamps; S13. Spatial Registration: A registration method based on feature point matching is used to spatially align the optical image data with the magnetic measurement data. Specifically, it includes: extracting feature points from the optical image, establishing a correspondence between magnetic measurement points and optical image pixels based on the position data in the magnetic measurement data, achieving spatial registration of the two types of data through affine transformation or perspective transformation, and generating a spatially aligned multimodal dataset.

3. The detection and identification method based on optical-magnetic fusion according to claim 1, characterized in that, Step S2 includes the following sub-steps: S21. Image preprocessing: Denoising the optical image data to eliminate random noise introduced during image acquisition; S22. Image Enhancement: The contrast of the denoised optical image is enhanced by using an adaptive histogram equalization method. The image is divided into several sub-regions, and histogram equalization is performed on each sub-region to generate the enhanced optical image. S23. Target detection: Using a deep learning target detection model or edge detection algorithm, target detection is performed on the enhanced optical image to identify surface targets such as buildings, vehicles, bridges, and roads. S24. Target Marking: Number each detected surface target, record its type name, center position coordinates and outline bounding box, and generate optical target detection results.

4. The detection and identification method based on optical-magnetic fusion according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Background magnetic field elimination: Based on the sampling rate of the magnetic measurement data and the flight speed of the motion platform, a detrending method is used to eliminate the influence of the Earth's background magnetic field and extract local magnetic anomaly signals. S32. Noise suppression: The local magnetic anomaly signal is filtered using wavelet transform, Kalman filtering, or adaptive filtering methods to eliminate environmental magnetic noise and obtain magnetic anomaly data. S33. Target localization: The magnetic anomaly data is processed using the analytical signal method, Euler inversion method, or matched filtering method to obtain the horizontal position coordinates and depth information of the magnetic anomaly target; S34. Grid Interpolation: Based on the density of the magnetic survey data and the type of survey line, select Kriging interpolation, inverse distance weighted interpolation, or spline interpolation to convert irregularly distributed magnetic anomaly data into regular grid data and generate a magnetic anomaly distribution map.

5. The detection and identification method based on optical-magnetic fusion according to claim 1, characterized in that, Step S4 involves preliminary classification of the targets based on the matching results, including the following sub-steps: S41. Image overlay: The magnetic anomaly distribution map and the enhanced optical image are overlaid and displayed according to the same spatial coordinate system to generate an optical-magnetic overlay image; S42. Spatial matching: Traverse each surface target in the optical target detection results and detect whether there is a magnetic anomaly signal at its location. At the same time, traverse each magnetic anomaly region in the magnetic anomaly distribution map and detect whether there is an optically visible target at its location. S43. Preliminary target classification: Based on the spatial matching results, targets are classified into three categories: camouflaged targets, i.e., targets visible in optical images but without magnetic anomaly response; visible ferromagnetic targets, i.e., targets visible in both optical images and magnetic anomaly distribution maps; and concealed targets, i.e. targets invisible in optical images but with magnetic anomaly response.

6. The detection and identification method based on optical-magnetic fusion according to claim 5, characterized in that, Step S4 involves modeling the magnetic field of visible ferromagnetic targets with large magnetic anomalies and calculating the distribution of magnetic anomalies generated by them within the mission area. This includes the following sub-steps: S44. Interference target screening: Select targets with magnetic anomaly intensity greater than a preset threshold from the visible ferromagnetic targets as magnetic interference targets; S45. Magnetic anomaly data extraction: Extract magnetic anomaly data within the contour region of each magnetic interference target; S46. Magnetic field modeling: A magnetic field model is established for each of the magnetic interference targets using the multi-magnetic dipole equivalent method. The position, magnetic moment magnitude and direction parameters of each magnetic dipole are determined by least squares fitting. S47. Magnetic Anomaly Calculation: Based on the established magnetic field model, calculate the magnetic anomaly value generated by each magnetic interference target at each grid point in the entire mission area, and sum up the magnetic anomaly values ​​generated by all magnetic interference targets to obtain the total magnetic anomaly distribution of the magnetic interference targets. S48. Magnetic Anomaly Stripping: Subtract the total magnetic anomaly distribution of the magnetic interference target from the total magnetic anomaly data of the mission area to obtain the remaining magnetic anomaly data, thereby enhancing the visualization effect of magnetic anomaly signals of weak magnetic targets and concealed targets.

7. The detection and identification method based on optical-magnetic fusion according to claim 6, characterized in that, In step S46, a magnetic field model is established for the magnetic interference target using the multi-magnetic dipole equivalent method. The specific implementation process is as follows: S461. Initialize magnetic dipole parameters: Determine the number of equivalent magnetic dipoles based on the outline size of the magnetic interference target. Distribute evenly within the target contour area The initial position of each magnetic dipole; S462. Constructing a forward model of the magnetic field: Calculate the theoretical value of the magnetic anomaly generated by each magnetic dipole at the measurement point according to the magnetic dipole magnetic field formula. The total field strength of the magnetic anomaly generated by the magnetic dipole at any point in space is calculated by the following formula: ； in, The magnetic anomaly (nT) generated by the magnetic dipole at the observation point; Let be the vacuum permeability, with a value of . (H / m); is the magnetic moment of the magnetic dipole (A·m²). The distance (m) from the observation point to the magnetic dipole. The angle (rad) between the direction of the observation point and the direction of the magnetic moment. Pi; S463. Parameter Inversion: Taking the minimum sum of squared residuals between measured magnetic anomaly data and theoretical calculations as the objective function, the position coordinates and magnetic moment parameters of each magnetic dipole are iteratively optimized using the least squares method until the residuals converge to the preset accuracy, thus completing the establishment of the magnetic field model.

8. The detection and identification method based on optical-magnetic fusion according to claim 1, characterized in that, Step S5 includes the following sub-steps: S51. Feature Extraction: Extract target features from the optical-magnetic fusion data. The target features include optical features and magnetic features. The optical features include target shape, texture and color information. The magnetic features include magnetic anomaly amplitude, magnetic anomaly range and magnetic anomaly gradient information. S52. Classification and Recognition: The extracted target features are input into a pre-trained classification model, which is a convolutional neural network or a support vector machine, and the target type label and confidence score are output. S53. Result Output: For targets with confidence scores higher than a preset threshold, output their target type, location contour information, and magnetic characteristic parameters. The target type includes underground fortifications, underground pipelines, camouflaged vehicles, and surface buildings. The magnetic characteristic parameters include equivalent magnetic moment and burial depth information.

9. The detection and identification method based on optical-magnetic fusion according to claim 3, characterized in that, In step S22, an adaptive histogram equalization method is used to enhance the image contrast. The specific implementation process is as follows: S221. Image segmentation: The denoised optical image is segmented into... There are three equal-sized subregions, among which This represents the number of blocks in the horizontal direction. This represents the number of blocks in the vertical direction; S222, Local Histogram Statistics: Calculate the grayscale histogram for each sub-region and the number of pixels at each grayscale level; S223. Gray-scale mapping transformation: Transform the gray-scale values ​​of pixels within each sub-region using the following formula: ； in, This is the grayscale value of the output pixel; This represents the total number of gray levels in the image; for an 8-bit image, the value is 256. Indicates to From 0 to Summation; For the sub-region, the gray level is The number of pixels (in units); This represents the total number of pixels within the sub-region. The gray level of the input pixel, with a value ranging from 0 to... ; S224. Bilinear Interpolation: The bilinear interpolation method is used to smooth the pixels at the boundaries of adjacent sub-regions, eliminating the discontinuity of the block boundaries and generating an enhanced optical image.

10. The detection and identification method based on optical-magnetic fusion according to claim 4, characterized in that, In step S33, the analytical signal method is used to locate the target in the magnetic anomaly data. The specific implementation process is as follows: S331. Calculate the magnetic anomaly gradient: Calculate the magnetic anomaly gradient in the magnetic anomaly data respectively. direction, direction and The partial derivatives in the direction are used to obtain the three-component gradient data of the magnetic anomaly; S332. Calculate the analytical signal amplitude: Calculate the analytical signal amplitude at each point based on the three-component gradient data. The calculation formula is as follows: ； in, For point The amplitude of the analytical signal at the location (nT / m); This represents the square root operation; Magnetic anomaly exist Partial derivative in direction (nT / m); Magnetic anomaly exist Partial derivative in direction (nT / m); Magnetic anomaly exist Partial derivative in direction (nT / m); The total magnetic anomaly field strength (nT); , , These are the east, north, and vertical coordinates (m), respectively. S333, Extreme point extraction: Search for local maxima in the amplitude map of the analytical signal. The horizontal position of the maxima is the horizontal position of the magnetic anomaly target. S334. Depth estimation: Estimate the burial depth of the magnetic anomaly target based on the half-amplitude width of the analytical signal amplitude, and complete the three-dimensional positioning of the magnetic anomaly target.

Images