3D Modeling Method and System for Slopes in Difficult and Dangerous Mountainous Areas Based on Multi-Source Data Fusion

Abstract

The present invention relates to the field of three-dimensional modeling technology, and in particular to a method and system for three-dimensional modeling of slopes in difficult mountainous areas based on multi-source data fusion. The method comprises the following steps: using a cluster of unmanned aerial vehicles to perform multi-angle acquisition of the mountain slope to obtain multi-angle image data of the slope; performing ground laser scanning on the mountain slope to obtain a slope point cloud dataset; performing seismic reflection wave detection on the mountain slope to obtain slope seismic reflection data; performing electromagnetic tomography detection on the mountain slope to obtain slope electromagnetic tomography data; fusing and inverting the slope seismic reflection data and the slope electromagnetic tomography data to obtain slope internal structure data; and performing coarse registration of the slope multi-angle image data with the slope point cloud dataset to obtain initial image-point cloud registration data. The present invention can achieve fine reconstruction of a complete three-dimensional slope model.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling, and in particular to a three-dimensional modeling method and system for slopes in difficult mountainous areas based on multi-source data fusion. Background Art

[0002] Traditional slope modeling methods primarily rely on ground surveying and geological exploration. While these methods offer advantages in flat or easily accessible areas, they face numerous challenges in rugged mountainous areas. First, the complex terrain, characterized by steep cliffs and ravines, makes it difficult for traditional ground surveying equipment to access these areas for data collection. For example, in some high-altitude mountainous areas, the steep terrain and harsh climate significantly restrict access for surveying personnel and equipment. This not only increases the difficulty and cost of surveying, but in some cases, even makes it impossible to obtain complete topographic data. Furthermore, the complexity of geological conditions poses challenges to modeling. The diverse geological structures in rugged mountainous areas, with fragmented rock formations and well-developed faults, make it difficult for traditional geological exploration methods to accurately capture these complex geological features. For example, in areas prone to geological disasters, the subsurface geological structure is complex and varied, making it difficult for traditional exploration methods to fully reveal these structures, thus compromising the accuracy and reliability of slope modeling. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a three-dimensional modeling method and system for slopes in difficult mountainous areas based on multi-source data fusion to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a 3D modeling method for slopes in difficult mountainous areas based on multi-source data fusion is proposed, which includes the following steps:

[0005] Step S1: Use a drone cluster to collect multi-angle data of the mountain slope to obtain multi-angle image data of the slope; perform ground laser scanning on the mountain slope to obtain a slope point cloud dataset;

[0006] Step S2: Perform seismic reflection wave detection on the mountain slope to obtain slope seismic reflection data; perform electromagnetic tomography detection on the mountain slope to obtain slope electromagnetic tomography data; fuse and invert the slope seismic reflection data and the slope electromagnetic tomography data to obtain slope internal structure data;

[0007] Step S3: performing coarse registration of the slope multi-angle image data and the slope point cloud data set to obtain initial image-point cloud registration data; performing tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data; performing spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data;

[0008] Step S4: performing improved watershed segmentation on the slope image-point cloud unified data to obtain slope unit division data; performing structural surface recognition on the slope unit division data to obtain slope structural surface feature data; performing feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data;

[0009] Step S5: Layer the slope hierarchical feature data to obtain slope layer data; reconstruct the surface of the slope layer data to obtain a slope spatial surface model; and reconstruct the mountain slope in three dimensions based on the slope spatial surface model to obtain a complete slope three-dimensional model.

[0010] The present invention can quickly obtain multi-angle image data of slopes in difficult mountainous areas through multi-angle acquisition by drone clusters, without the need for manual entry into dangerous areas, reducing the safety risks of personnel and equipment, while expanding the scope of data acquisition and ensuring the comprehensiveness and integrity of the data. The three-dimensional point cloud data set of the slope can be accurately obtained through ground laser scanning, which makes up for the deficiency of traditional ground measurement in obtaining complete data in complex terrain. Seismic reflection wave detection can effectively identify geological structures such as stratum interfaces and faults, while electromagnetic tomography detection can reveal the conductivity distribution inside the slope. After the fusion and inversion of the two data, the internal structural data of the slope can be obtained more accurately, overcoming the problem that traditional geological exploration methods are difficult to accurately capture geological features under complex geological conditions. After coarse alignment, the tensor field interpolation fine alignment method is used to further improve the alignment accuracy of the image and point cloud data, ensuring the spatial consistency of the data. The improved watershed segmentation method can more accurately divide the slope units, avoiding the problem of over-segmentation or under-segmentation that is prone to occur in traditional watershed algorithms in complex terrain. Through structural surface identification and feature extraction, structural surface feature data of the slope can be effectively extracted and integrated with internal structural data to obtain hierarchical slope feature data. By layering and reconstructing the surface layer of the slope's hierarchical feature data, a spatial surface model of the slope can be obtained. This model, combined with the reconstruction of the intermediate and deep layers, ultimately achieves a detailed reconstruction of a complete three-dimensional slope model. This model not only accurately reflects the slope's external morphology and internal structure, but also reveals the slope's hierarchical characteristics and geological details, offering greater accuracy and reliability than traditional modeling methods.

[0011] Preferably, the present invention further provides a 3D modeling system for slopes in difficult mountainous areas based on multi-source data fusion, which is used to execute the 3D modeling method for slopes in difficult mountainous areas based on multi-source data fusion as described above. The 3D modeling system for slopes in difficult mountainous areas based on multi-source data fusion comprises:

[0012] The data acquisition module is used to use a cluster of drones to collect multi-angle data on mountain slopes to obtain multi-angle image data of the slopes; and to perform ground laser scanning on mountain slopes to obtain slope point cloud data sets;

[0013] The structural detection module is used to perform seismic reflection wave detection on mountain slopes to obtain slope seismic reflection data; perform electromagnetic tomography detection on mountain slopes to obtain slope electromagnetic tomography data; and fuse and invert slope seismic reflection data and slope electromagnetic tomography data to obtain slope internal structure data.

[0014] The data registration and unification module is used to perform coarse registration of the slope multi-angle image data and the slope point cloud data set to obtain the initial image-point cloud registration data; perform tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data; and perform spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data.

[0015] The segmentation module is used to perform improved watershed segmentation on the unified slope image-point cloud data to obtain slope unit division data; perform structural surface recognition on the slope unit division data to obtain slope structural surface feature data; perform feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data;

[0016] The three-dimensional reconstruction module is used to stratify the hierarchical characteristic data of the slope to obtain the slope layer data; to reconstruct the surface of the slope layer data to obtain the slope spatial surface model; and to reconstruct the mountain slope in three dimensions based on the slope spatial surface model to obtain a complete slope three-dimensional model.

[0017] In the present invention, the data acquisition module, through the collaborative work of a drone cluster and ground-based laser scanning equipment, can automatically and efficiently acquire multi-angle image data and point cloud datasets for slopes in difficult mountainous areas, reducing the labor intensity and time cost of manual collection while improving the accuracy and consistency of data acquisition. The structural detection module, combined with seismic reflection wave detection and electromagnetic tomography detection technology, can comprehensively and accurately detect the geological structure within the slope. The fused and inverted data can more realistically reflect the geological characteristics within the slope. The data alignment and unification module, using a method that combines coarse alignment and fine alignment using tensor field interpolation, can efficiently and accurately achieve spatial alignment of image and point cloud data, and perform spatial unified transformation with internal structural data, ensuring the quality and consistency of data fusion. The segmentation module, utilizing an improved watershed segmentation algorithm and structural surface recognition, can intelligently divide slope units and extract structural surface feature data, improving the accuracy and efficiency of feature extraction. The three-dimensional reconstruction module, through layered processing and surface reconstruction of hierarchical feature data, can automatically achieve detailed reconstruction of the three-dimensional slope model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects and advantages of the present invention will become more apparent from reading the detailed description made with reference to the following drawings:

[0019] Figure 1 A schematic flow chart of the steps of a three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion according to one embodiment is shown.

[0020] Figure 2 A detailed flowchart of step S26 of an embodiment is shown.

[0021] Figure 3 A detailed flowchart of step S5 of an embodiment is shown. DETAILED DESCRIPTION

[0022] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0023] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0024] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0025] To achieve this, please refer to Figures 1 to 3 The present invention provides a three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion, comprising the following steps:

[0026] Step S1: Use a drone cluster to collect multi-angle data of the mountain slope to obtain multi-angle image data of the slope; perform ground laser scanning on the mountain slope to obtain a slope point cloud dataset;

[0027] Step S2: Perform seismic reflection wave detection on the mountain slope to obtain slope seismic reflection data; perform electromagnetic tomography detection on the mountain slope to obtain slope electromagnetic tomography data; fuse and invert the slope seismic reflection data and the slope electromagnetic tomography data to obtain slope internal structure data;

[0028] Step S3: performing coarse registration of the slope multi-angle image data and the slope point cloud data set to obtain initial image-point cloud registration data; performing tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data; performing spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data;

[0029] Step S4: performing improved watershed segmentation on the slope image-point cloud unified data to obtain slope unit division data; performing structural surface recognition on the slope unit division data to obtain slope structural surface feature data; performing feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data;

[0030] Step S5: Layer the slope hierarchical feature data to obtain slope layer data; reconstruct the surface of the slope layer data to obtain a slope spatial surface model; and reconstruct the mountain slope in three dimensions based on the slope spatial surface model to obtain a complete slope three-dimensional model.

[0031] In this embodiment, a swarm of drones equipped with high-resolution RGB cameras is used to capture multi-angle images of mountain slopes, generating multi-angle image data. Next, a terrestrial laser scanner is used to scan the slopes, generating a slope point cloud dataset. Simultaneously, seismic reflection wave detection equipment is used to detect the slopes, acquiring seismic reflection data. Electromagnetic tomography detection equipment is also used to detect the slopes, generating electromagnetic tomography data. Data processing software, such as GeoStudio, fuses and inverts the seismic reflection data and electromagnetic tomography data to obtain slope internal structure data. OpenCV software is used to coarsely align the multi-angle image data with the slope point cloud dataset, generating initial image-point cloud registration data. The initial image-point cloud registration data is then finely aligned using the tensor field interpolation method in MATLAB, generating image-point cloud registration data. Next, Leica GeoOffice software is used to spatially transform the image-point cloud registration data and the slope internal structure data to generate unified slope image-point cloud data. The improved watershed algorithm in MATLAB was used to segment the unified slope image-point cloud data to obtain slope unit division data. CloudCompare software was then used to identify the structural surface of the slope unit division data, obtaining slope structural surface feature data. The feature extraction algorithm in MATLAB was used to extract features from the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data. MeshLab software was used to perform layered processing on the slope hierarchical feature data to obtain slope layered data. The slope layered data was then surface reconstructed to obtain a slope spatial surface model. Finally, based on the slope spatial surface model, SolidWorks software was used to perform a three-dimensional reconstruction of the mountain slope to obtain a complete three-dimensional slope model.

[0032] Preferably, step S1 includes the following steps:

[0033] Step S11: collecting a side panoramic image of the mountain slope to obtain a side panoramic image of the slope, and performing regional mission planning for the UAV cluster based on the side panoramic image of the slope to obtain a UAV flight mission;

[0034] Specifically, a drone equipped with a high-resolution 360-degree panoramic camera can be selected, capable of capturing the entire slope in one shot. Before flight, the drone's flight parameters are set based on the slope's topographical characteristics and the expected flight altitude, including a speed of 5 meters per second and an altitude of 100 meters. During flight, the drone ascends in a spiral pattern along the slope's side, capturing panoramic images every 5 meters. This method yields a panoramic image of the slope's lateral surface. Using image processing software, the multiple panoramic images are stitched and corrected to create a complete panoramic image of the slope's lateral surface. Based on this panoramic image, regional mission planning is performed for the drone swarm. For example, the slope can be divided into three roughly equal areas. Different drone numbers and flight missions are assigned to each area based on the terrain complexity and data collection requirements. For example, two drones can be assigned to perform high-density image acquisition in areas with steeper terrain, while one drone can be assigned to perform regular image acquisition in areas with relatively flat terrain.

[0035] Step S12: assigning routes to the drone cluster based on the drone flight mission to obtain drone route planning data;

[0036] Specifically, the flight paths of a drone swarm can be planned based on a panoramic view of the slope's side profile and the drones' flight mission. Using drone route planning software, the software automatically generates the optimal flight path based on terrain data and mission requirements. The software inputs the slope's topographic data and mission parameters for each area, such as altitude, speed, and image capture interval. Based on these parameters, the software generates a detailed route for each drone. For example, in areas with steep terrain, the route is designed to follow the contour lines. In areas with flatter terrain, the route is designed to fly in a straight line. Furthermore, considering the drones' battery life and flight safety, multiple temporary landing points and emergency avoidance areas can be included in the route. For example, a temporary landing point is set in the middle of the slope so that drones can land and replace batteries in case of low battery. Emergency avoidance areas are set at the edges of the slope to cope with sudden wind changes or other flight obstacles. Through these steps, the drone route planning data is obtained.

[0037] Step S13: performing an aerial laser ranging scan on the side of the mountain slope to obtain slope characteristic data;

[0038] Specifically, a drone equipped with a lidar system can be used for aerial laser ranging scanning. The drone is equipped with a lidar with a 100-meter measurement range and 0.02-meter accuracy, as well as a high-capacity battery. Before flight, the drone's flight parameters are set based on the slope's topographical characteristics, including a speed of 8 meters per second and an altitude of 150 meters. The drone flies parallel to the slope's side, performing laser scans every 10 meters horizontally to acquire elevation data. During the scan, the lidar emits laser pulses at a frequency of 1000 times per second, measuring the time between emission and return, thereby calculating the elevation of each point on the slope. After the scan is complete, the acquired laser ranging data is imported into professional geographic information system (GIS) software, which generates a 3D point cloud model of the slope and calculates the slope value at each point. For example, in steep areas of the slope, the slope can exceed 65 degrees, while in flat areas, the slope is less than 15 degrees. Through the above steps, the slope characteristic data can be obtained.

[0039] Step S14: highly stratify the UAV route planning data according to the slope gradient characteristic data to obtain UAV layered collection data;

[0040] Specifically, slope gradient data can be imported into route planning software, which then divides the slope into different altitude levels based on the gradient. Specifically, areas with slopes between 0 and 25 degrees are classified as the low altitude level, with a flight altitude set at 80 meters; areas with slopes between 25 and 50 degrees are classified as the medium altitude level, with a flight altitude set at 120 meters; and areas with slopes between 50 and 70 degrees are classified as the high altitude level, with a flight altitude set at 160 meters. Within each altitude level, the drone's route is further refined based on terrain characteristics and data collection requirements. For example, in the flat areas of the low altitude level, the route is designed to fly in a straight line, with the drone flying at a speed of 6 meters per second and taking photos every 10 meters. In areas with moderate slopes in the medium altitude level, the route is designed to follow contour lines, with the drone flying at a speed of 7 meters per second and taking photos every 12 meters. In steep areas of the high altitude level, the route is designed to follow the slope lines, with the drone flying at a speed of 5 meters per second and taking photos every 8 meters. Through the above steps, the layered collection data of the drone can be obtained.

[0041] Step S15: using a drone cluster to collect multi-angle data on the mountain slope based on the layered data collected by the drone to obtain multi-angle image data of the slope;

[0042] Specifically, drones can be used to collect data in layers, allowing drone swarms to capture data from multiple angles on the slope. Based on the layered data, the slope is divided into three height layers: low (0-50 meters), middle (50-100 meters), and high (100-150 meters). Different numbers and types of drones are deployed for each layer. In the low layer, three quadrotor drones equipped with high-resolution RGB cameras with a resolution of 20 megapixels were deployed. The drones were set to fly at a speed of 5 meters per second and an altitude of 40 meters. The drones followed a pre-set route, photographing the slope at angles of 45, 90, and 135 degrees, taking photos every 5 meters. In the middle layer, two fixed-wing drones equipped with infrared cameras with a resolution of 10 megapixels were deployed. The drones flew at a speed of 10 meters per second and an altitude of 80 meters. The drones photographed the slope at angles of 30, 60, and 90 degrees, taking photos every 10 meters. At a high level, a multirotor drone equipped with a 5-megapixel hyperspectral camera was deployed at a speed of 8 meters per second at an altitude of 120 meters. The drone photographed the slope at angles of 15, 45, and 75 degrees, taking a picture every 15 meters. This process yielded multi-angle image data of the slope.

[0043] Step S16: Perform laser radar site planning on the mountain slope according to the slope gradient characteristic data to obtain laser scanning site distribution data, and perform multi-site scanning on the mountain slope according to the laser scanning site distribution data to obtain a slope point cloud dataset.

[0044] Specifically, the slope can be divided into several scanning areas based on the slope gradient data, each covering an area of approximately 100 square meters. LiDAR stations are planned at the center and four corners of each area. For example, in steep areas of the slope, the distance between stations is set at 50 meters; in flatter areas, the distance between stations is set at 100 meters. Multiple high-performance ground-based LiDAR devices are used for multi-site scanning. These LiDAR devices have a measurement range of 100 meters, an accuracy of 0.01 meters, and a scanning frequency of 50,000 times per second. During the scanning process, the LiDAR devices perform a 360-degree sweep of the slope, acquiring elevation data and 3D coordinate information of the slope surface. Each station takes approximately 30 minutes to scan. After completion, the point cloud data from each station is imported into professional point cloud processing software, which generates a 3D point cloud model of the slope based on this data. The above steps result in a slope point cloud dataset.

[0045] The present invention can comprehensively acquire the terrain features and slope information of the slope by performing panoramic image acquisition and aerial laser ranging scanning on the side of the slope, so that the acquisition process can more accurately cover the key areas of the slope, avoiding data omissions or errors caused by complex terrain. Through reasonable route planning and flight mission allocation, the UAV can fly safely according to the actual terrain characteristics of the slope. Through a highly layered data acquisition method, detailed data at different height levels of the slope can be obtained. Through reasonable LiDAR site planning, the scanning process can more efficiently cover the key areas of the slope. This not only improves the efficiency of LiDAR scanning, but also ensures that the acquired point cloud data has higher accuracy and integrity.

[0046] Preferably, step S2 includes the following steps:

[0047] Step S21: Arrange seismic wave sources on the mountain slopes to obtain seismic source arrangement data;

[0048] Specifically, a controllable seismic source device can be selected as a seismic wave source, which can generate seismic waves with a frequency range of 10-100Hz. During the layout process, the seismic wave source equipment is evenly distributed at the bottom and middle areas of the slope according to the terrain characteristics of the slope and the detection requirements. Specifically, a seismic source point is laid out every 50 meters at the bottom of the slope, and a seismic source point is laid out every 70 meters in the middle of the slope. At each seismic source point, the excitation parameters of the seismic source device are set, including an excitation frequency of 30Hz, an excitation energy of 500J, and an excitation interval of 2 seconds. During the layout process, a GPS positioning device is used to locate each seismic source point, and the coordinates, elevation, excitation parameters and other information of the seismic source point are recorded to finally obtain the seismic source layout data.

[0049] Step S22: Arranging receiver arrays on mountain slopes according to the seismic source layout data to obtain slope receiver layout data;

[0050] Specifically, a highly sensitive seismic geophone with a frequency response range of 5-200Hz can be selected as a receiver. When arranging the receiver array, a grid-like arrangement is designed based on the location of the earthquake source and the topographical characteristics of the slope. Specifically, at the bottom and middle areas of the slope, a row of receivers is arranged every 10 meters, with each row of receivers spaced 15 meters apart, forming a uniformly distributed grid. For example, near the first earthquake source at the bottom of the slope, five receivers are arranged, located directly above, to the upper left, to the upper right, to the lower left, and to the lower right of the earthquake source. Seven receivers are arranged near the earthquake source in the middle of the slope. A small pit is dug around each receiver, and the receiver is placed vertically in the pit and fixed with soil and stones. At the same time, a GPS positioning device is used to locate each receiver, and the receiver's coordinates, elevation, and layout parameters are recorded to ultimately obtain the slope receiver layout data.

[0051] Step S23: configuring the trigger timing of the receiver array according to the slope receiver layout data to obtain receiver trigger sequence data;

[0052] Specifically, the receiver layout data can be imported into the seismic wave detection control system, which automatically generates a trigger timing scheme based on the location and number of receivers. During the configuration process, the trigger timing parameters are set, including the trigger delay time, trigger interval, and trigger sequence. For example, the trigger delay time can be set to 0.1 seconds to compensate for the difference in seismic wave propagation speed in different media; the trigger interval can be set to 0.5 seconds; and the trigger sequence can be set to trigger from the receiver at the bottom of the slope and upwards. After the configuration is completed, the trigger timing scheme is simulated and tested to check the trigger time and signal synchronization of each receiver. Finally, the receiver trigger sequence data is obtained.

[0053] Step S24: performing seismic wave excitation on the mountain slope according to the receiver trigger sequence data to obtain slope seismic wave data, and performing first arrival wave identification on the slope seismic wave data to obtain slope wave arrival time data;

[0054] Specifically, a pre-placed vibroseis device can be used to sequentially excite seismic waves at each source point according to the timing in the trigger sequence data. For example, at the first source point at the bottom of the slope, the source device, under the control of a trigger signal, excites at a frequency of 30 Hz and an energy of 500 J, generating a seismic wave signal. Next, at the second source point in the middle of the slope, the source device, under the control of a trigger signal, excites at the same frequency and energy, and so on. During the excitation process, a seismic wave detection and control system monitors the propagation of seismic waves and the signal reception of receivers in real time. After the seismic wave excitation is completed, the seismic wave data received by each receiver is collected and imported into seismic data processing software. Based on the propagation characteristics of the seismic waves and the trigger timing of the receivers, the software performs first-arrival identification on the seismic wave data, identifying the initial time of arrival of the seismic wave at each receiver. For example, in the receiver array at the bottom of the slope, the first-arrival identification results show that the seismic wave arrives at the first receiver at 1.2 seconds, the second receiver at 1.5 seconds, and so on. Through first-arrival identification, slope wave arrival time data can be obtained.

[0055] Step S25: performing travel time inversion on the slope seismic wave data according to the slope wave arrival time data to obtain seismic wave velocity structure data, and performing waveform inversion on the seismic wave velocity structure data to obtain slope seismic reflection data;

[0056] Specifically, slope arrival time data and a preset seismic wave propagation model can be imported into the software. The software utilizes a travel-time inversion algorithm based on the least squares method, which can invert the velocity structure of seismic waves in the slope medium based on the time differences between the arrival of seismic waves at the receiver. During the inversion process, initial model parameters are set, including an initial velocity of 2000 m / s and a velocity gradient of 0.5 s. These parameters are determined based on the slope's geological background and previous empirical data. The software iteratively adjusts the velocity model parameters to minimize the error between the calculated theoretical arrival time and the actual arrival time. After multiple iterations, the slope's seismic wave velocity structure data is ultimately obtained. This data details the distribution of seismic wave propagation velocity in different media along the slope. For example, in the rock layer of the slope, the seismic wave velocity can reach 3000 m / s, while in the soil layer, the velocity is approximately 1500 m / s. Once the velocity structure data is obtained, waveform inversion is performed on the seismic wave velocity structure data. The software utilizes a waveform inversion algorithm based on the wave equation, capable of inverting seismic reflection data from the slope based on the propagation velocity and waveform characteristics of seismic waves. During the inversion process, an initial reflection coefficient model is set, and the slope's reflection interface and reflection waveform are calculated based on seismic wave velocity structure data and actual seismic wave data. For example, at a fault interface on the slope, the inversion results reveal a distinct reflection peak, indicating a high reflection coefficient at that interface. Through waveform inversion, seismic reflection data can be obtained from the slope.

[0057] Step S26: Perform electromagnetic tomography detection on the mountain slope to obtain slope electromagnetic tomography data, and fuse and invert the slope seismic reflection data and the slope electromagnetic tomography data to obtain slope internal structure data.

[0058] Specifically, please refer to the sub-steps of step S26 for the detailed implementation process of this embodiment.

[0059] The present invention can ensure the accurate excitation and reception of seismic wave signals through the layout of seismic wave sources and the arrangement of receiver arrays. The reasonable trigger timing configuration makes the acquisition of seismic wave data more stable and reliable, and also helps to improve the accuracy of seismic wave velocity structure data, so that the obtained seismic reflection data more accurately reflects the geological structure inside the slope. Through electromagnetic tomography detection, it is possible to penetrate deep into the slope and detect the distribution of conductivity at different depth levels. Through the fusion and inversion of seismic reflection data and electromagnetic tomography data, the data information of two different physical properties is integrated, and the internal structure of the slope can be analyzed and interpreted from multiple angles. This overcomes the limitations of a single data source under certain geological conditions, improves the comprehensiveness and accuracy of the internal structure of the slope, and makes the obtained internal structure data of the slope more real and reliable.

[0060] Preferably, step S26 includes the following steps:

[0061] Step S261: Arranging electromagnetic transmitting sources on the mountain slope to obtain electromagnetic source arrangement data, and arranging receiving electrodes on the mountain slope based on the electromagnetic source arrangement data to obtain electrode arrangement data;

[0062] Specifically, a GEM-2 electromagnetic induction instrument can be selected, which can generate electromagnetic waves with a frequency range of 1-10MHz. During the deployment process, the electromagnetic emission source equipment is evenly distributed at the bottom and middle areas of the slope based on the terrain characteristics of the slope and the detection requirements. Specifically, a transmission source point is deployed every 30 meters at the bottom of the slope, and a transmission source point is deployed every 50 meters in the middle of the slope. At each transmission source point, the transmission parameters of the transmission source equipment are set, including a transmission frequency of 5MHz, a transmission power of 500W, and a transmission duration of 10 seconds. During the deployment process, a GPS positioning device is used to accurately locate each transmission source point, and information such as the coordinates, elevation, and transmission parameters of the transmission source point are recorded to ultimately obtain the electromagnetic source deployment data. Based on the electromagnetic source deployment data, receiving electrodes are deployed on the mountain slope. High-sensitivity electrodes are selected as receivers, and the resistivity measurement range of the electrodes is 0.1-1000Ω·m. When arranging the receiving electrodes, a grid-like layout is adopted, with a row of electrodes placed every 10 meters, and each row spaced 15 meters apart, forming a uniformly distributed grid. For example, near the first transmitting source at the bottom of the slope, six receiving electrodes are placed directly above, to the upper left, to the upper right, to the lower left, to the lower right, and directly below the transmitting source. A small pit is dug around each receiving electrode, and the electrode is placed vertically in the pit, secured with soil and rocks. Simultaneously, GPS positioning equipment is used to precisely locate each receiving electrode, and information such as the electrode's coordinates, elevation, and layout parameters are recorded to ultimately generate electrode layout data.

[0063] Step S262: configuring the excitation waveform of the electromagnetic emission source according to the electrode layout data to obtain electromagnetic excitation waveform data;

[0064] Specifically, the electrode layout data can be imported into the Geonics EM-38 electromagnetic induction measurement system, which can automatically generate an excitation waveform scheme based on the position and number of electrodes. During the configuration process, the parameters of the excitation waveform are set, including the waveform type, frequency range, amplitude, and phase. For example, a sine wave is selected as the excitation waveform type, the frequency range is set to 2-8MHz, the amplitude is set to 2V, and the phase is set to 0 degrees. In addition, the excitation waveform can be optimized and adjusted according to the geological characteristics of the slope and the detection target. For example, in the rock layer area of the slope, the excitation frequency is appropriately increased; in the soil layer area of the slope, the excitation frequency is appropriately reduced. After the configuration is completed, the electromagnetic excitation waveform data is finally obtained.

[0065] Step S263: performing electromagnetic field measurement on the receiving electrode according to the electromagnetic excitation waveform data to obtain slope electromagnetic field data;

[0066] Specifically, the electromagnetic field of the mountain slope can be measured using receiving electrodes based on the electromagnetic excitation waveform data. First, ensure that the electromagnetic emission source equipment transmits stably according to the preset waveform parameters, with a transmission frequency of 2-8MHz and an amplitude of 2V. During the measurement process, each receiving electrode is connected to a high-precision electromagnetic field measuring instrument that can collect the electromagnetic signals received by the electrode in real time and convert them into electromagnetic field intensity data. Each receiving electrode is measured synchronously. For example, at the first receiving electrode at the bottom of the slope, the measurement results show that the electromagnetic field intensity at a frequency of 2MHz is 0.5V / m, and the electromagnetic field intensity at a frequency of 8MHz is 0.3V / m. The electromagnetic field intensity data of each electrode at different frequencies is recorded and combined with the electrode layout position and elevation information to obtain the slope electromagnetic field data.

[0067] Step S264: performing inversion and reconstruction on the slope electromagnetic field data to obtain slope conductivity distribution data;

[0068] Specifically, slope electromagnetic field data can be imported into the GeoElectrical module in GeoStudio software. Initial inversion model parameters are then set based on the slope's geological context and previous empirical data. For example, the initial conductivity is set to 0.1 S / m, and the conductivity gradient is set to 0.01 S / m. These parameters are determined based on the typical geological conditions of the slope area and previous inversion experience from similar projects. The GeoElectrical module employs an inversion algorithm based on the finite difference method, specifically a finite difference forward inversion algorithm. This algorithm constructs a three-dimensional mesh model of the slope, fits the electromagnetic field data to the model, and calculates the conductivity distribution of the slope medium. During the inversion process, the inversion results are iteratively optimized. Each iteration adjusts the conductivity model parameters based on the inversion error and model fit. For example, by comparing the error between the inversion results and the actual electromagnetic field data, the initial conductivity value and conductivity gradient are appropriately adjusted to reduce the error. After multiple iterations, the error between the slope conductivity distribution data and the actual electromagnetic field data is minimized. The inversion results show that the conductivity of the rock layer of the slope is low, approximately 0.05 S / m, while the conductivity of the water-bearing soil layer of the slope is high, approximately 0.5 S / m. Through the above steps, the slope conductivity distribution data is finally obtained.

[0069] Step S265: reconstructing the slope electromagnetic data according to the slope conductivity distribution data to obtain slope electromagnetic tomography data;

[0070] Specifically, slope conductivity distribution data can be imported into EIDORS Electrological Impedance Tomography and Diffuse Optical Tomography Reconstruction Software (EIDORS). Image reconstruction parameters are then set based on the slope's geometric characteristics and electromagnetic field measurement data. The image resolution is set to 1 meter by 1 meter, and the reconstruction range is set to cover the entire slope area. The reconstruction algorithm utilizes the algebraic reconstruction technique (ART) implemented in the EIDORS software. During the image reconstruction process, EIDORS uses the conductivity distribution data and electromagnetic field propagation model through iterative calculations to gradually optimize the electromagnetic field distribution within the slope and convert it into visual image data. For example, the rock layer of the slope appears in the image as a dark area with low conductivity, while the water-bearing soil layer appears as a light area with high conductivity. Through image reconstruction using the EIDORS software, slope electromagnetic tomography data can be obtained, which visually displays the conductivity distribution characteristics within the slope.

[0071] Step S266: fusing and inverting the slope seismic reflection data and the slope electromagnetic tomography data to obtain the slope internal structure data.

[0072] Specifically, slope seismic reflection data and slope electromagnetic tomography data can be imported into the GIM software. GIM software then uses a Bayesian inversion algorithm to jointly invert the seismic reflection and electromagnetic tomography data. During the inversion process, the algorithm considers the propagation characteristics of seismic and electromagnetic waves in the slope medium, such as the seismic wave reflection coefficient and the conductivity variation of the electromagnetic wave, as well as the correlation and complementarity between the two types of data. For example, at a fault interface on the slope, the seismic reflection data shows a distinct reflection peak, indicating a high reflection coefficient at this interface, while the electromagnetic tomography data reveals an anomalous change in conductivity, indicating a change in the material composition or water content at this interface. Through data fusion and inversion, GIM software integrates this information to determine the precise location, morphology, and properties of the fault interface. Ultimately, the resulting internal slope structural data provides detailed information about the slope's geological structure, rock strata distribution, and material composition. For example, the inversion results reveal the presence of multiple fault interfaces and rock strata interfaces within the slope, with the conductivity and reflection coefficient characteristics of each interface clearly visible.

[0073] The present invention ensures accurate excitation and reception of electromagnetic signals through the layout of electromagnetic emission sources and receiving electrodes. Reasonable electrode layout provides a stable foundation for electromagnetic field measurement. By inverting and reconstructing the slope electromagnetic field data, the slope's conductivity distribution data can be obtained. This can reveal the differences in conductivity between different materials within the slope, thereby more closely reflecting the geological structure and material distribution characteristics within the slope. Image reconstruction based on the conductivity distribution data yields slope electromagnetic tomography data with high clarity and readability. This can intuitively display the structural features within the slope, enabling technicians to more intuitively identify and analyze the slope's geological structure, improving the data's practicality and usability.

[0074] Preferably, step S3 includes the following steps:

[0075] Step S31: extracting feature points from the multi-angle image data of the slope to obtain a slope image feature point set, and extracting feature points from the slope point cloud data set to obtain a slope point cloud feature point set;

[0076] Specifically, for multi-angle image data, Adobe Photoshop was used as the image processing software, with the SIFT (Scale-Invariant Feature Transform) algorithm selected as the feature point extraction method. This algorithm can detect key points in the image and extract their feature descriptors. During the extraction process, the SIFT algorithm parameters were set, such as the number of scale space layers to 3, the sampling interval of each scale space to 1.2 times, and the feature point threshold to 0.01. Through these parameter settings, the software detected a large number of feature points in each image and generated a 128-dimensional feature vector for each feature point. For example, in an image of a slope, 500 feature points were extracted, and these feature points were primarily concentrated at the slope's edges, corners, and texture-rich areas. For the slope point cloud dataset, CloudCompare, a point cloud processing software capable of feature point extraction, was used. PCA (Principal Component Analysis) was selected as the feature point extraction method. This algorithm can identify feature points in the point cloud based on the local geometric characteristics of the point cloud data. During the extraction process, the PCA algorithm parameters were set, such as a local neighborhood radius of 0.1 meters and a curvature threshold of 0.5 for feature points. With these parameters, the software identified a large number of feature points in the point cloud data and calculated the normal vector and curvature value for each feature point. For example, 800 feature points were extracted from the slope point cloud data. These feature points were primarily distributed in areas of uneven slope and near structural surfaces. Ultimately, the slope image feature point set and the slope point cloud feature point set were obtained.

[0077] Step S32: matching the slope image feature point set with the slope point cloud feature point set to obtain initial slope matching point pair data;

[0078] Specifically, OpenCV can be used for feature point matching. OpenCV uses the FLANN (Fast Approximate Nearest Neighbor) algorithm as a feature point matching method, which can quickly find the nearest neighbor matching point pairs in large-scale feature point data. During the matching process, the parameters of the FLANN algorithm are set, such as the number of trees is 5, the branching factor is 32, and the number of search iterations is 10. Through the setting of parameters, the software establishes a large number of matching point pairs between the image feature point set and the point cloud feature point set. For example, in a certain area of the slope, a feature point in the image feature point set and a feature point in the point cloud feature point set are matched as a pair. They are very close in spatial position, and the distance between the feature descriptors is small. Through the above steps, the initial matching point pair data of the slope is finally obtained.

[0079] Step S33: screening the initial matching point pair data of the slope to obtain the effective matching point pair data of the slope;

[0080] Specifically, the initial matching point pair data of the slope can be imported into the OpenCV software, and the RANSAC (Random Sample Consensus) algorithm in OpenCV can be used to screen the initial matching point pair data. The RANSAC algorithm fits the model by randomly selecting a minimum point set and evaluates all data points to identify and eliminate incorrect matching point pairs and retain stable matching point pairs. During the implementation process, the parameters of the RANSAC algorithm are set, such as the number of iterations is 1000 times and the inlier threshold is 3 pixels. The setting of these parameters is based on the resolution of the slope image data and the distribution of feature points. For example, in the slope image, the average spacing of feature points is 10 pixels, so the inlier threshold is set to 3 pixels. Through the screening of the RANSAC algorithm, the OpenCV software identifies and eliminates approximately 30% of incorrect matching point pairs from a large number of initial matching point pairs, and finally obtains valid slope matching point pair data.

[0081] Step S34: performing ICP iterative registration on the slope image feature point set and the slope point cloud feature point set according to the slope effective matching point pair data to obtain a feature point registration transformation matrix;

[0082] Specifically, the valid matching point pair data as well as the image feature point set and point cloud feature point set can be imported into the CloudCompare software. Next, the ICP registration algorithm is started, which iteratively searches for the nearest point pair and calculates the transformation matrix. During the ICP registration process, the parameters of the algorithm are set, such as the maximum number of iterations is 50 times and the registration accuracy is 0.01 meters. These parameters are set based on the scale and accuracy requirements of the slope data. For example, the length of the slope is 100 meters and the width is 50 meters, and the registration accuracy is set to 0.01 meters. In each iteration, the CloudCompare software updates the position of the feature points according to the current transformation matrix and recalculates the distance error of the nearest point pair. After multiple iterations, the feature point registration transformation matrix is finally obtained. This matrix contains translation, rotation and scaling transformation parameters, which can accurately transform the image feature point set to a spatial position that matches the point cloud feature point set.

[0083] Step S35: performing spatial transformation on the slope multi-angle image data and the slope point cloud data set according to the feature point registration transformation matrix to obtain initial image-point cloud registration data;

[0084] Specifically, the multi-angle image data and point cloud data set of the slope can be imported into the Leica Geo Office software, and the feature point registration transformation matrix can be imported. The spatial transformation tool in the software can be used to perform precise spatial transformation on the image data and point cloud data according to the feature point registration transformation matrix. During the spatial transformation process, the transformation parameters are set, such as the interpolation method is selected as bilinear interpolation. For example, the original resolution of the slope image is 0.5 meters / pixel. The bilinear interpolation method can maintain the detailed features in the transformed image and avoid pixel distortion or blurring. At the same time, for the point cloud data, the nearest neighbor interpolation method is selected. For example, in the steep area of the slope, the point cloud density is high. The nearest neighbor interpolation method can maintain the high-density features of the area in the transformed point cloud data. After the spatial transformation processing of the Leica Geo Office software, the multi-angle image data and the point cloud data set of the slope are accurately transformed to the same spatial coordinate system, and the initial image-point cloud registration data is obtained.

[0085] Step S36: Perform tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data, and perform spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data.

[0086] Specifically, please refer to the sub-steps of step S36 for the detailed implementation process of this embodiment.

[0087] The present invention can accurately capture the key feature points of the slope surface by extracting feature points from the multi-angle image data and point cloud data set of the slope. This can ensure the consistency of the image data and the point cloud data at the feature point level, provide a reliable basis for data matching and registration, and improve the accuracy of data fusion. By using the RANSAC algorithm to screen the initial matching point pair data, the erroneous matching point pairs can be effectively removed and the valid matching point pairs can be retained. This improves the robustness and reliability of data matching, making the obtained valid matching point pair data more accurate. Through the ICP iterative registration method, the feature point registration transformation matrix can be accurately calculated, thereby achieving high-precision registration of the image feature point set and the point cloud feature point set. Through tensor field interpolation precise registration, the spatial consistency and smoothness of the registration data can be further improved. By performing a spatial unified transformation on the image-point cloud registration data and the slope internal structure data, the slope image-point cloud unified data obtained can effectively fuse data from different sources in space, achieving seamless connection and unified expression of data.

[0088] Preferably, step S36 includes the following steps:

[0089] Step S361: constructing a tensor field for the initial image-point cloud registration data to obtain image-point cloud tensor field data;

[0090] Specifically, the initial image-point cloud registration data, including the image pixel value matrix and the point cloud 3D coordinate matrix, can be imported into TensorFlow software. Using tensor operations in TensorFlow, the image and point cloud data are tensorized and converted into tensor field data. During the tensor field construction process, the tensor's dimensions and shape parameters are set. For example, the tensor dimensions for image data are set to (height, width, channels), where height is the image height, width is the image width, and channels is the number of image channels (e.g., RGB channels). The tensor dimensions for point cloud data are set to (points, features), where points is the number of points in the point cloud and features is the feature dimension of each point (e.g., 3D coordinates, normal vectors, etc.). Through TensorFlow's tensor operations, the image and point cloud data are fused into a unified tensor field, which contains both the image texture information and the point cloud geometry information. For example, in a certain area of a slope, the image tensor field contains the rock texture characteristics of that area, while the point cloud tensor field contains the terrain characteristics of that area. Through tensor field construction, image-point cloud tensor field data can be obtained.

[0091] Step S362: performing principal direction decomposition on the image-point cloud tensor field data to obtain principal direction data of the image-point cloud tensor;

[0092] Specifically, the image-point cloud tensor field data can be imported into MATLAB, and the data is stored in the form of a multi-dimensional matrix. The tensor decomposition function of MATLAB, such as the tensor_decomp function, is called to perform principal direction decomposition on the tensor field data. During the principal direction decomposition process, the decomposition parameters, such as the decomposition rank and the number of iterations, are set. For example, setting the decomposition rank to 5 means that the tensor field is decomposed into 5 principal direction components; the number of iterations is set to 100 times. MATLAB calculates the principal direction components and corresponding weight coefficients of the tensor field data through the alternating least squares method. For example, in a certain area of the slope, the principal direction decomposition results show that the first principal direction component mainly reflects the rock texture direction of the area, and the second principal direction component mainly reflects the slope direction of the terrain. Through principal direction decomposition, the principal direction data of the image-point cloud tensor can be obtained, which reveals the main structural features and directional information in the slope image and point cloud data.

[0093] Step S363: constructing image-point cloud radial basis functions on the initial image-point cloud registration data according to the main direction data of the image-point cloud tensor to obtain image-point cloud interpolation basis functions;

[0094] Specifically, the image-point cloud tensor principal direction data can be imported into SciPy. This data includes the texture feature points of the image and the geometric feature points of the point cloud, as well as their corresponding principal direction information. Using the radial basis function (RBF) module in SciPy, an interpolation basis function is constructed for the initial image-point cloud registration data based on the principal direction data. During the construction process, the type and parameters of the radial basis function are set, and the Gaussian radial basis function (RBF) is selected as the interpolation basis function. The shape parameter of the radial basis function is set to 1.0, which affects the smoothness of the function and the interpolation accuracy. For example, in steep areas of a slope, where the image and point cloud data are densely distributed, Gaussian RBF interpolation can accurately capture the topographic changes and image texture characteristics of this area. Based on the principal direction data of the image-point cloud tensor and the set radial basis function parameters, SciPy calculates the image-point cloud interpolation basis function. This basis function can smoothly interpolate the image and point cloud data spatially, filling gaps and discontinuities in the data.

[0095] Step S364: performing fine registration on the initial image-point cloud registration data according to the image-point cloud interpolation basis function to obtain image-point cloud registration data;

[0096] Specifically, the initial image-point cloud registration data and interpolation basis functions can be imported into CloudCompare. The software's registration tool is then activated, and the interpolation basis functions are used to perform fine spatial transformations and adjustments on the image and point cloud data. During the fine registration process, registration parameters are set, such as a registration accuracy of 0.005 meters, which is higher than the initial registration accuracy. For example, at a corner of a slope, there may be a slight misalignment or deviation between the image data and the point cloud data. Guided by the interpolation basis functions, CloudCompare can precisely adjust the pixel positions of the image data and the point coordinates of the point cloud data to ensure perfect spatial alignment. Furthermore, the software also performs local optimization on the registered data based on the smoothing properties of the interpolation basis functions to eliminate registration errors caused by data noise or irregular distribution. After fine registration, the image-point cloud registration data is obtained. The registration results show that the image and point cloud data are highly consistent in spatial position, with a registration error of less than 0.005 meters. For example, in the crack area of the slope, the crack texture in the image data is highly matched with the crack morphology in the point cloud data, clearly showing the direction and width of the crack.

[0097] Step S365: performing coordinate uniform conversion on the image-point cloud registration data and the slope internal structure data to obtain initial slope image-point cloud unified data;

[0098] Specifically, the image-point cloud registration data and slope internal structure data can be imported into Leica GeoOffice. The image-point cloud registration data contains precisely registered image texture information and point cloud geometry, while the slope internal structure data includes geological structural information derived from seismic reflection and electromagnetic tomography. The software's coordinate conversion tools are used to convert the two sets of data to a unified coordinate system. During the coordinate conversion process, the target coordinate system is set to WGS84UTM Zone45N. Based on the original coordinate system parameters of the image-point cloud registration data and the slope internal structure data, the software automatically calculates the transformation matrix and accurately converts both sets of data to the target coordinate system. For example, the original coordinate system of the image-point cloud registration data is a local rectangular coordinate system, while the original coordinate system of the slope internal structure data is a geographic coordinate system. Through coordinate conversion, both sets of data are converted to the WGS84UTM Zone 45N coordinate system. The converted data are perfectly aligned in space, ultimately resulting in the initial unified slope image-point cloud data.

[0099] Step S366: Perform spatial registration evaluation on the initial slope image-point cloud unified data to obtain registration accuracy evaluation data, and perform registration error compensation on the initial slope image-point cloud unified data based on the registration accuracy evaluation data to obtain slope image-point cloud unified data.

[0100] Specifically, the initial unified slope image-point cloud data can be imported into CloudCompare, and the software's spatial registration assessment tool can be launched. The assessment parameters, such as a registration error threshold of 0.01 meters, are set based on the accuracy requirements for slope modeling and the actual data. The software calculates the registration error for each point in the data—the difference between the point's actual and theoretical positions—to generate registration accuracy assessment data. For example, in a flat area of the slope, the registration error is generally less than 0.005 meters. However, in a steep area, the registration error is larger, with some points exceeding 0.01 meters. Based on the registration accuracy assessment data, CloudCompare's local optimization function fine-tunes the data in areas with large errors. By adjusting properties such as point positions and normal vectors, the registration error is reduced to within the threshold. For example, in steep areas, the normal vectors of the point cloud data are optimized to better conform to the terrain geometry, thereby reducing the registration error. After compensation for the registration error, the unified slope image-point cloud data is finally obtained.

[0101] The present invention can accurately capture the spatial variation characteristics and main direction information of the data by constructing an image-point cloud tensor field and performing main direction decomposition. By using the image-point cloud radial basis function for interpolation, the initial image-point cloud registration data can be effectively smoothed to fill the gaps and discontinuities in the data. This can better adapt to complex terrain and data distribution characteristics, making the registered data smoother and more continuous. Through unified coordinate conversion, the image-point cloud registration data and the slope internal structure data are accurately aligned in space, achieving seamless connection and unified expression between different data sources. Through spatial registration evaluation, registration accuracy evaluation data is obtained, and the registration results can be comprehensively tested and analyzed for accuracy. Registration error compensation based on the evaluation data can further improve the accuracy and reliability of the registration results and ensure the accuracy and consistency of the data in space.

[0102] Preferably, step S4 includes the following steps:

[0103] Step S41: performing spatial gradient distribution recognition on the slope image-point cloud unified data to obtain slope spatial gradient field data;

[0104] Specifically, the unified slope image-point cloud data can be imported into MATLAB. The data is stored as a multidimensional matrix, containing both image texture information and point cloud geometry. Using the Image Processing Toolbox in MATLAB, spatial gradient distribution identification is performed on the data. During the identification process, the Sobel operator is selected as the gradient detection operator. The Sobel operator effectively detects edges and gradient changes in the data. The Sobel operator's window size is set to 3×3 pixels. For example, in a rock edge region of the slope, the Sobel operator accurately detects the gradient change between the rock and soil layers, obtaining the gradient value for that region. MATLAB calculates the gradient magnitude and direction at each point in the data to obtain the slope's spatial gradient field data. Gradient field data is represented as vectors, with the length of each vector representing the gradient magnitude and the direction representing the gradient direction. For example, in a steep region of the slope, a long gradient vector pointing downward indicates a steeper slope and a downward slope.

[0105] Step S42: Selecting marker points from the slope spatial gradient field data to obtain slope watershed marker data;

[0106] Specifically, slope spatial gradient field data can be imported into ArcGIS and stored as raster data, with each grid cell corresponding to a gradient vector. Using ArcGIS's spatial analysis tools, local extreme value detection is performed on the gradient field data to select watershed marker points. During marker point selection, local extreme value threshold parameters are set, such as a gradient amplitude threshold of 10 for the extreme value and a 5×5 grid cell for the extreme value neighborhood. These parameters are determined based on the slope's topographic characteristics and the distribution of the gradient field data. For example, at the top of a slope, the gradient amplitude is large, while the gradient amplitude in the surrounding neighborhood is small, meeting the extreme value criteria and thus being selected as a watershed marker point. ArcGIS identifies local extreme value points that meet these criteria by calculating the gradient amplitude of each grid cell and the gradient variation within the neighborhood, and marks them as watershed marker points. For example, between two adjacent ridges on a slope, there is a clear watershed, and the marker points along this watershed appear as a continuous line. Through marker point selection, slope watershed marker data is obtained.

[0107] Step S43: performing distance transformation on the slope watershed marker data to obtain a slope watershed distance map;

[0108] Specifically, slope watershed marker data can be imported into OpenCV and stored as a binary image, with watershed markers as white pixels and background pixels as black pixels. Using OpenCV's distance transformation functions, such as cv2.distanceTransform, a distance transformation is performed on the watershed marker data. During the distance transformation, the distance type is set to DIST_L2 (Euclidean distance), which calculates the straight-line distance from a pixel to the nearest white pixel (watershed marker). The mask size is also set to 3×3 pixels. For example, in a certain area of the slope, the grayscale value of each pixel in the resulting image after distance transformation represents its distance to the nearest watershed marker. Pixels closer to the watershed marker have lower grayscale values, while pixels farther away have higher grayscale values. This distance transformation yields a slope watershed distance map, which displays the distribution of distances from each pixel to the watershed marker in the form of a grayscale image.

[0109] Step S44: performing watershed growth and diffusion on the marked points in the slope watershed marked data according to the slope watershed distance map to obtain initial slope segmentation data;

[0110] Specifically, the slope watershed marker data and the watershed distance map can be imported into MATLAB, and the region growing algorithm in MATLAB can be used to perform watershed growth and diffusion on the watershed marker points. During the watershed growth and diffusion process, the growth threshold parameter is set, such as the growth distance threshold is 5 pixels. This threshold is set based on the terrain characteristics of the slope and the distribution of the watershed distance map. For example, at the top of a certain mountain on the slope, starting from the watershed marker point, growth and diffusion are performed along the low grayscale value area in the distance map until a pixel point with a grayscale value greater than the growth distance threshold is encountered in the distance map. MATLAB gradually expands the watershed range of the watershed marker point through iterative calculations, merging adjacent pixels into the same watershed. For example, between two adjacent ridges on the slope, two independent watershed areas are formed after watershed growth and diffusion, corresponding to two different watershed marker points. Through watershed growth and diffusion, the initial segmentation data of the slope can be obtained.

[0111] Step S45: performing boundary optimization on the initial slope segmentation data to obtain slope unit division data;

[0112] Specifically, the initial segmentation data of the slope can be imported into ImageJ, and the data is stored in the form of a segmented binary image, in which different unit areas are represented by different pixel values. The boundary smoothing and optimization tools in ImageJ are used to refine the segmented boundaries. In the boundary optimization process, the opening and closing operations in morphological operations are selected as the main tools. The opening operation can remove small burrs and irregular details on the boundary, while the closing operation can fill small gaps and holes on the boundary. The structural element size of the opening and closing operations is set to 3×3 pixels. For example, in a certain unit area of the slope, there are some small bumps and burrs on the boundary after initial segmentation. Through the processing of opening and closing operations, the boundary becomes smoother and the shape of the unit area becomes more regular. Through the above steps, the slope unit division data can be obtained.

[0113] Step S46: performing structural surface identification on the slope unit division data to obtain slope structural surface feature data, and performing feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data.

[0114] Specifically, please refer to the sub-steps of step S46 for the detailed implementation process of this embodiment.

[0115] The present invention can accurately identify the natural dividing lines and unit boundaries of the slope through spatial gradient distribution recognition and watershed marking. Through the process of distance transformation and watershed growth and diffusion, the noise and discontinuity in the slope data can be effectively processed, making the segmentation result more stable and robust. Distance transformation can quantify the distance relationship between the marked point and the surrounding points, while watershed growth and diffusion can reasonably expand the area based on the distance information, which can reduce the segmentation error caused by data noise or local discontinuity. Through structural surface recognition, the structural surface features in the slope, such as joints and cracks, can be accurately identified. By combining the internal structural data of the slope for feature extraction, the hierarchical feature data of the slope can be further enriched and improved, so that the model can more realistically reflect the structural hierarchy and geological characteristics of the slope. By extracting and fusing the features of the structural surface feature data and the internal structural data, the hierarchical feature data of the slope obtained can effectively integrate and hierarchically express the surface features and internal structural features of the slope.

[0116] Preferably, step S46 includes the following steps:

[0117] Step S461: Calculating the surface curvature of the slope unit division data to obtain slope surface curvature distribution data;

[0118] Specifically, the slope unit division data can be imported into CloudCompare, and the data is stored in the form of a point cloud. Each point contains three-dimensional coordinate information, and the surface curvature calculation tool in the software is used to perform curvature analysis on the surface of each unit area. In the curvature calculation process, Gaussian curvature is selected as the calculation indicator. The neighborhood radius of the curvature calculation is set to 0.5 meters. For example, in a rock unit area of the slope, the surface is relatively steep and there are some uneven features. The curvature distribution map of the area is obtained through Gaussian curvature calculation. The map shows that the curvature value of the rock surface is high, and the curvature value changes significantly in the concave and convex parts. CloudCompare software generates slope surface curvature distribution data based on the curvature value of each point.

[0119] Step S462: extracting height difference characteristic lines of the mountain slope based on the slope surface curvature distribution data to obtain the slope height difference characteristic lines;

[0120] Specifically, slope surface curvature distribution data can be imported into MATLAB and stored as a two-dimensional matrix, where each element in the matrix corresponds to the curvature value of a point. The Canny edge detection algorithm in MATLAB is then used to extract height difference characteristic lines from the curvature distribution data. During the feature line extraction process, the threshold parameters of the Canny algorithm are set, such as a high threshold of 0.8 and a low threshold of 0.2. These thresholds are set based on the range of the curvature distribution data and the requirements for feature line extraction. The high threshold is used to detect edges with obvious height differences, while the low threshold is used to connect edge breakpoints. For example, in a certain area of the slope, there is a significant height difference at the junction of the rock layer and the soil layer. The Canny algorithm can accurately detect the edge of this junction and extract the height difference characteristic lines. MATLAB identifies the height difference characteristic lines by calculating the gradient amplitude and direction in the curvature distribution data and stores them as vector line elements, with each line element representing a height difference characteristic line.

[0121] Step S463: performing structural surface trend statistics on the slope height difference characteristic lines to obtain structural surface direction distribution data;

[0122] Specifically, the slope height difference characteristic line data can be imported into ArcGIS, and the "Azimuth" tool in ArcGIS can be used to calculate the direction of each characteristic line. This tool can automatically calculate the azimuth of the line feature, that is, the angle between the line feature and the north, so as to obtain the strike angle of each characteristic line. In the statistical process, the statistical interval angle is set to 10 degrees, and the azimuth range of 0-360 degrees is divided into 36 intervals, each interval representing a strike direction. For example, the interval of 0-10 degrees represents a strike direction close to the north. Through the spatial analysis function of ArcGIS, the strike angles of all characteristic lines are counted, and the number and proportion of characteristic lines in each strike direction are calculated. For example, in a certain area of the slope, the statistical results show that the number of characteristic lines with a strike angle of 45-55 degrees is the largest, accounting for 30%, indicating that the structural surface in this area is mainly northeast-southwest trending, and finally the structural surface direction distribution data is obtained.

[0123] Step S464: performing structural surface clustering on the structural surface direction distribution data to obtain slope structural surface family data;

[0124] Specifically, the structural surface direction distribution data can be imported into Python, and the data is stored in the form of feature vectors. Each feature vector contains a strike angle and the corresponding number of feature lines. The K-means clustering algorithm is selected as the clustering method. This algorithm can divide the data into multiple clusters based on the similarity of the data. In the clustering process, the number of clusters is set to 5. At the same time, the initialization method of the algorithm is set to "k-means++". Scikit-learn divides the structural surface into 5 clusters through iterative calculation based on the distance between the feature vectors. Each cluster represents a structural surface family. For example, one of the clusters contains structural surfaces with a strike angle in the range of 40-60 degrees and a large number of feature lines, indicating that this is a major structural surface family. Through the above steps, the slope structural surface family data can be obtained.

[0125] Step S465: performing spatial correlation measurement on the slope structural surface family data to obtain slope structural surface correlation data;

[0126] Specifically, the slope structural surface family data can be imported into the R language, and the data is stored in the form of spatial objects. Each spatial object represents a structural surface family, including its spatial position and geometric characteristics. The gDistance function in the rgeos package is used to calculate the spatial distance between different structural surface families. For example, the minimum distance between two adjacent structural surface families is calculated to evaluate their spatial proximity. The gIntersection function can also be used to calculate the spatial intersection area between structural surface families to evaluate their spatial overlap. Through the above steps, the slope structural surface correlation data can be obtained and stored in the form of a numerical matrix. Each element in the matrix represents the correlation measure between the two structural surface families. For example, the value of an element in the matrix is 0.05, which means that the minimum distance between the two structural surface families is 0.05 meters, indicating that they are relatively close in space and there is a certain degree of mutual influence.

[0127] Step S466: combining the structural surfaces according to the slope structural surface correlation data to obtain slope structural surface characteristic data;

[0128] Specifically, the structural surface correlation data can be imported into Python to construct a weighted undirected graph. The nodes in the graph represent structural surface families, the edges represent the correlation relationships between structural surface families, and the weights of the edges are the correlation metrics. The graph is clustered using NetworkX's graph clustering algorithm, such as the Louvain community discovery algorithm. This algorithm identifies the community structure in the graph by maximizing the modularity, thereby achieving the combination of structural surface families. During the clustering process, the resolution parameter of the algorithm is set to 1.0. For example, the clustering results show that several originally scattered structural surface families are combined into a large structural surface feature area. These families are spatially adjacent and have similar strike and dip characteristics, indicating that they belong to the same geological structural unit. Through the above steps, the slope structural surface feature data can be obtained, and the spatial distribution and correlation relationship of the structural surface can be displayed in the form of a graph.

[0129] Step S467: performing stress field evolution on the slope structural surface characteristic data to obtain slope stress field data, and identifying weak surfaces of the mountain slope based on the slope stress field data to obtain slope weak surface data;

[0130] Specifically, slope structural surface characteristic data can be imported into ABAQUS to construct a 3D geometric model of the slope. Based on geological and experimental data, mechanical parameters such as elastic modulus, Poisson's ratio, and shear strength are assigned to the various geotechnical elements in the model. Using ABAQUS's static analysis module, external loads and boundary conditions are applied to the slope model to simulate the stress distribution under different conditions. For example, when simulating rainfall conditions, the effect of water pressure on slope stability is considered by applying seepage pressure to the geotechnical elements to simulate water infiltration and saturation. Finite element calculations yield the slope's stress field data, including the stress tensor, principal stress directions, and stress concentration areas for each element. Using ABAQUS's post-processing capabilities, stress concentration areas and regions with a small angle between the principal stress direction and the structural surface strike are identified as weak points in the slope. For example, in a certain area of the slope, if the stress concentration factor reaches 1.5 and the angle between the principal stress direction and the structural surface strike is less than 30 degrees, this indicates a weak point with a risk of sliding or shear failure. Through the above steps, the weak surface data of the slope can be obtained.

[0131] Step S468: Perform stability assessment on the weak surface data of the slope to obtain weak surface stability assessment data, and perform multi-scale fusion of the weak surface stability assessment data and the internal structure data of the slope to obtain hierarchical characteristic data of the slope.

[0132] Specifically, the weak surface data of the slope can be imported into Slope / W. Combined with the simple geometric model of the slope and the parameters of the geotechnical mass, the strip method in the limit equilibrium method is selected for stability analysis. During the stability assessment process, the analysis parameters are set. For example, the safety factor is calculated using the Morgenstern-Price method, which considers the non-uniform sliding surfaces between strips and the interaction forces between strips. At the same time, the analysis conditions are set, such as considering the effects of water pressure under different rainfall conditions and the effects of seismic forces under different earthquake intensities. For example, when simulating heavy rainfall conditions, the rainfall intensity is input as 100 mm / hour to calculate the effect of water pressure on slope stability; when simulating earthquake conditions, the seismic acceleration is input as 0.3g to evaluate the effect of seismic forces on the stability of the weak surface of the slope. Through the calculation of Slope / W, the weak surface stability assessment data is obtained, including information such as the safety factor, the location of the sliding surface, and the volume of the sliding body. The weak surface stability assessment data is then integrated with the internal structure data of the slope at multiple scales. Using GeoStudio's multiphysics coupling analysis capabilities, we combined the stability assessment results with the slope's geological structure, geotechnical parameters, and stress field data to generate hierarchical slope characteristic data. For example, in one area of the slope, the fusion results showed a safety factor of 1.2 for the weak plane, and the presence of faults and weak interlayers in the area, indicating poor stability. Through these steps, we can obtain hierarchical slope characteristic data.

[0133] The present invention can accurately identify the structural surface features of the slope by calculating the surface curvature and extracting the height difference characteristic lines of the slope unit division data. This can comprehensively reveal the structural surface features of the slope, such as joints and cracks, and improve the accuracy and comprehensiveness of the structural surface feature extraction. Through structural surface strike statistics, the strike distribution of the slope structural surface can be accurately analyzed. By clustering and spatially correlating the structural surface direction distribution data, the structural surface data can be effectively organized and associated. This makes the structural surface data more logical and systematic. By combining the stress field evolution and weak surface identification, the stability of the slope can be evaluated and predicted, which enables the structural characteristic data of the slope to better serve the stability analysis and disaster prevention of the slope.

[0134] Preferably, step S5 includes the following steps:

[0135] Step S51: stratifying the hierarchical feature data of the slope to obtain layered slope data, and extracting surface point cloud from the layered slope data to obtain discrete surface point cloud data of the slope;

[0136] Specifically, the hierarchical feature data of the slope can be imported into Python, and the array slicing function of NumPy can be used to divide the data into different layered data according to the layer identifiers in the hierarchical feature data. For example, the hierarchical feature data of the slope contains three layers: surface, middle, and deep. Through array slicing operations, the surface data, middle data, and deep data are extracted respectively. Surface point clouds are extracted from the layered slope data. Clustering analysis is performed on the point clouds in the surface data using SciPy's clustering algorithms, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise). The parameters of the DBSCAN algorithm are set, such as a neighborhood radius of 0.1 meters and a minimum number of points of 10, to identify dense areas and discrete points in the surface point cloud. Through clustering analysis, discrete surface point cloud data of the slope are extracted, which contain surface terrain feature points and structural feature points.

[0137] Step S52: performing surface reconstruction on the discrete surface point cloud data of the slope to obtain a spatial surface model of the slope;

[0138] Specifically, you can import discrete surface point cloud data into MeshLab, start MeshLab's surface reconstruction tool, select the Poisson surface reconstruction algorithm, and during the reconstruction process, set the algorithm parameters, such as the depth to 9, that is, the maximum depth of the octree is 9; at the same time, set the normal vector consistency parameter to 1 to ensure that the normal vector of the reconstructed surface is consistent with the normal vector in the point cloud data. MeshLab calculates the normal vector field of the point cloud data and solves the Poisson equation to gradually construct the surface of the slope. For example, in a corner area of the slope, the surface reconstruction results show clear corner shapes and edge details. Finally, the spatial surface model of the slope is obtained, which is presented in the form of a three-dimensional mesh.

[0139] Step S53: extracting middle-layer features from the slope spatial surface model to obtain slope spatial middle-layer feature data, and performing implicit surface middle-layer reconstruction on the slope spatial middle-layer feature data to obtain a slope spatial middle-layer model;

[0140] Specifically, the spatial surface model of the slope can be imported into MATLAB, and the image processing toolbox in MATLAB can be used to extract the middle-layer features of the surface model. The Laplacian of Gaussian (LoG) operator is selected as the feature extraction operator. The LoG operator can detect the edge and curvature features in the model. The parameters of the LoG operator, such as σ (standard deviation), are set to 2 pixels. For example, in a certain rock layer area of the slope, the LoG operator can accurately extract the edge features and curvature change features of the rock layer, and obtain the middle-layer feature data of the area. Using the implicit surface reconstruction function in MATLAB, such as the isosurface function, the isovalue surface is extracted as the implicit surface according to the value of the middle-layer feature data. The isovalue threshold is set to 0.5, that is, the isovalue surface with the middle-layer feature data value of 0.5 is extracted. Through the above steps, the spatial middle-layer model of the slope is obtained.

[0141] Step S54: performing deep interpolation grid division on the slope space middle layer model to obtain the slope space deep layer grid, and performing deep reconstruction on the slope space deep layer grid to obtain the slope space deep layer model;

[0142] Specifically, the middle-layer model of the slope space can be imported into ANSYS, and the meshing tool of ANSYS can be used to perform deep interpolation meshing on the middle-layer model. Select tetrahedral mesh as the mesh type, and set the mesh size parameters, such as the maximum mesh size is 0.5 meters and the minimum mesh size is 0.1 meters. For example, in a steep area of the slope, the mesh size is smaller during meshing; while in the gentle area of the slope, the mesh size is larger. Through meshing, the deep mesh of the slope space is obtained. Then, the finite element analysis function of ANSYS is used to optimize and reconstruct the deep mesh. For example, the shape and distribution of the mesh are optimized, and singular points and too small mesh units in the mesh are eliminated. Through deep reconstruction, the deep model of the slope space is obtained.

[0143] Step S55: integrating the slope spatial surface model, the slope spatial middle layer model and the slope spatial deep layer model to obtain a complete slope three-dimensional model.

[0144] Specifically, the slope surface model, the slope middle model, and the slope deep model can be imported separately into SolidWorks. Using SolidWorks' assembly functionality, the three models can be precisely aligned and combined. First, the slope surface model is selected as the base and placed at the bottom of the assembly to ensure its alignment with the actual terrain. Then, the slope middle model is placed above the surface model. The position and rotation of the middle model are adjusted to align it with the edges and structural features of the surface model. For example, at a certain corner of the slope, the edge of the middle model is aligned with the corner edge of the surface model. Finally, the slope deep model is placed below the middle model and its depth and range are adjusted to connect it with the bottom structure of the middle model. For example, at a certain rock layer area of the slope, the rock layer of the deep model connects with the rock layer of the middle model, forming a complete rock structure. Through these steps, a complete three-dimensional slope model is obtained, presented as a unified assembly.

[0145] The present invention can effectively separate and reconstruct the different levels of features of the slope by performing layered processing on the hierarchical feature data of the slope. This enables the model to more finely reflect the hierarchical structure and feature changes of the slope, improving the detail expression and accuracy of the model. By using the surface point cloud data for surface reconstruction, the external morphology and surface features of the slope can be accurately reconstructed. This can capture the subtle changes and local features of the slope. Through implicit surface reconstruction and deep grid interpolation reconstruction, the internal structural features of the slope can be effectively expressed. Implicit surface reconstruction can smoothly represent the complex surface features of the middle layer, while interpolation reconstruction can accurately depict the structural details of the deep layer, so that the middle and deep layer models can better reflect the internal structure and hierarchical relationship of the slope, improving the structural expression ability of the model. By integrating the surface, middle and deep layer models, the complete three-dimensional slope model obtained can effectively integrate and uniformly express the external morphology, internal structure and hierarchical features of the slope.

[0146] Preferably, the present invention further provides a 3D modeling system for slopes in difficult mountainous areas based on multi-source data fusion, which is used to execute the 3D modeling method for slopes in difficult mountainous areas based on multi-source data fusion as described above. The 3D modeling system for slopes in difficult mountainous areas based on multi-source data fusion comprises:

[0147] The data acquisition module is used to use a cluster of drones to collect multi-angle data on mountain slopes to obtain multi-angle image data of the slopes; and to perform ground laser scanning on mountain slopes to obtain slope point cloud data sets;

[0148] The structural detection module is used to perform seismic reflection wave detection on mountain slopes to obtain slope seismic reflection data; perform electromagnetic tomography detection on mountain slopes to obtain slope electromagnetic tomography data; and fuse and invert slope seismic reflection data and slope electromagnetic tomography data to obtain slope internal structure data.

[0149] The data registration and unification module is used to perform coarse registration of the slope multi-angle image data and the slope point cloud data set to obtain the initial image-point cloud registration data; perform tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data; and perform spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data.

[0150] The segmentation module is used to perform improved watershed segmentation on the unified slope image-point cloud data to obtain slope unit division data; perform structural surface recognition on the slope unit division data to obtain slope structural surface feature data; perform feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data;

[0151] The three-dimensional reconstruction module is used to stratify the hierarchical characteristic data of the slope to obtain the slope layer data; to reconstruct the surface of the slope layer data to obtain the slope spatial surface model; and to reconstruct the mountain slope in three dimensions based on the slope spatial surface model to obtain a complete slope three-dimensional model.

[0152] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0153] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion, characterized in that: The following steps are involved: Step S1: Use a drone cluster to collect multi-angle data of the mountain slope to obtain multi-angle image data of the slope; perform ground laser scanning on the mountain slope to obtain a slope point cloud dataset; Step S2: Perform seismic reflection wave detection on the mountain slope to obtain slope seismic reflection data; perform electromagnetic tomography detection on the mountain slope to obtain slope electromagnetic tomography data; fuse and invert the slope seismic reflection data and the slope electromagnetic tomography data to obtain slope internal structure data; Step S3: performing coarse registration of the slope multi-angle image data and the slope point cloud data set to obtain initial image-point cloud registration data; performing tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data; performing spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data; Step S4: performing improved watershed segmentation on the slope image-point cloud unified data to obtain slope unit division data; performing structural surface recognition on the slope unit division data to obtain slope structural surface feature data; performing feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data; Step S5: Layer the slope hierarchical feature data to obtain slope layer data; reconstruct the surface of the slope layer data to obtain a slope spatial surface model; and reconstruct the mountain slope in three dimensions based on the slope spatial surface model to obtain a complete slope three-dimensional model.

2. The three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: collecting a side panoramic image of the mountain slope to obtain a side panoramic image of the slope, and performing regional mission planning for the UAV cluster based on the side panoramic image of the slope to obtain a UAV flight mission; Step S12: assigning routes to the drone cluster based on the drone flight mission to obtain drone route planning data; Step S13: performing an aerial laser ranging scan on the side of the mountain slope to obtain slope characteristic data; Step S14: highly stratify the UAV route planning data according to the slope gradient characteristic data to obtain UAV layered collection data; Step S15: using a drone cluster to collect multi-angle data on the mountain slope based on the layered data collected by the drone to obtain multi-angle image data of the slope; Step S16: Perform laser radar site planning on the mountain slope according to the slope gradient characteristic data to obtain laser scanning site distribution data, and perform multi-site scanning on the mountain slope according to the laser scanning site distribution data to obtain a slope point cloud dataset.

3. The three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: Arrange seismic wave sources on the mountain slopes to obtain seismic source arrangement data; Step S22: Arranging receiver arrays on mountain slopes according to the seismic source layout data to obtain slope receiver layout data; Step S23: configuring the trigger timing of the receiver array according to the slope receiver layout data to obtain receiver trigger sequence data; Step S24: performing seismic wave excitation on the mountain slope according to the receiver trigger sequence data to obtain slope seismic wave data, and performing first arrival wave identification on the slope seismic wave data to obtain slope wave arrival time data; Step S25: performing travel time inversion on the slope seismic wave data according to the slope wave arrival time data to obtain seismic wave velocity structure data, and performing waveform inversion on the seismic wave velocity structure data to obtain slope seismic reflection data; Step S26: Perform electromagnetic tomography detection on the mountain slope to obtain slope electromagnetic tomography data, and fuse and invert the slope seismic reflection data and the slope electromagnetic tomography data to obtain slope internal structure data.

4. The three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion according to claim 3 is characterized in that: Step S26 includes the following steps: Step S261: Arranging electromagnetic transmitting sources on the mountain slope to obtain electromagnetic source arrangement data, and arranging receiving electrodes on the mountain slope based on the electromagnetic source arrangement data to obtain electrode arrangement data; Step S262: configuring the excitation waveform of the electromagnetic emission source according to the electrode layout data to obtain electromagnetic excitation waveform data; Step S263: performing electromagnetic field measurement on the receiving electrode according to the electromagnetic excitation waveform data to obtain slope electromagnetic field data; Step S264: performing inversion and reconstruction on the slope electromagnetic field data to obtain slope conductivity distribution data; Step S265: reconstructing the slope electromagnetic data according to the slope conductivity distribution data to obtain slope electromagnetic tomography data; Step S266: fusing and inverting the slope seismic reflection data and the slope electromagnetic tomography data to obtain the slope internal structure data.

5. The three-dimensional modeling method for slopes in difficult mountainous areas based on multi-source data fusion according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: extracting feature points from the multi-angle image data of the slope to obtain a slope image feature point set, and extracting feature points from the slope point cloud data set to obtain a slope point cloud feature point set; Step S32: matching the slope image feature point set with the slope point cloud feature point set to obtain initial slope matching point pair data; Step S33: screening the initial matching point pair data of the slope to obtain the effective matching point pair data of the slope; Step S34: performing ICP iterative registration on the slope image feature point set and the slope point cloud feature point set according to the slope effective matching point pair data to obtain a feature point registration transformation matrix; Step S35: performing spatial transformation on the slope multi-angle image data and the slope point cloud data set according to the feature point registration transformation matrix to obtain initial image-point cloud registration data; Step S36: Perform tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data, and perform spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data.

6. The method for three-dimensional modeling of slopes in difficult mountainous areas based on multi-source data fusion according to claim 5 is characterized in that: Step S36 includes the following steps: Step S361: constructing a tensor field for the initial image-point cloud registration data to obtain image-point cloud tensor field data; Step S362: performing principal direction decomposition on the image-point cloud tensor field data to obtain principal direction data of the image-point cloud tensor; Step S363: constructing image-point cloud radial basis functions on the initial image-point cloud registration data according to the main direction data of the image-point cloud tensor to obtain image-point cloud interpolation basis functions; Step S364: performing fine registration on the initial image-point cloud registration data according to the image-point cloud interpolation basis function to obtain image-point cloud registration data; Step S365: performing coordinate uniform conversion on the image-point cloud registration data and the slope internal structure data to obtain initial slope image-point cloud unified data; Step S366: Perform spatial registration evaluation on the initial slope image-point cloud unified data to obtain registration accuracy evaluation data, and perform registration error compensation on the initial slope image-point cloud unified data based on the registration accuracy evaluation data to obtain slope image-point cloud unified data.

7. The method for three-dimensional modeling of slopes in difficult mountainous areas based on multi-source data fusion according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: performing spatial gradient distribution recognition on the slope image-point cloud unified data to obtain slope spatial gradient field data; Step S42: Selecting marker points from the slope spatial gradient field data to obtain slope watershed marker data; Step S43: performing distance transformation on the slope watershed marker data to obtain a slope watershed distance map; Step S44: performing watershed growth and diffusion on the marked points in the slope watershed marked data according to the slope watershed distance map to obtain initial slope segmentation data; Step S45: performing boundary optimization on the initial slope segmentation data to obtain slope unit division data; Step S46: performing structural surface identification on the slope unit division data to obtain slope structural surface feature data, and performing feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data.

8. The method for three-dimensional modeling of slopes in difficult mountainous areas based on multi-source data fusion according to claim 7 is characterized in that: Step S46 includes the following steps: Step S461: Calculating the surface curvature of the slope unit division data to obtain slope surface curvature distribution data; Step S462: extracting height difference characteristic lines of the mountain slope based on the slope surface curvature distribution data to obtain the slope height difference characteristic lines; Step S463: performing structural surface trend statistics on the slope height difference characteristic lines to obtain structural surface direction distribution data; Step S464: performing structural surface clustering on the structural surface direction distribution data to obtain slope structural surface family data; Step S465: performing spatial correlation measurement on the slope structural surface family data to obtain slope structural surface correlation data; Step S466: combining the structural surfaces according to the slope structural surface correlation data to obtain slope structural surface characteristic data; Step S467: performing stress field evolution on the slope structural surface characteristic data to obtain slope stress field data, and identifying weak surfaces of the mountain slope based on the slope stress field data to obtain slope weak surface data; Step S468: Perform stability assessment on the weak surface data of the slope to obtain weak surface stability assessment data, and perform multi-scale fusion of the weak surface stability assessment data and the internal structure data of the slope to obtain hierarchical characteristic data of the slope.

9. The method for three-dimensional modeling of slopes in difficult mountainous areas based on multi-source data fusion according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: stratifying the hierarchical feature data of the slope to obtain layered slope data, and extracting surface point cloud data from the layered slope data to obtain discrete surface point cloud data of the slope; Step S52: performing surface reconstruction on the discrete surface point cloud data of the slope to obtain a spatial surface model of the slope; Step S53: extracting middle-layer features from the slope spatial surface model to obtain slope spatial middle-layer feature data, and performing implicit surface middle-layer reconstruction on the slope spatial middle-layer feature data to obtain a slope spatial middle-layer model; Step S54: performing deep interpolation grid division on the slope space middle layer model to obtain the slope space deep layer grid, and performing deep reconstruction on the slope space deep layer grid to obtain the slope space deep layer model; Step S55: integrating the slope spatial surface model, the slope spatial middle layer model and the slope spatial deep layer model to obtain a complete slope three-dimensional model.

10. A 3D modeling system for slopes in difficult mountainous areas based on multi-source data fusion, characterized by: For executing the method for three-dimensional modeling of slopes in difficult mountainous areas based on multi-source data fusion as claimed in claim 1, the three-dimensional modeling system for slopes in difficult mountainous areas based on multi-source data fusion comprises: The data acquisition module is used to use a cluster of drones to collect multi-angle data on mountain slopes to obtain multi-angle image data of the slopes; and to perform ground laser scanning on mountain slopes to obtain slope point cloud data sets; The structural detection module is used to perform seismic reflection wave detection on mountain slopes to obtain slope seismic reflection data; perform electromagnetic tomography detection on mountain slopes to obtain slope electromagnetic tomography data; and fuse and invert slope seismic reflection data and slope electromagnetic tomography data to obtain slope internal structure data. The data registration and unification module is used to perform coarse registration of the slope multi-angle image data and the slope point cloud data set to obtain the initial image-point cloud registration data; perform tensor field interpolation fine registration on the initial image-point cloud registration data to obtain image-point cloud registration data; and perform spatial unified transformation on the image-point cloud registration data and the slope internal structure data to obtain slope image-point cloud unified data. The segmentation module is used to perform improved watershed segmentation on the slope image-point cloud unified data to obtain slope unit division data; Perform structural surface identification on the slope unit division data to obtain slope structural surface feature data; perform feature extraction on the slope structural surface feature data and the slope internal structure data to obtain slope hierarchical feature data; The three-dimensional reconstruction module is used to stratify the hierarchical characteristic data of the slope to obtain the slope layer data; to reconstruct the surface of the slope layer data to obtain the slope spatial surface model; and to reconstruct the mountain slope in three dimensions based on the slope spatial surface model to obtain a complete slope three-dimensional model.

Images