A method and system for geological disaster monitoring based on UAV vision and target coordination

Abstract

This invention provides a geological disaster monitoring method and system based on UAV vision and target collaboration, belonging to the field of disaster monitoring technology. The method includes: acquiring geological images at different times; performing terrain modeling based on the geological images and target locations to obtain a digital surface model; identifying vegetated and bare areas based on the geological images; constructing a Delaunay triangulation based on targets in the vegetated areas to calculate vegetation displacement; calculating bare land displacement based on the digital surface model; fusing vegetation and bare land displacements using thin-plate spline functions to obtain global displacement information; detecting cracks using an improved Canny-Zernike operator based on the digital surface model and identifying crack skeleton lines using a morphological thinning algorithm; and modeling displacement gradient tensors based on crack skeleton lines and global displacement information to construct a crack opening distribution map and obtain monitoring results. This invention creatively combines physical noise reduction, intelligent fusion, and precise analysis to achieve high-precision positioning in harsh environments.

Description

Technical Field

[0001] This invention relates to the field of disaster monitoring technology, and more specifically, to a geological disaster monitoring method and system based on the collaboration of UAV vision and target. Background Technology

[0002] The field of geological disaster displacement monitoring mainly relies on two types of technical systems: contact and non-contact. Contact technologies, such as crack gauges and inclinometers, are based on the principle of deformation transmission in solid mechanics, achieving displacement measurement through direct coupling between physical sensors and the ground surface. Non-contact technologies are primarily based on unmanned aerial vehicle (UAV) photogrammetry, whose core principle is based on multi-view geometric reconstruction models—using the matching of corresponding feature points between multi-view images to calculate the three-dimensional spatial coordinates of the target point.

[0003] Existing monitoring methods that combine drones with pre-set artificial targets suffer from severe data breaks and missing data when natural surface textures are largely obscured or disturbed by movement in areas covered by shrubs, grass, and other vegetation. This leads to a sharp deterioration in the overall reconstruction accuracy, with the monitoring accuracy of planar and elevation displacements falling far below the sub-centimeter level, failing to meet the basic requirements for early warning. Summary of the Invention

[0004] The purpose of this invention is to provide a geological disaster monitoring method and system based on UAV vision and target coordination, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] In a first aspect, the present invention provides a geological disaster monitoring method based on the coordination of UAV vision and target, comprising:

[0006] Geological images from different times are acquired, and terrain modeling is performed based on the geological images and the target locations in them to obtain a digital surface model.

[0007] Based on geological imagery, vegetation zones and bare areas are identified. A Delaunay triangulation network is constructed based on vegetation zone targets to calculate vegetation displacement. Bare area displacement is calculated based on a digital surface model.

[0008] The displacement information of the entire region is obtained by fusing vegetation displacement and bare ground displacement using thin plate spline functions;

[0009] Based on the digital surface model, an improved Canny-Zernike operator is used for crack detection, and a morphological thinning algorithm is used to identify crack skeleton lines.

[0010] Based on the crack skeleton line and global displacement information, the displacement gradient of the crack region is inverted by solving the Poisson equation, and the displacement gradient tensor is modeled. The crack opening distribution map is constructed by projecting along the normal of the skeleton line to obtain the monitoring results.

[0011] Secondly, the present invention also provides a geological disaster monitoring system based on UAV vision and target coordination, comprising:

[0012] The first construction module is used to acquire geological images at different times, and to perform terrain modeling based on the geological images and the target locations in them to obtain a digital surface model.

[0013] The first calculation module is used to identify vegetated areas and bare areas based on geological images, construct a Delaunay triangulation based on vegetated area targets, and calculate vegetation displacement; it also calculates bare land displacement based on digital surface models.

[0014] The first fusion module is used to fuse vegetation displacement and bare ground displacement using thin plate spline functions to obtain global displacement information;

[0015] The first identification module is used to detect cracks based on a digital surface model, using an improved Canny-Zernike operator, and to identify crack skeleton lines using a morphological thinning algorithm.

[0016] The second module is used to solve the Poisson equation to invert the displacement gradient of the crack region based on the crack skeleton line and global displacement information, perform displacement gradient tensor modeling, and construct the crack opening distribution map by projection along the skeleton line normal to obtain the monitoring results.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention achieves breakthroughs in both displacement field global fusion and crack gradient analysis algorithms. For displacement field fracturing caused by vegetation shading, a target-constrained Delaunay triangulation interpolation model is developed. A virtual displacement field in the vegetated area is generated using the displacement vectors of neighboring targets and the weighted centroid coordinates, and seamlessly fused with the measured point cloud displacement in the bare area using a thin-plate spline function. For micro-crack monitoring, Canny-Zernike sub-pixel edge detection and displacement gradient tensor modeling are combined, and the Poisson equation is solved with target displacement as the boundary condition, achieving accurate inversion of crack opening at the ±0.2mm level.

[0019] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a geological disaster monitoring method based on UAV vision and target collaboration, as described in this application.

[0022] Figure 2 This is a schematic diagram of the geological disaster monitoring equipment based on UAV vision and target collaboration, as described in an embodiment of this application.

[0023] The diagram is labeled as follows: 800 - Geological disaster monitoring equipment based on UAV vision and target collaboration; 801 - Processor; 802 - Memory; 803 - Multimedia component; 804 - I / O interface; 805 - Communication component. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0026] In the prior art, the solution most similar to this invention is a time-series photogrammetric monitoring method based on the collaboration of unmanned aerial vehicles (UAVs) and pre-set artificial targets. The hardware system of this method typically includes three key components: an automated UAV flight platform equipped with an RTK / PPK positioning module and a high-resolution visible light camera; an array of artificial targets regularly deployed within the monitoring area, made of weather-resistant materials and coated with a highly reflective coating; and a data processing and analysis terminal equipped with professional photogrammetric and point cloud processing software. This method typically involves four consecutive stages in its implementation: First, target deployment and absolute coordinate determination are performed, using a total station or GNSS RTK device to precisely measure the geodetic coordinates of each target center as a strict control point; then, periodic aerial survey data acquisition tasks are carried out, with UAVs periodically taking images of the target area along a pre-set high-overlap flight path, while accurately recording the positioning and attitude data at the time of each image acquisition; next, high-precision 3D reconstruction calculations are performed, including combining the measured target absolute coordinates to perform regional network adjustment (aerial triangulation) to construct an initial sparse point cloud, and applying a sub-pixel positioning algorithm to accurately extract the target center from each image. The pixel coordinates of the target images are used as strong constraints to optimize the adjustment model, which can significantly improve the positioning accuracy of the control points and their adjacent small areas. Finally, based on the optimized aerial triangulation results, a digital surface model (DSM) covering the entire area is generated through a dense matching algorithm. In the final stage, the surface displacement is calculated. The core is to use the first high-precision DSM (or the selected baseline model) as a spatial reference benchmark, and to accurately register the subsequent DSMs with it through a point cloud registration algorithm. The planar displacement and elevation displacement of each artificial target center are calculated respectively, and the spatial distribution map of the displacement field of the entire surface is calculated based on the overall point cloud difference obtained after registration.

[0027] While this closest approximation scheme has some application value under conditions of exposed terrain and stable environment, it suffers from significant technical bottlenecks and inherent defects, preventing it from effectively addressing the core technical concerns of this invention, particularly in achieving the sub-centimeter accuracy necessary for geological disaster early warning in complex environments. The primary problem lies in its over-reliance on artificial targets and its severely insufficient spatial expansion capability. In areas covered by shrubs, grasses, and other vegetation, natural surface textures are largely obscured or disturbed by movement, leading to severe data breaks and gaps in the DSM generated based on texture feature matching, resulting in a sharp deterioration in overall reconstruction accuracy. Although artificial targets themselves can achieve centimeter-level (e.g., approximately ±3cm planar accuracy) point measurement accuracy under ideal conditions, their accuracy advantage cannot be effectively transferred spatially to the surrounding vegetated natural surface areas. This results in low-accuracy displacement estimation in these critical areas due to reliance on low-quality DSMs, often exceeding 10cm, failing to meet requirements. Another prominent bottleneck is the method's weak resistance to environmental interference and poor operational stability. Under dynamic lighting conditions, the imaging quality and edge sharpness of the target in the image exhibit uncontrollable fluctuations. This imaging instability directly leads to drastic fluctuations in the core sub-pixel positioning accuracy, even resulting in positioning failures, thus introducing significant point extraction errors. More seriously, this error propagates and accumulates widely through the parameter transfer mechanism of the regional network adjustment process, ultimately manifesting as a significant increase in the error of the target center coordinates, causing unacceptably large deviations in the calculated planar displacement results. The third key deficiency lies in its lack of ability to monitor micro-cracks. Limited by the ground sampling interval resolution of a single image and the insufficient number and density of actual targets, this scheme struggles to effectively capture, identify, and quantify the highly localized, minute changes in displacement gradients occurring at the edges of small cracks (often only millimeters wide). Therefore, it completely lacks the ability to finely monitor and quantify crack initiation signs and initial slip signals. It is particularly important to emphasize that all these deficiencies ultimately lead to a fundamental problem: the actual accuracy achievable by this scheme is vastly different from the stringent requirements of early warning of geological disasters. Even in well-maintained, exposed areas where limited accuracy (planar displacement monitoring accuracy of approximately ±3cm) can be achieved, in critical areas that are often prone to geological disasters, such as densely vegetated areas with rugged terrain or complex and variable lighting, the monitoring accuracy of planar and elevation displacements is far below the sub-centimeter level (i.e., it cannot meet the basic requirements for early warning with planar displacement ≤2cm and elevation displacement ≤5cm), falling far short of the goal of reliably supporting accurate early warning decisions.

[0028] To implement the method of the present invention, this embodiment adopts the following device composition and physical basis:

[0029] Ground target array: A group of physical reference points deployed in areas prone to geological disasters, using anti-interference polarized targets, with a single point coverage area ≥30cm×30cm;

[0030] Unmanned aerial vehicle (UAV) data acquisition platform: a hexacopter equipped with a polarization imaging module (including an adjustable filter group) and an RTK positioning module (Real Time Kinematic). The flight altitude is 50-100m and the cruising speed is 2-5m / s.

[0031] Central processing terminal: A server equipped with an NVIDIA Quadro RTX 6000 GPU (graphics processor) that receives data collected by the drone via a wireless transmission module.

[0032] Among them, the anti-interference polarization target is an innovative design of this invention. It adopts a four-layer composite structure design and achieves dynamic noise suppression through precision mechanical adjustment. The composite structure design is as follows:

[0033] Weather-resistant engineering plastic base layer: provides impact-resistant support, with pre-drilled slots at the edges for fixing the adjustment mechanism; the protective frame assembly integrates an STM32 microcontroller (20mm×20mm) and a light sensor, which are assembled with the base layer through the slots; dimensions 30cm×30cm×0.5cm (length×width×thickness), surface roughness Ra≤3.2μm, temperature range 30℃-80℃, UV aging rate <5% / year (ASTM G154 standard);

[0034] Adjustable micro-polarizing film layer: ±15° rotation is achieved through a bottom micro stepper motor (15mm in diameter). The micro stepper motor is mounted on the back of the substrate layer. The stepper motor (output torque 0.25N·m) is connected to the polarizing film shaft through a coupling. The shaft is nested in the bearing groove of the substrate layer. The film thickness is linearly adjustable from 50-200μm (achieved by rotating the adjustment ring), with a polarization efficiency >99%, an extinction ratio of 1:1000, and an angle adjustment range of ±15° (manual or motor driven).

[0035] Subpixel positioning marker layer: a concentric circle-cross composite pattern chemically etched on a tunable micro-polarization film layer, with a concentric circle diameter of 10cm, a cross line width of 1.0mm±0.05mm, and an edge serration error of <5μm;

[0036] High diffuse reflection coating: BaSO4 (barium sulfate) coating, which is bonded to the polarizing film layer by UV curing adhesive; thickness 50μm, reflectivity 92%±2% (wavelength 450-900nm), diffuse reflection characteristics (Lambertian) ensure no color distortion in wide-angle imaging.

[0037] Working principle of anti-interference polarization target:

[0038] Strong light suppression principle: The micro-polarizing film is rotated to a 135° direction (orthogonal to the solar azimuth angle) under an illumination of >100,000 lux, filtering polarized stray light reflected from the water surface and rock surface;

[0039] Low-light enhancement principle: BaSO4 coating increases target brightness by 30%-50% on cloudy or rainy days (illuminance < 10,000 lux) (compared to ordinary paint targets);

[0040] Positioning reference principle: The cross-point coordinate fitting error is ≤0.1 pixels (based on a 20mm focal length lens), which meets the sub-pixel level measurement requirements.

[0041] Control process: The light sensor detects the ambient illuminance in real time → STM32 calculates the optimal polarization angle → drives the stepper motor to rotate to the target angle (0° or 135°), with a response delay of <0.1 seconds.

[0042] Example 1

[0043] See Figure 1 This embodiment provides a geological disaster monitoring method based on UAV vision and target collaboration, including steps S100, S200, S300, and S400.

[0044] S100. Acquire geological images at different times, and perform terrain modeling based on the geological images and the target locations therein to obtain a digital surface model.

[0045] Different time periods are divided into a baseline period (initial period) and a monitoring period (periodic monitoring of geological changes), obtaining two periods of DSM data (t0 is the time marker for the baseline period, t...). i (For use as a time marker during the monitoring period);

[0046] S110. Extract the region of interest from the image based on the geological image and target location;

[0047] Centered on the target's predicted position (the target's absolute coordinates measured during system initialization and the POS data collected by the UAV in real time, and the target's initial pixel coordinates in the image are automatically calculated based on spatial forward intersection and collinearity equations), a 100×100 pixel region of interest (ROI) is extracted.

[0048] The image of the region of interest is processed, and an edge-preserving filter is used to eliminate rain and fog interference:

[0049]

[0050] Ifilter: Outputs the filtered pixel value (dimensionless grayscale value);

[0051] i: Pixel row direction offset index (unit: pixels);

[0052] j: Pixel column offset index (unit: pixels);

[0053] W: Weight coefficient matrix (dimensionless, range [0,1]);

[0054] i,j: Subscripts of w (indicating the position of the weight in the matrix);

[0055] Iraw: The original input image matrix;

[0056] (i,j): The subscript of Iraw (identifying the coordinate position of the pixel in the image);

[0057] S120. Based on the spatial position and attitude parameters of geological images measured by satellite navigation and positioning, Zernike moment subpixel fitting is performed on the region of interest, the subpixel offset of the target center is calculated, and the world coordinates of the target center are obtained by combining the exterior orientation elements of the image through spatial resection transformation.

[0058] Zernike subpixel fitting:

[0059]

[0060] Δx: Subpixel offset in the x-direction (unit: pixels);

[0061] Δy: Subpixel offset in the y-direction (unit: pixels);

[0062] k: Zernike moment order index (k = 0, 1, 2, 3, 4);

[0063] Re: Operator for extracting the real part of a complex number;

[0064] Im: The operator for extracting the imaginary part of a complex number;

[0065] Z k1 Complex numerical values ​​of Zernike moments for the k-th order m=1 mode;

[0066] Z k0 Complex numerical values ​​of Zernike moments for the k-th order m=0 mode;

[0067] The calculated Δx and Δy are subpixel level corrections relative to the initial coordinates; the final precise pixel coordinates in the image are: x = x0 + Δx, y = y0 + Δy; (x0, y0) are the initial coordinates.

[0068] Satellite navigation and positioning measures the spatial position of the image center (geological image spatial position) at the moment of the drone's capture. Based on the spatial position of the geological image measured by satellite navigation and positioning and the attitude parameters of the drone, a rotation matrix R and translation vector T from the world coordinate system to the camera coordinate system are constructed.

[0069] Convert the coordinates in the image to coordinates in the world coordinate system:

[0070]

[0071] K: Camera intrinsic parameter matrix;

[0072] [RT]: Extrinsic parameter matrix;

[0073] R: Rotation matrix (3×3, rotation transformation from world coordinate system to camera coordinate system);

[0074] T: Translation vector (3×1, translation from the origin of the world coordinate system to the camera coordinate system);

[0075] x, y: Image plane coordinates (unit: pixels), obtained by Zernike moment subpixel fitting correction;

[0076] Zc: Depth value in camera coordinate system (unit: meters);

[0077] K: Camera intrinsic parameter matrix (3×3 matrix);

[0078] Xw, Yw, Zw: Coordinates of the target point in the world coordinate system (unit: meters);

[0079] This step allows you to obtain the world coordinates of the bullseye.

[0080] S130. Construct an initial geological point cloud based on geological images;

[0081] By combining data acquired from RGB cameras and LiDAR sensors, 3D reconstruction is performed through multi-view stereo matching to generate a complete point cloud of the monitored area.

[0082] S140, combined with image pose parameters from satellite navigation and positioning measurements, bullseye world coordinates, and LiDAR terrain data, the initial geological point cloud was optimized using bundle adjustment terrain constraints to construct a digital surface model;

[0083] LiDAR terrain data includes terrain curvature, and terrain regularization constraints are added based on the terrain curvature:

[0084]

[0085] X: Represents the parameter vector to be optimized, which includes the elevation adjustment of all target points;

[0086] A: Design matrix (coefficient matrix of the observation equation); A is a mathematical tool for inferring the position of the theoretical image point from the current world coordinate estimate (bullseye world coordinates).

[0087] L: Observation vector (containing image point coordinates); an image point is the two-dimensional projected coordinate of a point on the surface of the target object onto the image plane; L provides the measured value of the actual imaging position of the bullseye; A and L work together to calculate the optimal solution for the spatial position of the bullseye;

[0088] λ: Regularization weight coefficient (dimensionless);

[0089] Kcurv: Curvature weight matrix (a weight matrix calculated based on the curvature of the terrain);

[0090] ΔZw: Elevation adjustment vector (unit: meters);

[0091] The preceding term of the above formula For traditional adjustment residual constraints (ensuring control point matching accuracy); the latter term For terrain curvature constraints;

[0092] The optimized terrain elevation change (ΔZw) is consistent with the prior LiDAR terrain curvature distribution; the elevation abrupt change caused by matching failure in the vegetation area is eliminated (the error is compressed from >10cm to ≤5cm), resulting in a high-precision digital surface model;

[0093] S200: Based on geological images, identify vegetated and bare areas, construct a Delaunay triangulation based on vegetated area targets, and calculate vegetation displacement; calculate bare land displacement based on digital surface models.

[0094] S210. Use UNet++ semantic segmentation network to process geological images to obtain vegetation probability maps;

[0095] Geological images are RGB (red, green, and blue three-channel color model) images collected by drones. The UNet++ (nested densely connected U-shaped network structure) semantic segmentation network is used to process the images. This network adopts a nested densely connected U-shaped structure (the encoder uses 64 / 128 / 256 / 512 convolutional kernels with 2×2 max pooling downsampling, and the decoder uses 256 / 128 / 64 convolutional kernels with bilinear interpolation upsampling) to output a vegetation probability map.

[0096] S220. Divide the vegetation probability map based on a preset threshold to obtain vegetated areas and bare areas;

[0097] A binary mask is generated by setting a threshold of 0.7, where a value of 1 marks vegetated areas and a value of 0 marks bare areas, with the false positive rate strictly controlled to within 5%. This processing logic ensures that the mask is independent of the DSM data, directly utilizing the texture and color features of the optical image, and avoiding interference from missing point clouds in vegetated areas on mask generation.

[0098] S230. Construct a Delaunay triangulation based on the target points of the vegetation zone. Based on the target displacement of the vertices of the triangulation, perform displacement interpolation on any point within the triangulation to obtain the displacement of the vegetation zone.

[0099] For displacement calculation in bare areas, the displacement vector of the point cloud is obtained directly:

[0100]

[0101] D bare (x,y): Displacement vector of bare ground area (unit: meters);

[0102] Pti(x,y): Monitoring period t i Point cloud coordinates at any given time (unit: meters);

[0103] Pt0(x,y): Point cloud coordinates at time t0 (unit: meters);

[0104] For displacement calculation in vegetated areas, a Delaunay triangulation network is constructed using the target points (triangulation rule: maximum empty circle criterion); for any point u (spatial location point) within the triangular facet, the displacement interpolation is:

[0105]

[0106] Where, α k Coordinate weights, D target (k): Displacement of the target at the vertex of the triangle; k represents the k-th vertex of the triangle; D virtual (u) represents vegetation displacement;

[0107] S300: By fusing vegetation displacement and bare ground displacement using thin plate spline functions, global displacement information is obtained.

[0108] Thin-plate spline global fusion (TPS) achieves continuous representation of the global displacement field by uniformly processing the measured point cloud displacement in bare areas and the virtual displacement control points in vegetated areas, as detailed below:

[0109] S310. Using vegetation displacement and bare ground displacement as two types of control points respectively, construct a system of linear equations based on the two types of control points, interpolation conditions, and orthogonality conditions.

[0110] Interpolation condition: The function must pass through control points;

[0111] Orthogonality condition: The weight w must be orthogonal to the affine part to avoid redundancy;

[0112] Rewriting the above constraints in matrix form yields a system of linear equations:

[0113]

[0114] Φ: an n×n kernel matrix, where the elements φ jk =φ(‖u j -u k ||);u j u k These represent control points j and k, which originate from two types of control points;

[0115] P: n×3 polynomial term matrix ([1, x...) j ,y j ]);

[0116] w: n×1 weight vector (w1, w2, ..., w n );

[0117] a: 3×1 polynomial coefficient vector (a0, a1, a2);

[0118] d: n×1 target displacement observation vector (d(u1), d(u2), ..., d(u...)) n ));

[0119] 0: A 3×3 zero matrix (corresponding to orthogonality constraints);

[0120] S320. Determine the weights of the radial basis functions and the affine transformation parameters of the thin plate spline by inversely solving the linear equations through target displacement.

[0121] Transform the linear equation system to solve for a and the weight w:

[0122]

[0123] Using two types of control points as constraints, numerical methods (such as LU decomposition, SVD or least squares method) are used to solve the linear equation system to obtain a0, a1, a2 and weight w;

[0124] S330. Based on the radial basis functions, corresponding weights, and affine transformation parameters, a fusion function is constructed; the fusion function is:

[0125]

[0126] Among them, D total (u) represents the total displacement at point u (unit: millimeters); u j Here, n represents the target control point (center point), and n represents the number of target control points; w j For u j The weights; u is any point in the global domain;

[0127] S340. Solve for the global displacement value based on the fusion function to obtain the global displacement information.

[0128] The TPS function takes these two types of control points as input and utilizes the radial basis function φ(r) = r 2 The spatial interpolation properties of logr can fuse discrete bare ground displacement observations with virtual displacement constraint points in vegetated areas into a continuous smooth surface.

[0129] S400, based on a digital surface model, uses an improved Canny-Zernike operator for crack detection and a morphological thinning algorithm to identify crack skeleton lines;

[0130] S410. Use Canny to extract crack edge points;

[0131] S420. The Zernike moment is used to calculate the sub-pixel position of the crack edge point to obtain the binary edge map of the crack;

[0132] Calculate the sub-pixel position of the edge using orthogonal moments:

[0133]

[0134] Where θ: edge normal direction angle (unit: radians);

[0135] μ 20 μ 11 μ 02 The second central moment of an image (describing the pixel grayscale distribution);

[0136] Δ: Subpixel edge offset (unit: pixels);

[0137] Z31Re, Z11Re: Real parts of Zernike moments (used to calculate edge sub-pixel positions);

[0138] The original edge point position is corrected by sub-pixel edge offset Δ to obtain the accurate sub-pixel edge position, and the crack binary edge map is output (crack area is 1, background is 0);

[0139] S430. The binary edge map of the crack is processed by a morphological thinning algorithm to obtain the crack skeleton line.

[0140] The morphological thinning algorithm generates a single-pixel wide skeleton, which is a crucial preprocessing stage for crack gradient calculation. Its output is not a direct numerical parameter value, but rather structured intermediate data—the single-pixel wide skeleton line of the crack. This skeleton line serves as the basis for subsequent displacement gradient tensor calculations. In other words, it transforms the crack edge point set into a simplified topological structure (skeleton line), providing a spatial framework for subsequent gradient calculations.

[0141] S500: Based on the crack skeleton line and global displacement information, solve the Poisson equation to invert the displacement gradient of the crack region, perform displacement gradient tensor modeling, and construct the crack opening distribution map by projection along the skeleton line normal to obtain the monitoring results.

[0142] S510. Based on global displacement information, extract displacement field data at the crack edge and calculate the displacement gradient tensor.

[0143] After identifying the crack edge, the displacement field data of the crack edge is extracted based on the global displacement information, and the displacement gradient tensor is calculated:

[0144]

[0145] Where G is the displacement gradient tensor. Let be the first-order partial derivatives of the displacement field in the x and y directions;

[0146] S520. Calculate the strain field tensor based on the displacement gradient tensor at the crack edge, and construct the Poisson equation based on the strain field tensor.

[0147] The divergence is obtained by solving the displacement gradient tensor (summing over the main diagonal elements of the tensor). Estrain is the strain field tensor; this divergence serves as the source term of the Poisson equation, driving the inversion of the global displacement field d of the crack (the Laplace of the forced reconstruction of the displacement field d is equal to the strain divergence).

[0148] Poisson equation:

[0149]

[0150] Laplace operator (second-order differential operator), finite element mesh discretization solution;

[0151] d: Displacement field function (unit: meters), derived from global displacement information;

[0152] Divergence;

[0153] g: Boundary displacement condition (unit: meters)

[0154] Boundary curve (crack outline);

[0155] S530. Using the displacement at the crack edge as a boundary constraint, the displacement field function of the crack region is solved by the Poisson equation to obtain the continuous displacement field of the crack region.

[0156] Given the displacement at the crack edge, the displacement field function d is obtained by solving the Poisson equation using the finite element method. This allows us to obtain the displacement vector at any point in the crack region. By using boundary driving, we can achieve continuous and smooth reconstruction of the displacement field inside the crack.

[0157] S540. Project the continuous displacement field based on the normal of the crack skeleton line to obtain the crack normal opening amount, and construct the crack opening amount distribution map.

[0158] The crack skeleton line, also known as the centerline, provides a geometric reference (normal direction) for the crack's orientation. Based on the displacement vector of any point within the crack region, its displacement component along the normal direction is calculated.

[0159] δ N =N T G N ·l0

[0160] δ n Crack normal opening (unit: mm);

[0161] N T The transpose of the local normal unit vector (dimensionless) of the crack;

[0162] l0: Crack baseline length (unit: meters); Crack length at the baseline period (initial). If it is an irregular crack, the crack is divided into continuous short straight lines, and the length of the straight line segment where the point is located is taken here; if it is a short straight crack, the total length of the crack is taken here.

[0163] G N : Displacement gradient components along the crack normal (2×2 matrix);

[0164] The normal crack opening can be calculated using this formula, thereby constructing a crack opening distribution map (accuracy ±0.2mm) and marking the location of the maximum crack and its propagation direction.

[0165] The method of the present invention was tested, and the experimental area was set up as shown in Table 1:

[0166] Table 1 Experimental Area Setup

[0167] parameter Vegetated mixed slope area bare land control area slope 32°±2° (measured with a slope meter) 28°±1° vegetation coverage 65% ± 5% (UAV orthophoto interpretation) 0％ Target deployment 36 (spaced 20m apart) 36 (spaced 20m apart) Test crack Three artificially created cracks (1-5mm wide) Tongzuo Aerial survey parameters Height 50m, speed 3m / s, overlap 85% Tongzuo Environmental conditions Daytime dynamic light intensity (fluctuating between 100,000 and 120,000 lux) Tongzuo

[0168] The cracks in the experimental area were monitored using the method of this invention and a conventional method, and the results were compared as follows:

[0169] As shown in Table 2:

[0170] Crack number Actual opening Traditional method measurement values Measurement values ​​of this invention absolute error Test Crack No. 1 1.02 1.98±0.21 1.05±0.03 0.03 F2 Test Crack 2.15 3.67±0.34 2.18±0.05 0.03 Test Crack No. 3 4.89 6.12±0.47 4.92±0.08 0.03

[0171] It can be seen that the present invention has higher measurement accuracy compared with traditional technical solutions.

[0172] The core advantage of this invention stems from its systematic solution to persistent technical bottlenecks in geological disaster monitoring. Existing contact-based equipment relies on manual operation in high-risk environments and has sparse monitoring points. Furthermore, UAV aerial surveys in vegetated areas face accuracy collapse due to texture loss. This invention creatively combines physical noise suppression, intelligent fusion, and precise analysis: through a target structure design using an adjustable micro-polarization film layer (50-200μm) and a high-reflectivity coating, it achieves, for the first time, 0.1 pixel-level target center positioning accuracy under dynamic lighting conditions, completely solving the sub-pixel positioning failure problem caused by glare and low contrast—the measured planar displacement error under strong light is reduced from >3cm to ≤2cm; addressing the issue of vegetation obstruction... To address displacement field fracture, a Delaunay triangulation virtual displacement field was constructed based on target spatial topological constraints. This was combined with thin-plate spline functions to fuse measured point clouds from bare land, increasing vegetation coverage from <60% in traditional methods to 100%, while controlling the continuous error of the entire displacement field to within 2cm. For the millimeter-level accuracy requirement of micro-crack monitoring, a Poisson equation-driven displacement gradient tensor inversion model was innovatively adopted. By strongly constraining crack deformation boundaries through target displacement, the measurement error of crack opening was reduced from >1.5mm in traditional methods to the order of 0.2mm, breaking through the sub-millimeter analytical limit. These three technological breakthroughs form a closed-loop effect, ultimately maintaining a stable accuracy of ≤5cm elevation displacement error even under harsh environments such as heavy rain and seasonal transitions, enabling displacement monitoring of vegetated slopes to reach the practical standard for engineering early warning for the first time.

[0173] Example 2

[0174] A geological disaster monitoring system based on UAV vision and target coordination includes:

[0175] The first construction module is used to acquire geological images at different times, and to perform terrain modeling based on the geological images and the target locations in them to obtain a digital surface model.

[0176] The first calculation module is used to identify vegetated areas and bare areas based on geological images, construct a Delaunay triangulation based on vegetated area targets, and calculate vegetation displacement; it also calculates bare land displacement based on digital surface models.

[0177] The first fusion module is used to fuse vegetation displacement and bare ground displacement using thin plate spline functions to obtain global displacement information;

[0178] The first identification module is used to detect cracks based on a digital surface model, using an improved Canny-Zernike operator, and to identify crack skeleton lines using a morphological thinning algorithm.

[0179] The second module is used to solve the Poisson equation to invert the displacement gradient of the crack region based on the crack skeleton line and global displacement information, perform displacement gradient tensor modeling, and construct the crack opening distribution map by projection along the skeleton line normal to obtain the monitoring results.

[0180] As an optional implementation, the first building module includes:

[0181] The extraction unit is used to extract the region of interest from the image based on the geological image and the target location.

[0182] The fitting unit is used to perform Zernike moment subpixel fitting on the region of interest based on the spatial position and attitude parameters of the geological image measured by satellite navigation positioning. It calculates the subpixel offset of the target center and obtains the world coordinates of the target center by combining the exterior orientation elements of the image through spatial resection transformation.

[0183] The first building unit is used to construct an initial geological point cloud based on geological images;

[0184] The second building unit is used to combine the image pose parameters of satellite navigation and positioning measurements, the world coordinates of the target center, and LiDAR terrain data. The initial geological point cloud is optimized by using bundle adjustment terrain constraints to build a digital surface model.

[0185] As an optional implementation, the first computing module includes:

[0186] The first processing unit is used to process geological images using the UNet++ semantic segmentation network to obtain a vegetation probability map.

[0187] The second processing unit is used to divide the vegetation probability map based on a preset threshold to obtain vegetated areas and bare areas.

[0188] The first calculation unit is used to construct a Delaunay triangulation based on the target points of the vegetation area. Based on the target displacement of the vertices of the triangulation, displacement interpolation is performed on any point within the triangulation to obtain the displacement of the vegetation area.

[0189] As an optional implementation, the first fusion module includes:

[0190] The third building unit is used to construct two types of control points based on vegetation displacement and bare ground displacement, respectively, and to construct a linear equation system based on the two types of control points, interpolation conditions, and orthogonality conditions.

[0191] The second calculation unit is used to determine the weights of the radial basis functions and the affine transformation parameters of the thin plate spline by inversely solving the linear equations through target displacement;

[0192] The fourth building unit is used to construct the fusion function based on the radial basis functions, the corresponding weights, and the affine transformation parameters;

[0193] The fourth calculation unit is used to solve for the global displacement value based on the fusion function to obtain global displacement information.

[0194] Example 3

[0195] Corresponding to the above method embodiments, this embodiment also provides a geological disaster monitoring device based on UAV vision and target collaboration. The geological disaster monitoring device based on UAV vision and target collaboration described below can be referred to in correspondence with the geological disaster monitoring method based on UAV vision and target collaboration described above.

[0196] Figure 2 This is a block diagram illustrating a geological disaster monitoring device 800 based on UAV vision and target collaboration, according to an exemplary embodiment. Figure 2 As shown, the UAV-based vision and target-based geological disaster monitoring device 800 includes a processor 801 and a memory 802. The UAV-based vision and target-based geological disaster monitoring device 800 may also include one or more of the following: a multimedia component 803, an input / output (I / O) interface 804, and a communication component 805. The processor 801 controls the overall operation of the UAV-based vision and target-based geological disaster monitoring device 800 to complete all or part of the steps in the aforementioned UAV-based vision and target-based geological disaster monitoring method. The memory 802 stores various types of data to support the operation of the UAV-based vision and target-based geological disaster monitoring device 800. This data may include, for example, commands for any application or method operating on the UAV-based vision and target-based geological disaster monitoring device 800, and application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0197] Multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals.

[0198] The received audio signals can be further stored in memory 802 or transmitted via communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, buttons, etc. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the UAV-based vision and target-based geological disaster monitoring device 800 and other devices. Wireless communication methods include Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof; therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0199] Example 4

[0200] Corresponding to the above embodiment of the geological disaster monitoring method based on UAV vision and target collaboration, this embodiment also provides a readable storage medium. The readable storage medium described below can be referred to in correspondence with the geological disaster monitoring method based on UAV vision and target collaboration described above.

[0201] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described embodiment of the geological disaster monitoring method based on UAV vision and target collaboration.

[0202] Specifically, the readable storage medium can be a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or any other readable storage medium capable of storing program code.

[0203] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0204] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A geological disaster monitoring method based on UAV vision and target coordination, characterized in that, include: Geological images from different times are acquired, and terrain modeling is performed based on the geological images and the target locations in them to obtain a digital surface model. Based on geological imagery, vegetated and bare areas were identified, and a Delaunay triangulation was constructed using vegetation area targets to calculate vegetation displacement. Digital surface modeling calculates bare ground displacement; The displacement information of the entire region is obtained by fusing vegetation displacement and bare ground displacement using thin plate spline functions; Based on the digital surface model, an improved Canny-Zernike operator is used for crack detection, and a morphological thinning algorithm is used to identify crack skeleton lines. Based on the crack skeleton line and global displacement information, the Poisson equation is solved to invert the displacement gradient of the crack region. Displacement gradient tensor modeling is then performed, and a crack opening distribution map is constructed by projecting along the skeleton line normal. Monitoring results are obtained, including: Based on global displacement information, displacement field data at the crack edge is extracted, and the displacement gradient tensor is calculated. The strain field tensor is calculated based on the displacement gradient tensor at the crack edge, and the Poisson equation is constructed based on the strain field tensor. By using the displacement at the crack edge as a boundary constraint, the displacement field function of the crack region is solved using the Poisson equation to obtain the continuous displacement field of the crack region. By projecting the continuous displacement field onto the normal of the crack skeleton line, the crack normal opening is obtained, and a crack opening distribution map is constructed.

2. The geological disaster monitoring method based on UAV vision and target coordination according to claim 1, characterized in that, Based on geological images and the locations of targets within them, terrain modeling is performed to obtain a digital surface model, including: Extract regions of interest from geological images and target locations; Based on the spatial position and attitude parameters of geological images measured by satellite navigation and positioning, Zernike moment subpixel fitting is performed on the region of interest, the subpixel offset of the target center is calculated, and the world coordinates of the target center are obtained by spatial resection transformation in combination with the exterior orientation elements of the image. Construct an initial geological point cloud based on geological images; By combining the image pose parameters from satellite navigation and positioning measurements, the bullseye world coordinates, and LiDAR terrain data, the initial geological point cloud was optimized using bundle adjustment terrain constraints to construct a digital surface model.

3. The geological disaster monitoring method based on UAV vision and target coordination according to claim 1, characterized in that, Based on geological imagery, vegetated and bare areas are identified. A Delaunay triangulation network is constructed using vegetated area targets, and vegetation displacement is calculated. This includes: Geological images were processed using the UNet++ semantic segmentation network to obtain vegetation probability maps; The vegetation probability map is divided based on a preset threshold to obtain vegetated areas and bare areas; A Delaunay triangulation is constructed based on the target points of the vegetation zone. Based on the target displacement of the vertices of the triangulation, displacement interpolation is performed on any point within the triangulation to obtain the displacement of the vegetation zone.

4. The geological disaster monitoring method based on UAV vision and target coordination according to claim 1, characterized in that, By fusing vegetation displacement and bare ground displacement using thin-plate spline functions, global displacement information is obtained, including: Vegetation displacement and bare ground displacement are used as two types of control points, respectively. Based on the two types of control points, interpolation conditions and orthogonality conditions, a system of linear equations is constructed. The weights of the radial basis functions and the affine transformation parameters of the thin plate spline are determined by inverse solving of the linear equations based on the target displacement. A fusion function is constructed based on the radial basis functions, their corresponding weights, and the affine transformation parameters. Global displacement information is obtained by solving the global displacement value based on the fusion function.

5. The geological disaster monitoring method based on UAV vision and target coordination according to claim 1, characterized in that, Based on a digital surface model, an improved Canny-Zernike operator is used for crack detection, and a morphological thinning algorithm is employed to identify crack skeleton lines, including: Canny was used to extract crack edge points; The sub-pixel positions of crack edge points are calculated using Zernike moments to obtain a binary edge map of the crack; The crack skeleton line is obtained by processing the binary edge map of the crack using a morphological thinning algorithm.

6. A geological disaster monitoring system based on UAV vision and target coordination, characterized in that, include: The first construction module is used to acquire geological images at different times, and to perform terrain modeling based on the geological images and the target locations in them to obtain a digital surface model. The first calculation module is used to identify vegetated areas and bare areas based on geological images, construct a Delaunay triangulation based on vegetated area targets, and calculate vegetation displacement; it also calculates bare land displacement based on digital surface models. The first fusion module is used to fuse vegetation displacement and bare ground displacement using thin plate spline functions to obtain global displacement information; The first identification module is used to detect cracks based on a digital surface model, using an improved Canny-Zernike operator, and to identify crack skeleton lines using a morphological thinning algorithm. The second construction module is used to solve the Poisson equation to invert the displacement gradient of the crack region based on the crack skeleton line and global displacement information, perform displacement gradient tensor modeling, and construct a crack opening distribution map by projecting along the skeleton line normal to obtain monitoring results, including: Based on global displacement information, displacement field data at the crack edge is extracted, and the displacement gradient tensor is calculated. The strain field tensor is calculated based on the displacement gradient tensor at the crack edge, and the Poisson equation is constructed based on the strain field tensor. By using the displacement at the crack edge as a boundary constraint, the displacement field function of the crack region is solved using the Poisson equation to obtain the continuous displacement field of the crack region. By projecting the continuous displacement field onto the normal of the crack skeleton line, the crack normal opening is obtained, and a crack opening distribution map is constructed.

7. The geological disaster monitoring system based on UAV vision and target coordination according to claim 6, characterized in that, The first building module includes: The extraction unit is used to extract the region of interest from the image based on the geological image and the target location. The fitting unit is used to perform Zernike moment subpixel fitting on the region of interest based on the spatial position and attitude parameters of the geological image measured by satellite navigation positioning. It calculates the subpixel offset of the target center and obtains the world coordinates of the target center by combining the exterior orientation elements of the image through spatial resection transformation. The first building unit is used to construct an initial geological point cloud based on geological images; The second building unit is used to combine the image pose parameters of satellite navigation and positioning measurements, the world coordinates of the target center, and LiDAR terrain data. The initial geological point cloud is optimized by using bundle adjustment terrain constraints to build a digital surface model.

8. The geological disaster monitoring system based on UAV vision and target coordination according to claim 6, characterized in that, The first computing module includes: The first processing unit is used to process geological images using the UNet++ semantic segmentation network to obtain a vegetation probability map. The second processing unit is used to divide the vegetation probability map based on a preset threshold to obtain vegetated areas and bare areas. The first calculation unit is used to construct a Delaunay triangulation based on the target points of the vegetation area. Based on the target displacement of the vertices of the triangulation, displacement interpolation is performed on any point within the triangulation to obtain the displacement of the vegetation area.

9. The geological disaster monitoring system based on UAV vision and target coordination according to claim 6, characterized in that, The first fusion module includes: The third building unit is used to construct two types of control points based on vegetation displacement and bare ground displacement, respectively, and to construct a linear equation system based on the two types of control points, interpolation conditions, and orthogonality conditions. The second calculation unit is used to determine the weights of the radial basis functions and the affine transformation parameters of the thin plate spline by inversely solving the linear equations through target displacement; The fourth building unit is used to construct the fusion function based on the radial basis functions, the corresponding weights, and the affine transformation parameters; The fourth calculation unit is used to solve for the global displacement value based on the fusion function to obtain global displacement information.

Images