Mine slope monitoring methods based on drone inspection

Abstract

This invention relates to the field of slope monitoring technology, and particularly to a method for monitoring mine slopes based on UAV (Unmanned Aerial Vehicle) inspection. The method includes: acquiring and preprocessing slope images; introducing a perturbation excitation dilation mechanism; performing a spatiotemporal nonlinear transformation on the preprocessed slope images; calculating the perturbation excitation response; and obtaining the dilated slope image. Based on the dilated slope image, a residual energy diffusion modeling mechanism is introduced to construct a residual energy tensor. Based on the residual energy tensor, an entropy density splitting mechanism is introduced to construct a stacked spatial projection function to obtain the nonlinear landslide information response intensity. Based on the nonlinear landslide information response intensity, a stacked discriminator is constructed to classify landslide risk and obtain the monitoring results. This method solves the problems of inaccurate processing of UAV-acquired image data in traditional mine slope monitoring methods, leading to low accuracy in locating and judging landslide risk areas, as well as insufficient adaptability and sensitivity.

Description

Technical Field

[0001] This invention relates to the field of slope monitoring technology, and in particular to a method for monitoring mine slopes based on unmanned aerial vehicle (UAV) inspection. Background Technology

[0002] With the continuous exploitation of mining resources, slope engineering is widely used in open-pit mines, steep slopes, and spoil heaps. Its stability directly affects personnel safety and the continuity of mining operations. However, due to various factors such as loose geological structures, poor surface drainage, blasting disturbances, and rainwater seepage, mine slopes are highly susceptible to landslides and collapses. Currently, the mainstream monitoring methods for addressing these hazards include manual inspections, displacement sensor measurements, GNSS measurements, laser scanning, and photogrammetry. While these methods can achieve localized or periodic monitoring to some extent, they suffer from limitations such as poor real-time performance, limited spatial coverage, and difficulty in capturing early, subtle deformation signs. Especially in high-risk areas or complex terrain environments, they cannot effectively support the rapid detection and intelligent identification of landslide precursors across the entire area.

[0003] In recent years, with the maturity of UAV platforms and the development of image processing technology, using UAVs to acquire high-resolution images and perform landslide identification has become a research hotspot. Most related studies are based on traditional image classification, feature point matching, and change detection methods, or combine convolutional neural networks from deep learning for image recognition modeling. While these methods have made some progress in stable boundary identification and large-scale landslide area detection, they still suffer from insufficient sensitivity, high false detection rates, and a lack of physical interpretation capabilities when facing challenges such as identifying minute geological deformations in multi-temporal image sequences, feature extraction under strong interference environments, and continuous risk assessment against complex texture backgrounds. Furthermore, traditional deep learning methods generally rely on large-scale labeled sample training, making them poorly adaptable to the characteristics of geological disaster problems, such as the difficulty in obtaining samples and the ambiguity of category boundaries. Moreover, most methods directly apply general models in their structural design, lacking in-depth modeling of landslide induction mechanisms and the spatiotemporal evolution of image features.

[0004] In summary, traditional mine slope monitoring methods still suffer from technical problems such as inaccurate processing of image data acquired by UAVs, resulting in low accuracy in locating and judging landslide risk areas, as well as insufficient adaptability and sensitivity. Summary of the Invention

[0005] This invention provides a mine slope monitoring method based on UAV inspection, which solves the technical problems of inaccurate processing of image data acquired by UAVs in traditional mine slope monitoring methods, resulting in low accuracy in locating and judging landslide risk areas, as well as insufficient adaptability and sensitivity.

[0006] The mine slope monitoring method based on unmanned aerial vehicle (UAV) inspection of the present invention specifically includes the following technical solutions:

[0007] The method for monitoring mine slopes based on drone inspection includes the following steps:

[0008] S1. Acquire slope images and preprocess them to obtain preprocessed slope images; introduce a perturbation excitation dilation mechanism to perform spatiotemporal nonlinear transformation on the preprocessed slope images, calculate the perturbation excitation response, and obtain the dilated slope images.

[0009] S2. Based on the slope image after dilation transformation, a residual energy diffusion modeling mechanism is introduced to obtain local residual energy and construct a residual energy tensor. Based on the residual energy tensor, an entropy density splitting mechanism is introduced to construct a stacked spatial projection function to obtain the nonlinear landslide information response intensity. Based on the nonlinear landslide information response intensity, a stacked discriminator is constructed to determine the landslide risk category and obtain the monitoring results.

[0010] Preferably, S1 specifically includes:

[0011] In the implementation of the perturbation excitation dilation mechanism, based on the preprocessed slope image, the rate of change of pixels in the time direction and the second-order spatial gradient in the diagonal direction are calculated. Combined with trigonometric functions and logarithmic functions, a joint space-time transformation is performed to obtain the perturbation excitation response. Based on the perturbation excitation response, the preprocessed slope image is dilated to obtain the dilated slope image.

[0012] Preferably, S2 specifically includes:

[0013] In the implementation of the residual energy diffusion modeling mechanism, the gradient vector, structural complexity index, and gradient vector of the perturbation potential field of the slope image after dilation transformation are calculated, and the local residual energy is calculated by combining the radial perturbation intensity in the main texture direction, and the residual energy tensor is constructed.

[0014] Preferably, S2 specifically includes:

[0015] The structural complexity index is obtained by introducing a local window and weighting and fusing the local standard deviation, information entropy, and directional gradient divergence of the slope image after dilation transformation.

[0016] Preferably, S2 specifically includes:

[0017] Based on the preprocessed slope image and the disturbance excitation response, the disturbance potential field is obtained, and the gradient is calculated to obtain the gradient vector of the disturbance potential field.

[0018] Preferably, S2 specifically includes:

[0019] The radial perturbation intensity along the main texture direction is calculated based on the perturbation excitation response and the main texture direction of the slope image after dilation transformation.

[0020] Preferably, S2 specifically includes:

[0021] In the process of implementing the entropy density splitting mechanism, a normalization adjustment term for the information entropy response is introduced based on the residual energy tensor, and a stacked spatial projection function is constructed in combination with the local information entropy density to obtain the nonlinear landslide information response intensity.

[0022] Preferably, S2 specifically includes:

[0023] The stacked discriminator is a multi-layered nested nonlinear combination model, which is constructed based on the nonlinear landslide information response intensity and by introducing a weight matrix and a bias term.

[0024] The beneficial effects of the technical solution of the present invention are:

[0025] 1. By introducing a perturbation-excitation expansion mechanism and performing a joint space-time transformation, the response signal of the geological deformation area is stretched in the preprocessed slope image, and the non-slip region is compressed, thereby improving the recognizability of subtle geomorphic evolution signals and effectively solving the problems of weak response capability and indistinct features in the traditional method for progressive displacement regions.

[0026] 2. By using the residual energy diffusion modeling mechanism, combined with the square of the inner product between the gradient vector of the slope image after dilation transformation and the gradient vector of the perturbation potential energy field, as well as the radial perturbation intensity and energy flow regulator in the main texture direction, the multi-physical variables of potential landslide-induced structures in the image are jointly modeled. It has high-dimensional coupling expression capability and effectively enhances the distinguishability of landslide risk areas and the accuracy of pattern recognition.

[0027] 3. By constructing a stacked spatial projection function, the square of the local information entropy density is combined with the normalization adjustment term of the information entropy response, and the hyperbolic tangent activation function is used to adjust the boundary response. This enables the slope image to maintain strong risk sensitivity in areas with high uncertainty, such as structural boundaries, crack areas, and mesh structure areas, thereby improving the ability to identify fuzzy structural boundaries. Attached Figure Description

[0028] Figure 1 This is a flowchart of the mine slope monitoring method based on UAV inspection as described in this invention. Detailed Implementation

[0029] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] The specific scheme of the mine slope monitoring method based on UAV inspection provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0032] See attached document Figure 1 The diagram illustrates a flowchart of a mine slope monitoring method based on unmanned aerial vehicle (UAV) inspection, according to an embodiment of the present invention. The method includes the following steps:

[0033] S1. Acquire slope images and preprocess them to obtain preprocessed slope images; introduce a perturbation excitation dilation mechanism to perform spatiotemporal nonlinear transformation on the preprocessed slope images, calculate the perturbation excitation response, and obtain the dilated slope images.

[0034] The drone inspects the mine slope according to a fixed route, attitude, and time period preset by the technicians, collecting slope images. The slope images are then preprocessed to obtain preprocessed slope images. The preprocessing process includes using existing technologies such as IMU-camera attitude synchronization calibration algorithms for heading control, using existing technologies such as geodetic coordinate projection algorithms based on RPC (Rational Polynomial Coefficients) models for orthorectification, and using existing technologies such as image feature point matching and registration algorithms for georegistration. The preprocessed slope image sequence is denoted as... ,in This indicates the length of the time series data acquisition for the slope images. This represents the two-dimensional pixel coordinates of the preprocessed slope image. Indicates at time Preprocessed slope image, This indicates the pixel position of the preprocessed slope image. The pixel values ​​at the location; the above preprocessing process can ensure that all slope images are under a unified spatial reference system. The methods used in the preprocessing process are all technical means well known to those skilled in the art, and will not be described in detail here;

[0035] Furthermore, a spatiotemporal nonlinear transformation is performed on the preprocessed slope image, which significantly enhances the response of potential landslide change areas while suppressing spurious change signals generated by non-geological disturbances. During the spatiotemporal nonlinear transformation, a perturbation excitation dilation mechanism is introduced to construct a joint space-time transformation. This allows pixels in the preprocessed slope image to evolve with time, generating a non-uniform transformation response. This stretches the response signal of geological deformation areas and compresses non-slip regions in the preprocessed slope image.

[0036] The perturbation excitation expansion mechanism is implemented through the following formula:

[0037] ,

[0038] in, This represents the response to the disturbance stimulus, i.e., at time t. In pixels Nonlinear perturbation enhancement factor at the location; This is the disturbance intensity amplification factor, used to adjust the overall amplitude of the disturbance excitation response. The reference value range is... The energy density of the local structure is determined by estimation of the local structure energy density, which is a well-known technique in the art and will not be described in detail here. This is a disturbance response frequency adjustment factor used to control the sinusoidal response frequency; the reference value range is... The result was obtained through Fourier transform spectrum analysis, a technique well-known to those skilled in the art, and will not be elaborated upon here. At any moment Preprocessed slope image at pixel location Pixel value at; It is the rate of change of a pixel in the time direction, representing the intensity of brightness change of an image pixel between consecutive frames. It is calculated by inter-frame difference, which is a well-known technique in the art and will not be elaborated here. It is the second-order spatial gradient of a pixel in the diagonal direction, which can reflect the edge intersection area. It is calculated by the second-order differential operator of the image, which is a well-known technique in the art and will not be described in detail here. This means that the brightness changes of image pixels in the time direction are periodically mapped, which can periodically amplify the time-varying signal and suppress the abrupt signal, which is beneficial for detecting areas with continuous and gradual changes in terrain. It represents the enhancement of cross-texture variations in the spatial direction of the image, which can highlight risk structure patterns as "mesh", "oblique crack" and "radial", and is especially suitable for identifying crack networks before slope instability; This is used to integrate change detection in the time domain with structure detection in the spatial domain to generate a composite risk response mechanism.

[0039] The above process, through the combined use of trigonometric and logarithmic functions, enables it to respond to changes at different scales and frequencies, thereby achieving nonlinear excitation of image sequences.

[0040] Furthermore, based on the aforementioned perturbation excitation response, a dilation transformation is performed on the preprocessed slope image to obtain a dilated slope image. The dilated slope image at any pixel location... The expression at the location is as follows:

[0041] ,

[0042] in, This indicates the pixel position of the slope image after dilation transformation. The pixel values ​​at the location; the slope image after dilation transformation shows an amplitude enhancement effect in the landslide feature area, which can provide clearer differences for subsequent modeling.

[0043] S2. Based on the slope image after dilation transformation, a residual energy diffusion modeling mechanism is introduced to obtain local residual energy and construct a residual energy tensor. Based on the residual energy tensor, an entropy density splitting mechanism is introduced to construct a stacked spatial projection function to obtain the nonlinear landslide information response intensity. Based on the nonlinear landslide information response intensity, a stacked discriminator is constructed to determine the landslide risk category and obtain the monitoring results.

[0044] To further describe the change behavior of slope images after dilation transformation in the spatial domain, a residual energy diffusion modeling mechanism is introduced. The technical objective is to construct a residual energy tensor coupled with multiple physical quantities at the image pixel level to reflect the subtle expression of geological structure changes in the dilation-transformed slope image. This residual energy diffusion modeling mechanism, based on multiple features such as image gradient, structural complexity, and perturbation potential, constructs local residual energy at spatial points, specifically defined as follows:

[0045] ,

[0046] in, It is a pixel. At any moment The local residual energy density response value under disturbance-driven conditions, i.e., the local residual energy; It is the gradient vector of the slope image after dilation transformation, representing the gradient at each pixel. The rate of change of the spatial first derivative at a given point can reflect the edge intensity of the image. It is calculated using the Sobel, Scharr, or Prewitt operators. The calculation methods are well-known to those skilled in the art and will not be elaborated here. It is a pixel. At any moment The gradient vector of the perturbation potential field is used to describe the changing trend of the direction and amplitude of the potential response to the perturbation excitation response in response to the preprocessed slope image. It is a pixel. At any moment The perturbation potential field, ; It is the square of the inner product of the gradient vector of the slope image after dilation transformation and the gradient vector of the perturbation potential field, used to represent a measure of the consistency between the image gradient and the perturbation direction. It is the slope image after dilation transformation at the pixel level. The dominant texture direction at the location represents the dominant angle of the edge direction. It is calculated using existing methods based on structure tensor analysis or Gabor filter groups. The calculation method is a well-known technique in the art and will not be elaborated here. The perturbation excitation response is in the radial direction (i.e., the main texture direction). The gradient value of ) represents the propagation ability or concentration of the disturbance in that direction. It is obtained based on the main texture direction using the existing directional derivative kernel method. The calculation method is a well-known technique in the art and will not be elaborated here. It is the radial perturbation intensity in the main texture direction, used to control the degree of intrusion of the perturbation into the main texture direction; This is a structural complexity index of the dilated slope image within a local window (the size of the local window is determined according to specific needs, such as 5×5 or 7×7). It is obtained by calculating the local standard deviation, information entropy, and directional gradient divergence of the dilated slope image and then performing weighted fusion. The reference value range is... The methods for calculating local standard deviation, information entropy, and directional gradient divergence are well-known to those skilled in the art and will not be elaborated here. It is the slope image after dilation transformation at the pixel level. The second derivative of the Laplacian at point is used to characterize the overall curvature of the image region in the slope image after dilation transformation. It is obtained through the Laplacian operator, and the calculation method is a well-known technique to those skilled in the art, which will not be elaborated here; exponential term As an energy flow regulator, it is used to regulate local residual energy;

[0047] Based on local residual energy A residual energy tensor is constructed. Furthermore, since the residual energy tensor has high dimensionality and strong internal heterogeneity, it cannot be directly used for classification. Therefore, an entropy density splitting mechanism is introduced. This entropy density splitting mechanism first obtains the information entropy at the pixel level based on the Shannon information entropy formula. At the moment The uncertainty of classification (information complexity) , that is, the local information entropy density; the Shannon information entropy formula is a technical means well known to those skilled in the art, and will not be elaborated here;

[0048] Furthermore, based on information entropy theory and combined with the physical modeling concept of disturbance response, the local information entropy density and residual energy tensor are jointly modeled to construct a stacked spatial projection function, thereby obtaining the nonlinear landslide information response intensity. The specific formula is as follows:

[0049] ,

[0050] in, Indicates at time pixel The nonlinear landslide information response intensity at a point is used to characterize the likelihood of a landslide or deformation occurring at that point. The closer the value of the nonlinear landslide information response intensity is to 1, the more likely that the point is to be a potential landslide risk point. It is the normalization adjustment term of the information entropy response, used to balance the information expression under different degrees of disturbance; It is a hyperbolic tangent activation function, used as a normalization and boundary reinforcement mechanism to ensure the intensity of the nonlinear landslide information response. It has a suppressive effect on larger disturbances and a sensitive response to medium and low disturbances;

[0051] Finally, based on the nonlinear landslide information response intensity, a stacked discriminator is constructed. The stacked discriminator structure is a multi-layered nested nonlinear combination model, defined as follows:

[0052] ,

[0053] in, At any moment pixel The results of landslide risk classification at the location, i.e., monitoring results; This represents the total number of classification labels for landslide risk, predefined by professional technicians based on actual circumstances. It is a classification label index of landslide risk, such as This indicates that the slope shows no obvious signs of sliding and is within the normal range; This indicates that there are initial signs of displacement in the area, which is considered a minor risk. This indicates that the regional structure has become loose or cracks have formed, which is considered a medium risk. This indicates that the area has experienced landform deformation or severe damage to the slope structure, and is considered high-risk. This is the number of discriminator stacking layers, i.e. the number of sub-classifiers. It is determined according to the application requirements and is not limited here. Values ​​such as 5 or 7 are acceptable. It is the first The contribution of each subclassifier to the overall decision is learned through backpropagation, with a reference value range of [value missing]. And the sum is 1. The back propagation mechanism is a technical means well known to those skilled in the art, and will not be described in detail here. It is a non-linear activation function, such as the Sigmoid function; It is the first Subclassifiers are used to distinguish categories. The weight parameter matrix is ​​obtained by optimization using the gradient descent method; It is the first Subclassifiers are used to distinguish categories. The bias term can shift and adjust the output of each sub-classifier, and is obtained by optimization through gradient descent; the gradient descent method is a well-known technique in the art and will not be described in detail here.

[0054] In summary, a method for monitoring mine slopes based on unmanned aerial vehicle (UAV) inspection has been developed.

[0055] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0056] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0057] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for monitoring mine slopes based on unmanned aerial vehicle (UAV) inspection, characterized in that, Includes the following steps: S1. Acquire slope images and preprocess them to obtain preprocessed slope images; introduce a perturbation excitation dilation mechanism, calculate the rate of change of pixels in the time direction and the second-order spatial gradient in the diagonal direction based on the preprocessed slope images, and perform a space-time joint transformation by combining trigonometric functions and logarithmic functions to obtain the perturbation excitation response; based on the perturbation excitation response, perform a dilation transformation on the preprocessed slope images to obtain dilated slope images. S2. A residual energy diffusion modeling mechanism is introduced. By calculating the gradient vector, structural complexity index, and gradient vector of the perturbation potential field of the slope image after dilation transformation, and combining the radial perturbation intensity in the main texture direction, the local residual energy is calculated, and a residual energy tensor is constructed. The structural complexity index is obtained by introducing a local window and weighting and fusing the local standard deviation, information entropy, and directional gradient divergence of the slope image after dilation transformation. The gradient vector of the perturbation potential field is obtained by calculating the gradient of the perturbation potential field constructed based on the preprocessed slope image and the perturbation excitation response. An entropy density splitting mechanism is introduced. Based on the residual energy tensor, a normalization adjustment term of the information entropy response is introduced, and combined with the local information entropy density, a stacked spatial projection function is constructed to obtain the nonlinear landslide information response intensity. Based on the nonlinear landslide information response intensity, a stacked discriminator is constructed to distinguish the landslide risk category and obtain the monitoring results.

2. The mine slope monitoring method based on UAV inspection according to claim 1, characterized in that, S2 specifically includes: The radial perturbation intensity along the main texture direction is calculated based on the perturbation excitation response and the main texture direction of the slope image after dilation transformation.

3. The mine slope monitoring method based on UAV inspection according to claim 1, characterized in that, S2 specifically includes: The stacked discriminator is a multi-layered nested nonlinear combination model, which is constructed based on the nonlinear landslide information response intensity and by introducing a weight matrix and a bias term.

Images