Unmanned aerial vehicle geological surveying and mapping method and system based on image analysis

Abstract

The invention relates to the technical field of geological survey surveying and mapping, and discloses an unmanned aerial vehicle geological survey surveying and mapping method and system based on image analysis, and the method comprises the steps: obtaining a geological body appearance original image shot by an unmanned aerial vehicle, and carrying out the denoising of the appearance original image, and obtaining a preprocessed image; extracting feature points between the preprocessed adjacent images based on a feature point matching algorithm and matching the feature points, splicing the images based on a feature point matching result, forming a complete investigation area large-size image at a splicing gap by adopting an image fusion technology, and aiming at the large-size image of the investigation area, carrying out image fusion on the large-size image of the investigation area; obtaining similar images collected by different cameras, and performing mosaic according to geographic coordinates to construct a multi-dimensional geological image data set; performing training identification on images in the multi-dimensional address image data set to construct a geological investigation identification model, and outputting a geological investigation surveying and mapping graph containing rock types, geological structures and mineral components through the geological investigation identification model; the working efficiency is improved, and the workload and risk of manual surveying are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of geological survey and mapping, and in particular to an unmanned aerial vehicle (UAV) geological survey and mapping method and system based on image analysis. Background Art

[0002] In traditional geological survey and mapping work, manual field surveys are often relied upon, and workers need to cross mountains and ridges and cross complex terrain. This is not only inefficient, but also difficult for personnel to reach some dangerous areas, such as cliffs, deep mountains and dense forests, resulting in incomplete data collection. At the same time, the information obtained by traditional surveying and mapping methods is limited, and it is difficult to conduct detailed geological analysis of large areas quickly and accurately. With the development of drone technology, its application in the field of geological survey has gradually emerged, but most of the existing drone geological survey methods remain at the simple image acquisition level, lacking a systematic and efficient image analysis process to deeply mine geological information, and cannot fully meet the urgent needs of modern geological research, preliminary surveys for engineering construction, and other aspects for high-precision and high-resolution geological data. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a UAV geological survey and mapping method and system based on image analysis.

[0004] A first aspect of the present invention provides a method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis, the method comprising the following steps:

[0005] Based on the scope of the survey area and the topographic features, the GIS system is used to formulate the UAV flight route, obtain the original images of the geological body taken by the UAV equipped with high-resolution optical cameras, multispectral cameras and infrared thermal imaging cameras, and perform denoising on the original images to obtain the pre-processed images;

[0006] Based on the feature point matching algorithm, feature points are extracted and matched between adjacent preprocessed images. The images are then spliced based on the feature point matching results. At the splicing gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset.

[0007] The images in the multi-dimensional address image dataset are trained and recognized to build a geological survey recognition model, which is used to output geological survey maps containing rock types, geological structures, and mineral compositions.

[0008] Optionally, in a first implementation of the first aspect of the present invention, the method of formulating a UAV flight route based on the scope of the survey area and the topographic features using a GIS system, and obtaining original images of the geological body appearance taken by a UAV equipped with a high-resolution optical camera, a multispectral camera, and an infrared thermal imaging camera, includes:

[0009] The survey area is divided into grids in the GIS system according to the geographic coordinate information, and the topographic features of the survey area are obtained using the DEM model, and the average altitude value and terrain slope value within each grid are extracted;

[0010] The greedy algorithm is used to optimize the UAV flight path. Starting from the UAV take-off point, at each decision node, the nearest unsurveyed key area or grid to the current position is selected as the next flight target. The flight route is gradually constructed until all areas are surveyed. The decision nodes are the grid boundary intersections or key area boundary points.

[0011] According to the optimal flight path, the original images of the geological body appearance are obtained by using a drone equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera.

[0012] Optionally, in a second implementation of the first aspect of the present invention, performing denoising on the original appearance image to obtain a preprocessed image includes:

[0013] Perform wavelet decomposition on the original image of the geological body appearance, and decompose the image signal into wavelet coefficients of different scales and directions:

[0014]

[0015] Among them, I(a,b) is the pixel value of the original image at coordinate (a,b), It is the conjugate function of the wavelet basis function in scale e, translation k, and direction d. is the corresponding wavelet coefficient;

[0016] After decomposition, the low-frequency information of the original image is concentrated in the large-scale wavelet coefficients, and the high-frequency information is distributed in the small-scale wavelet coefficients. The soft threshold method is used to remove the noise contained in the high-frequency wavelet coefficients, and the processed wavelet coefficients are used for wavelet reconstruction to obtain the pre-processed image.

[0017] when When , the new high-frequency wavelet coefficients are:

[0018]

[0019] Among them, T S is the threshold, The new coefficient is the positive or negative sign of the difference between the original high-frequency wavelet coefficient and the threshold multiplied by the threshold.

[0020] Optionally, in a third implementation of the first aspect of the present invention, extracting and matching feature points between adjacent preprocessed images based on a feature point matching algorithm includes:

[0021] A Gaussian difference pyramid is constructed for each adjacent image after preprocessing. In each layer of the Gaussian difference pyramid, each pixel is compared with its 8 adjacent pixels in the same layer and 9 adjacent pixels in the upper and lower layers to determine the potential feature points.

[0022] The potential feature points are located by fitting a three-dimensional quadratic function, Taylor expansion is performed on the three-dimensional quadratic function and the derivative is taken, the position offset of the potential feature points is solved by setting the derivative to 0, and the feature points at the edge of the image and in the low-contrast area are removed from the potential feature points using the Hessian matrix to obtain the first feature point;

[0023] With the first feature point as the center, calculate the gradient direction and amplitude of the pixels in the neighborhood of the first feature point, calculate the histogram of the gradient direction in the area centered on the first feature point, and take the direction corresponding to the peak of the histogram as the main direction of the feature point;

[0024] Taking the first feature point as the center, take a 16×16 neighborhood, rotate the feature point to the main direction, divide the neighborhood into 4×4 sub-regions, calculate the gradient histogram of 8 directions in each sub-region, and obtain a 128-dimensional feature descriptor;

[0025] Based on the feature points and their descriptors extracted from the two adjacent preprocessed images, the RANSAC algorithm is used to match the feature points to obtain the feature point matching results.

[0026] Optionally, in a fourth implementation of the first aspect of the present invention, the feature points and their descriptors extracted from the two adjacent preprocessed images are matched using a RANSAC algorithm to obtain a feature point matching result, including:

[0027] The nearest neighbor algorithm is used to perform preliminary matching on the feature points and their descriptors extracted from the two adjacent preprocessed images I1 and I2. The Euclidean distance between the descriptor of each feature point in image I1 and the descriptors of all feature points in image I2 is calculated, and the feature point with the smallest distance is taken as the matching point pair.

[0028] Randomly select 4 pairs of matching points, calculate the homography matrix, transform all matching points in image I1 according to the homography matrix, obtain their predicted positions in image I2, and use Euclidean distance to calculate the error between the predicted positions and the actual matching point positions;

[0029] When the statistical error is less than the set threshold of the number of matching point pairs, the matching point pair is the inlier point and the rest are the outliers. Repeat the iteration multiple times, and record the homography matrix with the largest number of inliers and the corresponding inlier point set after each iteration;

[0030] The inlier point set determined based on the homography matrix with the largest number of inliers is used as the optimized matching point pair to obtain the feature point matching results between adjacent images.

[0031] Optionally, in a fifth implementation of the first aspect of the present invention, the images are spliced based on the feature point matching results, and at the splicing gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multidimensional geological image dataset, including:

[0032] Assume that the spliced image is I, and for a pixel point p at the splicing gap, its neighborhood is N p , respectively calculate the two images I1 and I2 in the neighborhood N p The grayscale value and texture information within the neighborhood, where the texture information is described by calculating the gradient magnitude and direction within the neighborhood;

[0033] The weighting coefficient is determined based on the grayscale value difference and texture information difference, and all pixels at the splicing gap are weighted averaged to form a complete large-scale image of the survey area;

[0034] For large-scale images of the survey area, similar images collected by different cameras are obtained and mosaicked based on geographic coordinates. Let the coordinates of a certain pixel point in the geographic coordinate system be (X, Y). The images L collected by different cameras are c1 and L c2 The corresponding pixel coordinates are (x c1 ,y c1 ) and (x c2 ,y c2 );

[0035] A mapping relationship between geographic coordinates and image pixel coordinates is established, and images collected by different cameras are aligned and fused according to the geographic coordinates to construct a multi-dimensional geological image dataset.

[0036] Optionally, in a sixth implementation of the first aspect of the present invention, the training and recognition of images in the multi-dimensional address image dataset, the construction of a geological survey recognition model, and the output of a geological survey map containing predicted information on rock type, geological structure, and mineral composition through the geological survey recognition model include:

[0037] The image samples in the multidimensional geological image dataset are used as nodes of the graph, and the connection relationship between the nodes is determined based on the cosine similarity;

[0038] Using graph convolutional network as the basic architecture, for the node feature H of the lth layer (l) , and its propagation formula is:

[0039]

[0040] in, is the adjacency matrix after adding the self-loop, a is the original adjacency matrix, and I is the identity matrix; yes The degree matrix, W (l) is the learnable weight matrix of layer l, σ(·) is the activation function;

[0041] For each node i, calculate its attention coefficient α to neighbor node j ij , the feature representation of nodes i and j is obtained through linear transformation:

[0042] z i =W1h i ;

[0043] z j =W1h j ;

[0044] Among them, h i and h j are the feature vectors of nodes i and j respectively, W1 is the learnable weight matrix, z i is the feature representation of node i, z j is the feature representation of node j;

[0045] Use the attention coefficient to update the features of node i, integrate the attention mechanism into the propagation process of the graph convolutional network, and use the attention mechanism to update the node features in each layer of the graph convolutional network;

[0046] After the fusion of multi-layer graph convolutional network propagation and attention mechanism, the node features are input into the fully connected layer for prediction output;

[0047] According to the prediction output results, the rock type, geological structure, and mineral composition information are marked, and a geological survey map containing rock type, geological structure, and mineral composition information is generated.

[0048] The second aspect of the present invention provides an unmanned aerial vehicle geological survey and mapping system based on image analysis, wherein the unmanned aerial vehicle geological survey and mapping system based on image analysis includes an image acquisition module, a feature point matching module and a model recognition module, wherein:

[0049] The image acquisition module is used to formulate the UAV flight route based on the scope of the survey area and the topographic features using the GIS system, obtain the original images of the geological body taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera, and perform denoising on the original images to obtain the pre-processed images;

[0050] The feature point matching module is used to extract and match feature points between adjacent preprocessed images based on the feature point matching algorithm. The images are then stitched together based on the feature point matching results. At the stitching gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset.

[0051] The model recognition module is used to train and recognize images in the multi-dimensional address image dataset, build a geological survey recognition model, and output a geological survey map containing rock type, geological structure, and mineral composition through the geological survey recognition model.

[0052] Optionally, in a first implementation of the second aspect of the present invention, the image acquisition module includes an extraction submodule, a path optimization submodule and an acquisition submodule, wherein:

[0053] The extraction submodule is used to grid the survey area according to the geographic coordinate information in the GIS system, obtain the topographic features of the survey area using the DEM model, and extract the average altitude value and terrain slope value within each grid;

[0054] The path optimization submodule is used to optimize the UAV flight path using a greedy algorithm. Starting from the UAV's takeoff point, at each decision node, the nearest unsurveyed key area or grid to the current location is selected as the next flight target, and the flight route is gradually constructed until all areas are surveyed. The decision nodes are the intersection points of the grid boundaries or the boundary points of the key areas.

[0055] The acquisition submodule is used to obtain the original images of the geological body appearance taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera according to the optimal flight path.

[0056] Optionally, in a second implementation of the second aspect of the present invention, the feature point matching module includes a comparison submodule, a positioning submodule, a calculation submodule, a division submodule and a matching submodule, wherein:

[0057] The comparison submodule is used to construct a Gaussian difference pyramid for each of the two adjacent preprocessed images. In each layer of the Gaussian difference pyramid, each pixel is compared with its eight adjacent pixels in the same layer and its nine adjacent pixels in the upper and lower layers to determine potential feature points.

[0058] The positioning submodule is used to locate potential feature points by fitting a three-dimensional quadratic function, perform Taylor expansion on the three-dimensional quadratic function and take its derivative, set the derivative to 0 to solve the position offset of the potential feature points, and remove the feature points at the edge of the image and in the low-contrast area using the Hessian matrix to obtain the first feature point;

[0059] A calculation submodule is used to calculate the gradient direction and amplitude of the pixel points in the neighborhood of the first feature point with the first feature point as the center, and to generate a histogram of the gradient direction in the area centered on the first feature point, and to take the direction corresponding to the peak of the histogram as the main direction of the feature point;

[0060] The partitioning submodule is used to take a 16×16 neighborhood centered on the first feature point, rotate the feature point to the main direction, divide the neighborhood into 4×4 sub-regions, and calculate the gradient histogram in 8 directions in each sub-region to obtain a 128-dimensional feature descriptor;

[0061] The matching submodule is used to match the feature points based on the feature points and their descriptors extracted from the two adjacent images after preprocessing, and obtain the feature point matching results using the RANSAC algorithm.

[0062] In the technical solution provided by the present invention, the GIS system is used to formulate the flight route of the UAV according to the scope of the survey area and the topographic features, and the original image of the geological body taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera is obtained. The original image of the appearance is denoised to obtain a preprocessed image; based on the feature point matching algorithm, feature points are extracted and matched between adjacent preprocessed images, and the images are spliced based on the feature point matching results. At the splicing gaps, the image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area, and for the large-scale image of the survey area, the images collected by different cameras are obtained. Similar images are mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset; images in the multi-dimensional address image dataset are trained and recognized to construct a geological survey recognition model, and the geological survey recognition model is used to output geological survey maps containing rock types, geological structures, and mineral compositions; the UAV of the present invention achieves all-round coverage and fine imaging of complex terrain through intelligent dynamic flight planning, adopts multi-source image fusion acquisition technology to simultaneously acquire different types of data, improve data acquisition efficiency and richness, and the deep reinforcement learning model output enhances the depth and accuracy of geological feature recognition. The intelligent operation of the entire process significantly improves work efficiency, reduces the workload and risks of manual surveys, and optimizes the cost structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0064] Figure 1 A schematic diagram of a first embodiment of a UAV geological survey and mapping method based on image analysis provided by an embodiment of the present invention;

[0065] Figure 2 A schematic diagram of a second embodiment of the UAV geological survey and mapping method based on image analysis provided in an embodiment of the present invention;

[0066] Figure 3 A schematic diagram of a third embodiment of the UAV geological survey and mapping method based on image analysis provided by an embodiment of the present invention;

[0067] Figure 4 A schematic structural diagram of an image analysis-based UAV geological survey and mapping system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The terms "first," "second," "third," "fourth," and the like (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, apparatus, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0069] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of a first embodiment of a method for geological survey and mapping using a UAV based on image analysis provided by an embodiment of the present invention, wherein the method specifically comprises the following steps:

[0070] Step 101: Develop a UAV flight route using a GIS system based on the scope of the survey area and the topographic features, obtain original images of the geological body's appearance taken by a UAV equipped with a high-resolution optical camera, a multispectral camera, and an infrared thermal imaging camera, and perform denoising on the original images to obtain preprocessed images.

[0071] In this embodiment, wavelet decomposition is performed on the original image of the geological body appearance, and the image signal is decomposed into wavelet coefficients of different scales and directions:

[0072]

[0073] Among them, I(a,b) is the pixel value of the original image at coordinate (a,b), It is the conjugate function of the wavelet basis function in scale e, translation k, and direction d. is the corresponding wavelet coefficient;

[0074] After decomposition, the low-frequency information of the original image is concentrated in the large-scale wavelet coefficients, and the high-frequency information is distributed in the small-scale wavelet coefficients. The soft threshold method is used to remove the noise contained in the high-frequency wavelet coefficients, and the processed wavelet coefficients are used for wavelet reconstruction to obtain the pre-processed image.

[0075] when When , the new high-frequency wavelet coefficients are:

[0076]

[0077] Among them, T S is the threshold, The new coefficient is the positive or negative sign of the difference between the original high-frequency wavelet coefficient and the threshold multiplied by the threshold.

[0078] Step 102: Based on a feature point matching algorithm, feature points are extracted and matched between adjacent preprocessed images. Based on the feature point matching results, the images are spliced together. At the splicing gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset.

[0079] In this embodiment, let the spliced image be I, and for a certain pixel point p at the splicing gap, its neighborhood is N p , respectively calculate the two images I1 and I2 in the neighborhood N p The grayscale value and texture information within the neighborhood, where the texture information is described by calculating the gradient magnitude and direction within the neighborhood;

[0080] The weighting coefficient is determined based on the grayscale value difference and texture information difference, and all pixels at the splicing gap are weighted averaged to form a complete large-scale image of the survey area;

[0081] For large-scale images of the survey area, similar images collected by different cameras are obtained and mosaicked based on geographic coordinates. Let the coordinates of a certain pixel point in the geographic coordinate system be (X, Y). The images L collected by different cameras are c1 and L c2 The corresponding pixel coordinates are (x c1 ,y c1 ) and (x c2 ,y c2 );

[0082] A mapping relationship between geographic coordinates and image pixel coordinates is established, and images collected by different cameras are aligned and fused according to the geographic coordinates to construct a multi-dimensional geological image dataset.

[0083] Step 103: Perform training recognition on the images in the multi-dimensional address image dataset, build a geological survey recognition model, and output a geological survey map containing rock types, geological structures, and mineral compositions through the geological survey recognition model.

[0084] In this embodiment, the image samples in the multidimensional geological image dataset are used as nodes of the graph, and the connection relationship between the nodes is determined based on the cosine similarity;

[0085] Using graph convolutional network as the basic architecture, for the node feature H of the lth layer (l) , and its propagation formula is:

[0086]

[0087] in, is the adjacency matrix after adding the self-loop, A is the original adjacency matrix, and I is the identity matrix; yes The degree matrix, W (l) is the learnable weight matrix of layer l, σ(·) is the activation function;

[0088] For each node i, calculate its attention coefficient α to neighbor node j ij , the feature representation of nodes i and j is obtained through linear transformation:

[0089] z i =W1h i ;

[0090] z j =W1h j ;

[0091] Among them, h i and h j are the feature vectors of nodes i and j respectively, W1 is the learnable weight matrix, z i is the feature representation of node i, z j is the feature representation of node j;

[0092] Use the attention coefficient to update the features of node i, integrate the attention mechanism into the propagation process of the graph convolutional network, and use the attention mechanism to update the node features in each layer of the graph convolutional network;

[0093] After the fusion of multi-layer graph convolutional network propagation and attention mechanism, the node features are input into the fully connected layer for prediction output;

[0094] According to the prediction output results, the rock type, geological structure, and mineral composition information are marked, and a geological survey map containing rock type, geological structure, and mineral composition information is generated.

[0095] See also Figure 2 , a schematic diagram of a second embodiment of a UAV geological survey and mapping method based on image analysis provided by an embodiment of the present invention, the method comprising:

[0096] Step 201: Divide the survey area into grids in the GIS system according to geographic coordinate information, use the DEM model to obtain the topographic features of the survey area, and extract the average altitude value and terrain slope value within each grid;

[0097] Step 202: Optimize the flight path of the UAV using a greedy algorithm. Starting from the take-off point of the UAV, at each decision node, select the key area or grid closest to the current position that has not been surveyed as the next flight target. The flight path is gradually constructed until all areas are surveyed. The decision nodes are the intersection points of the grid boundaries or the boundary points of the key areas.

[0098] Step 203: Acquire original images of the geological body appearance taken by a UAV equipped with a high-resolution optical camera, a multispectral camera, and an infrared thermal imaging camera according to the optimal flight path.

[0099] See also Figure 3 , a schematic diagram of a third embodiment of a method for geological survey and mapping using a UAV based on image analysis provided by an embodiment of the present invention, the method comprising:

[0100] Step 301: construct a Gaussian difference pyramid for each of the two adjacent preprocessed images. In each layer of the Gaussian difference pyramid, compare each pixel with its eight adjacent pixels in the same layer and its nine adjacent pixels in the upper and lower layers to determine potential feature points.

[0101] Step 302: Position the potential feature points by fitting a three-dimensional quadratic function, perform Taylor expansion on the three-dimensional quadratic function and take its derivative, set the derivative to 0 to solve for the position offset of the potential feature points, and remove the feature points at the edge of the image and in low-contrast areas using the Hessian matrix to obtain the first feature point.

[0102] Step 303: Calculate the gradient direction and amplitude of the pixels in the neighborhood of the first feature point with the first feature point as the center, calculate the histogram of the gradient direction in the area centered on the first feature point, and take the direction corresponding to the peak of the histogram as the main direction of the feature point;

[0103] Step 304: Take a 16×16 neighborhood with the first feature point as the center, rotate the feature point to the main direction, divide the neighborhood into 4×4 sub-regions, calculate the gradient histogram in 8 directions in each sub-region, and obtain a 128-dimensional feature descriptor;

[0104] Step 305: Based on the feature points and their descriptors extracted from the two adjacent preprocessed images, the RANSAC algorithm is used to match the feature points to obtain a feature point matching result.

[0105] In this embodiment, the nearest neighbor algorithm is used to perform preliminary matching on the feature points and their descriptors extracted from the two adjacent preprocessed images I1 and I2. The Euclidean distance between the descriptor of each feature point in image I1 and the descriptors of all feature points in image I2 is calculated, and the feature point with the smallest distance is taken as the matching point pair.

[0106] Randomly select 4 pairs of matching points, calculate the homography matrix, transform all matching points in image I1 according to the homography matrix, obtain their predicted positions in image I2, and use Euclidean distance to calculate the error between the predicted positions and the actual matching point positions;

[0107] When the statistical error is less than the set threshold of the number of matching point pairs, the matching point pair is the inlier point and the rest are the outliers. Repeat the iteration multiple times, and record the homography matrix with the largest number of inliers and the corresponding inlier point set after each iteration;

[0108] The inlier point set determined based on the homography matrix with the largest number of inliers is used as the optimized matching point pair to obtain the feature point matching results between adjacent images.

[0109] See also Figure 4 , a structural diagram of a UAV geological survey and mapping system based on image analysis provided by an embodiment of the present invention, the system includes an image acquisition module, a feature point matching module and a model recognition module, wherein,

[0110] The image acquisition module is used to formulate the UAV flight route based on the scope of the survey area and the topographic features using the GIS system, obtain the original images of the geological body taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera, and perform denoising on the original images to obtain the pre-processed images;

[0111] The feature point matching module is used to extract and match feature points between adjacent preprocessed images based on the feature point matching algorithm. The images are then stitched together based on the feature point matching results. At the stitching gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset.

[0112] The model recognition module is used to train and recognize images in the multi-dimensional address image dataset, build a geological survey recognition model, and output a geological survey map containing rock type, geological structure, and mineral composition through the geological survey recognition model.

[0113] In this embodiment, the image acquisition module includes an extraction submodule, a path optimization submodule and an acquisition submodule, wherein:

[0114] The extraction submodule is used to grid the survey area according to the geographic coordinate information in the GIS system, obtain the topographic features of the survey area using the DEM model, and extract the average altitude value and terrain slope value within each grid;

[0115] The path optimization submodule is used to optimize the UAV flight path using a greedy algorithm. Starting from the UAV's takeoff point, at each decision node, the nearest unsurveyed key area or grid to the current location is selected as the next flight target, and the flight route is gradually constructed until all areas are surveyed. The decision nodes are the intersection points of the grid boundaries or the boundary points of the key areas.

[0116] The acquisition submodule is used to obtain the original images of the geological body appearance taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera according to the optimal flight path.

[0117] In this embodiment, the feature point matching module includes a comparison submodule, a positioning submodule, a calculation submodule, a division submodule and a matching submodule, wherein:

[0118] The comparison submodule is used to construct a Gaussian difference pyramid for each of the two adjacent preprocessed images. In each layer of the Gaussian difference pyramid, each pixel is compared with its eight adjacent pixels in the same layer and its nine adjacent pixels in the upper and lower layers to determine potential feature points.

[0119] The positioning submodule is used to locate potential feature points by fitting a three-dimensional quadratic function, perform Taylor expansion on the three-dimensional quadratic function and take its derivative, set the derivative to 0 to solve the position offset of the potential feature points, and remove the feature points at the edge of the image and in the low-contrast area using the Hessian matrix to obtain the first feature point;

[0120] A calculation submodule is used to calculate the gradient direction and amplitude of the pixel points in the neighborhood of the first feature point with the first feature point as the center, and to generate a histogram of the gradient direction in the area centered on the first feature point, and to take the direction corresponding to the peak of the histogram as the main direction of the feature point;

[0121] The partitioning submodule is used to take a 16×16 neighborhood centered on the first feature point, rotate the feature point to the main direction, divide the neighborhood into 4×4 sub-regions, and calculate the gradient histogram in 8 directions in each sub-region to obtain a 128-dimensional feature descriptor;

[0122] The matching submodule is used to match the feature points based on the feature points and their descriptors extracted from the two adjacent images after preprocessing, and obtain the feature point matching results using the RANSAC algorithm.

[0123] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis, characterized in that: The UAV geological survey and mapping method based on image analysis includes the following steps: Based on the scope of the survey area and the topographic features, the GIS system is used to formulate the UAV flight route, obtain the original images of the geological body taken by the UAV equipped with high-resolution optical cameras, multispectral cameras and infrared thermal imaging cameras, and perform denoising on the original images to obtain the pre-processed images; Based on the feature point matching algorithm, feature points are extracted and matched between adjacent preprocessed images. The images are then spliced based on the feature point matching results. At the splicing gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset. The images in the multi-dimensional address image dataset are trained and recognized to build a geological survey recognition model, which is used to output geological survey maps containing rock types, geological structures, and mineral compositions.

2. The method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis according to claim 1, wherein: The method of using a GIS system to formulate a UAV flight route based on the scope of the survey area and the topographic features, and obtaining original images of the geological body's appearance taken by a UAV equipped with a high-resolution optical camera, a multispectral camera, and an infrared thermal imaging camera, includes: The survey area is divided into grids in the GIS system according to the geographic coordinate information, and the topographic features of the survey area are obtained using the DEM model, and the average altitude value and terrain slope value within each grid are extracted; The greedy algorithm is used to optimize the UAV flight path. Starting from the UAV take-off point, at each decision node, the nearest unsurveyed key area or grid to the current position is selected as the next flight target. The flight route is gradually constructed until all areas are surveyed. The decision nodes are the grid boundary intersections or key area boundary points. According to the optimal flight path, the original images of the geological body appearance are obtained by using a drone equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera.

3. The method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis according to claim 1, wherein: The denoising process is performed on the original appearance image to obtain a pre-processed image, including: Perform wavelet decomposition on the original image of the geological body appearance, and decompose the image signal into wavelet coefficients of different scales and directions: Among them, I(a,b) is the pixel value of the original image at coordinate (a,b), It is the conjugate function of the wavelet basis function in scale e, translation k, and direction d. is the corresponding wavelet coefficient; After decomposition, the low-frequency information of the original image is concentrated in the large-scale wavelet coefficients, and the high-frequency information is distributed in the small-scale wavelet coefficients. The soft threshold method is used to remove the noise contained in the high-frequency wavelet coefficients, and the processed wavelet coefficients are used for wavelet reconstruction to obtain the pre-processed image. when When , the new high-frequency wavelet coefficients are: Among them, T S is the threshold, The new coefficient is the positive or negative sign of the difference between the original high-frequency wavelet coefficient and the threshold multiplied by the threshold.

4. The method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis according to claim 1, wherein: The feature point matching algorithm is based on extracting feature points between adjacent pre-processed images and matching them, including: A Gaussian difference pyramid is constructed for each adjacent image after preprocessing. In each layer of the Gaussian difference pyramid, each pixel is compared with its 8 adjacent pixels in the same layer and 9 adjacent pixels in the upper and lower layers to determine the potential feature points. The potential feature points are located by fitting a three-dimensional quadratic function, Taylor expansion is performed on the three-dimensional quadratic function and the derivative is taken, the position offset of the potential feature points is solved by setting the derivative to 0, and the feature points at the edge of the image and in the low-contrast area are removed from the potential feature points using the Hessian matrix to obtain the first feature point; With the first feature point as the center, calculate the gradient direction and amplitude of the pixels in the neighborhood of the first feature point, calculate the histogram of the gradient direction in the area centered on the first feature point, and take the direction corresponding to the peak of the histogram as the main direction of the feature point; Taking the first feature point as the center, take a 16×16 neighborhood, rotate the feature point to the main direction, divide the neighborhood into 4×4 sub-regions, calculate the gradient histogram of 8 directions in each sub-region, and obtain a 128-dimensional feature descriptor; Based on the feature points and their descriptors extracted from the two adjacent preprocessed images, the RANSAC algorithm is used to match the feature points to obtain the feature point matching results.

5. The method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis according to claim 4, wherein: The feature points and their descriptors extracted from the two adjacent images after preprocessing are matched using the RANSAC algorithm to obtain feature point matching results, including: The nearest neighbor algorithm is used to perform preliminary matching on the feature points and their descriptors extracted from the two adjacent preprocessed images I1 and I2. The Euclidean distance between the descriptor of each feature point in image I1 and the descriptors of all feature points in image I2 is calculated, and the feature point with the smallest distance is taken as the matching point pair. Randomly select 4 pairs of matching points, calculate the homography matrix, transform all matching points in image I1 according to the homography matrix, obtain their predicted positions in image I2, and use Euclidean distance to calculate the error between the predicted positions and the actual matching point positions; When the statistical error is less than the set threshold of the number of matching point pairs, the matching point pair is the inlier point and the rest are the outliers. Repeat the iteration multiple times, and record the homography matrix with the largest number of inliers and the corresponding inlier point set after each iteration; The inlier point set determined based on the homography matrix with the largest number of inliers is used as the optimized matching point pair to obtain the feature point matching results between adjacent images.

6. The method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis according to claim 1, wherein: The images are stitched together based on the feature point matching results. At the stitching gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset, including: Assume that the spliced image is I, and for a pixel point p at the splicing gap, its neighborhood is N p , respectively calculate the two images I1 and I2 in the neighborhood N p The grayscale value and texture information within the neighborhood, where the texture information is described by calculating the gradient magnitude and direction within the neighborhood; The weighting coefficient is determined based on the grayscale value difference and texture information difference, and all pixels at the splicing gap are weighted averaged to form a complete large-scale image of the survey area; For large-scale images of the survey area, similar images collected by different cameras are obtained and mosaicked based on geographic coordinates. Let the coordinates of a certain pixel point in the geographic coordinate system be (X, Y). The images L collected by different cameras are c1 and L c2 The corresponding pixel coordinates are (x c1 ,y c1 ) and (x c2 ,y c2 ); A mapping relationship between geographic coordinates and image pixel coordinates is established, and images collected by different cameras are aligned and fused according to the geographic coordinates to construct a multi-dimensional geological image dataset.

7. The method for geological survey and mapping using an unmanned aerial vehicle (UAV) based on image analysis according to claim 1, wherein: The training and recognition of images in the multi-dimensional address image dataset, the construction of a geological survey recognition model, and the output of a geological survey map containing predicted information on rock type, geological structure, and mineral composition through the geological survey recognition model include: The image samples in the multidimensional geological image dataset are used as nodes of the graph, and the connection relationship between the nodes is determined based on the cosine similarity; Using graph convolutional network as the basic architecture, for the node feature H of the lth layer (l) , and its propagation formula is: in, is the adjacency matrix after adding the self-loop, A is the original adjacency matrix, and I is the identity matrix; yes The degree matrix, W (l) is the learnable weight matrix of layer l, σ(·) is the activation function; For each node i, calculate its attention coefficient α to neighbor node j ij , the feature representation of nodes i and j is obtained through linear transformation: from i =W1h i ; from j =W1h j ; Among them, h i and h j are the feature vectors of nodes i and j respectively, W1 is the learnable weight matrix, z i is the feature representation of node i, z j is the feature representation of node j; Use the attention coefficient to update the features of node i, integrate the attention mechanism into the propagation process of the graph convolutional network, and use the attention mechanism to update the node features in each layer of the graph convolutional network; After the fusion of multi-layer graph convolutional network propagation and attention mechanism, the node features are input into the fully connected layer for prediction output; According to the prediction output results, the rock type, geological structure, and mineral composition information are marked, and a geological survey map containing rock type, geological structure, and mineral composition information is generated.

8. An unmanned aerial vehicle geological survey and mapping system based on image analysis, characterized in that: The UAV geological survey and mapping system based on image analysis includes an image acquisition module, a feature point matching module and a model recognition module, wherein: The image acquisition module is used to formulate the UAV flight route based on the scope of the survey area and the topographic features using the GIS system, obtain the original images of the geological body taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera, and perform denoising on the original images to obtain the pre-processed images; The feature point matching module is used to extract and match feature points between adjacent preprocessed images based on the feature point matching algorithm. The images are then stitched together based on the feature point matching results. At the stitching gaps, image fusion technology is used to perform weighted averaging based on the grayscale value and texture information of the pixel neighborhood to form a complete large-scale image of the survey area. For the large-scale image of the survey area, similar images collected by different cameras are obtained and mosaicked according to geographic coordinates to construct a multi-dimensional geological image dataset. The model recognition module is used to train and recognize images in the multi-dimensional address image dataset, build a geological survey recognition model, and output a geological survey map containing rock type, geological structure, and mineral composition through the geological survey recognition model.

9. The UAV geological survey and mapping system based on image analysis according to claim 8, characterized in that: The image acquisition module includes an extraction submodule, a path optimization submodule and an acquisition submodule, wherein: The extraction submodule is used to grid the survey area according to the geographic coordinate information in the GIS system, obtain the topographic features of the survey area using the DEM model, and extract the average altitude value and terrain slope value within each grid; The path optimization submodule is used to optimize the UAV flight path using a greedy algorithm. Starting from the UAV's takeoff point, at each decision node, the nearest unsurveyed key area or grid to the current location is selected as the next flight target, and the flight route is gradually constructed until all areas are surveyed. The decision nodes are the intersection points of the grid boundaries or the boundary points of the key areas. The acquisition submodule is used to obtain the original images of the geological body appearance taken by the UAV equipped with a high-resolution optical camera, a multispectral camera and an infrared thermal imaging camera according to the optimal flight path.

10. The UAV geological survey and mapping system based on image analysis according to claim 8, characterized in that: The feature point matching module includes a comparison submodule, a positioning submodule, a calculation submodule, a division submodule and a matching submodule, wherein: The comparison submodule is used to construct a Gaussian difference pyramid for each of the two adjacent preprocessed images. In each layer of the Gaussian difference pyramid, each pixel is compared with its eight adjacent pixels in the same layer and its nine adjacent pixels in the upper and lower layers to determine potential feature points. The positioning submodule is used to locate potential feature points by fitting a three-dimensional quadratic function, perform Taylor expansion on the three-dimensional quadratic function and take its derivative, set the derivative to 0 to solve the position offset of the potential feature points, and remove the feature points at the edge of the image and in the low-contrast area using the Hessian matrix to obtain the first feature point; A calculation submodule is used to calculate the gradient direction and amplitude of the pixel points in the neighborhood of the first feature point with the first feature point as the center, and to generate a histogram of the gradient direction in the area centered on the first feature point, and to take the direction corresponding to the peak of the histogram as the main direction of the feature point; The partitioning submodule is used to take a 16×16 neighborhood centered on the first feature point, rotate the feature point to the main direction, divide the neighborhood into 4×4 sub-regions, and calculate the gradient histogram in 8 directions in each sub-region to obtain a 128-dimensional feature descriptor; The matching submodule is used to match the feature points based on the feature points and their descriptors extracted from the two adjacent images after preprocessing, and obtain the feature point matching results using the RANSAC algorithm.

Images