An intelligent UAV mapping system based on remote sensing technology

Abstract

The present invention relates to the field of surveying and mapping geographic information technology. Specifically, it relates to an intelligent unmanned aerial vehicle surveying and mapping system based on remote sensing technology, which includes a remote sensing image acquisition module, an image correction and stitching module, an orthorectification module, a ground object recognition module, an attribute information input module, a map production module, and a management database. This system obtains surface remote sensing images, extracts descriptors of each surface remote sensing image to generate matching pairs, determines the corresponding relationships of the same ground objects between different images, provides a basis for image stitching, trains a correction model to convert the ground control point coordinates into image coordinates, creates a basic attribute table through the basic attributes corresponding to the types of each ground object, improves the information content and practicality of the map, combines and interpolates to calculate the elevation of each interpolation point in the surveying and mapping area to generate a complete map, enabling the map to not only present the distribution of ground objects but also reflect the terrain undulation, thus forming a map with rich content.

Description

Technical Field

[0001] The present invention relates to the field of surveying and mapping geographic information technology. Specifically, it is an intelligent unmanned aerial vehicle (UAV) surveying and mapping system based on remote sensing technology. Background Art

[0002] In the current era of rapid development of digitalization and informatization, the acquisition and processing of geospatial information play a crucial role in the development of many fields. From urban planning, land resource management to agricultural monitoring, disaster emergency response, etc., accurate and timely surveying and mapping data are important bases for scientific decision-making. Traditional surveying and mapping technologies are often inefficient when dealing with large-area and complex terrain areas and are difficult to meet the rapidly changing real-world needs. With the continuous innovation of remote sensing technology, UAVs have gradually become a new force in the surveying and mapping field due to their flexibility, high efficiency, low cost and other advantages.

[0003] However, existing surveying and mapping technologies have many limitations. For example, the Chinese patent with the application number 202410198586.0 discloses a surveying and mapping system and method for dynamic remote sensing monitoring. This solution can make multiple different remote sensing images have reference points by marking the map with data reference, enabling more accurate image stitching. At the same time, the reference points can provide references for surveying and mapping data and the position trajectory of the UAV, facilitating multiple UAVs to confirm the flight trajectory and the accurate stitching of subsequent remote sensing images. By setting up a base station group, it is used to obtain more accurate spatial data feedback of the UAV and mutually reference the feedback data to calibrate the remote sensing images obtained by the UAV and provide references for surveying and mapping data.

[0004] But this solution has the following deficiencies: When stitching the remote sensing images taken by multiple UAVs to obtain a complete regional remote sensing image, there may be problems with mismatched geographical coordinates. Although there are surveying and mapping marks as references, in large-scale surveying and mapping areas, due to factors such as image deformation, the geographical coordinates at the image stitching may be inaccurate, affecting the overall accuracy of the map.

[0005] For example, the Chinese patent with the application number 202211304414.4 discloses a remote sensing camera three-dimensional surveying and mapping method and system based on multi-source data fusion. This solution collects the external contour image information of the surveyed object from bottom to top, the floor height and specific layout of a certain floor inside the surveyed object, and after data integration and processing, conducts three-dimensional modeling, solving the problem of single measurement data, resulting in the inability to display all aspects of information of the surveyed object in the three-dimensional surveying and mapping results, so that the modeling results can greatly restore the ontology of the surveyed object and improve the surveying and mapping effect.

[0006] However, this solution has the following deficiencies: This solution mainly focuses on aspects such as the height of the surveyed object, the floor height measurement inside the surveyed object, and the calculation of the floor area of the surveyed object. However, when collecting the image information of the outer contour of the surveyed object, the elevation information of the location is not considered, which may affect the accuracy of the final three-dimensional modeling and cannot accurately restore the true spatial relationship between the surveyed object and the surrounding terrain. Summary of the Invention

[0007] In order to overcome the deficiencies in the background technology, the embodiments of the present invention provide an intelligent unmanned aerial vehicle (UAV) mapping system based on remote sensing technology, which can effectively solve the problems involved in the above-mentioned background technology.

[0008] The object of the present invention can be achieved through the following technical solutions: The present invention provides an intelligent UAV mapping system based on remote sensing technology, including: a remote sensing image acquisition module, which is used to set UAV parameters, use the UAV to acquire surface remote sensing images, and perform geometric correction on the surface remote sensing images using each ground control point.

[0009] An image correction and stitching module, which is used to correct the surface remote sensing images through the time stamp correspondence relationship, extract the descriptors of each surface remote sensing image to generate each matching pair, and form a complete remote sensing image of the mapping area.

[0010] An orthorectification module, which is used to train a correction model to convert the ground control point coordinates into image coordinates, and use the K-fold cross-validation method to adjust the model parameters within the allowable error range.

[0011] A ground object recognition module, which is used to build a recognition model based on a convolutional neural network, identify the types of each ground object in the mapping area and output them, and delineate the edge contours of each known type of ground object.

[0012] An attribute information entry module, which is used to create a basic attribute table, associate the basic attributes with the ground objects in the remote sensing image of the mapping area, and enter the basic attributes corresponding to the types of each ground object.

[0013] A map production module, through a stereo matching algorithm, generates point cloud data containing three-dimensional coordinate information, calculates the elevation of each interpolation point in the mapping area through interpolation calculation to form a continuous surface model, and combines the ground object vector data to generate a complete map.

[0014] A management database, which is used to store surface remote sensing image data.

[0015] Preferably, the UAV parameters include flight height and acquisition time interval.

[0016] Preferably, the remote sensing image acquisition module further includes: the drone takes surface remote sensing images at preset time intervals, and the ground control point measurement device synchronously records the coordinates of ground control points at the same time intervals, and timestamps are respectively marked on the acquired surface remote sensing images and each ground control point.

[0017] Match the timestamps of the surface remote sensing images with the timestamps of the ground control points, and accurately mark the coordinates of each ground control point on the corresponding surface remote sensing image through the corresponding relationship of the timestamps.

[0018] Perform geometric correction on the corresponding surface remote sensing image through the coordinates of the ground control points.

[0019] Preferably, the specific analysis method of the image correction and stitching module is: extract feature points from each surface remote sensing image and generate descriptors, calculate the distances between the descriptors of each surface remote sensing image and the descriptors of other surface remote sensing images, find the descriptor with the smallest distance from the descriptors of other surface remote sensing images to the descriptor of this surface remote sensing image, mark this pair of descriptors as a matching pair, obtain each matching pair, set a distance threshold, filter out the matching pairs with distances greater than the threshold, use the filtered matching pairs to estimate the geometric transformation relationship between each surface remote sensing image, and stitch the geometrically corrected surface remote sensing images according to their positions and directions in the survey area to form a complete remote sensing image of the survey area.

[0020] Preferably, the specific analysis method of the orthorectification module is: obtain the surface remote sensing images imported with the coordinates of each ground control point, select a correction model according to the type of the surface corresponding to the remote sensing image of the survey area, import the coordinates of each ground control point into the correction model, for any given coordinates of the ground control point, substitute it into the selected transformation model equation to obtain the corresponding image coordinates.

[0021] Calculate the error between the coordinates of the ground control point and its corresponding image coordinates. If the error is greater than the set allowable error range, divide the coordinates of each ground control point and its corresponding image coordinates into K subsets. Each subset is used as a validation set in a certain iteration, and the remaining K - 1 subsets are used as training sets. Input the training sets into the correction model until each subset is used as a validation set once, train the model and record the performance indicators on the validation set until the model performance of the correction model meets the preset requirements, and the training is completed to obtain a correction model with adjusted parameters.

[0022] Use the correction model with adjusted parameters to perform coordinate transformation again, and calculate the transformation error again until the error is within the set allowable error range.

[0023] Preferably, the specific analysis method of the feature recognition module is as follows: construct a recognition model based on a convolutional neural network, obtain a large amount of image data and divide it into a training set and a test set, and input the training set into the recognition model for training until the loss function is stable.

[0024] Input the test set into the recognition model until the model parameters of the recognition model meet the preset requirements, and the training is completed to obtain a trained recognition model.

[0025] Input the remote sensing image of the surveying and mapping area into the trained recognition model for recognition to identify each feature in the surveying and mapping area, and output the types of each feature in the surveying and mapping area.

[0026] Preferably, the specific analysis method for delineating the edge contours of each known type of feature is as follows: perform grayscale processing on the remote sensing image of the surveying and mapping area, select sample areas of each known feature type from it, calculate the grayscale values of the pixels in the sample areas of each known feature type to obtain the grayscale value ranges of the pixels in each sample area, and select the intermediate value between the minimum and maximum grayscale values as the grayscale value threshold for the sample areas of each known feature type.

[0027] According to the set grayscale value thresholds of the sample areas of each known feature type, perform binary processing on the remote sensing image of the surveying and mapping area respectively, divide the pixels in the remote sensing image of the surveying and mapping area into pixels of each known type of feature and background pixels, and use an edge detection operator to extract the edge contours of each known type of feature.

[0028] Preferably, the specific analysis method of the attribute information input module is as follows: convert the types of each feature in the surveying and mapping area obtained into pixel values, obtain the basic attributes of the types of each feature, create a basic attribute table, calculate the spatial distances between the coordinates of each feature in the basic attribute table and the coordinates of each ground control point in the remote sensing image of the surveying and mapping area, establish a mapping relationship between the ground control points and the features, perform vectorization processing on the remote sensing image of the surveying and mapping area to obtain feature vector boundary data containing feature boundary information, use the vector boundary data as the target layer and the basic attribute table as the connection layer, and perform matching according to the mapping relationship between the ground control points and the features to input the basic attributes corresponding to the types of each feature.

[0029] Preferably, the specific operation method of the map production module is as follows: through a stereo matching algorithm, for the image pairs with overlapping areas in the remote sensing image of the surveying and mapping area, find homologous feature points to generate point cloud data containing three-dimensional coordinate information, and use the elevation information of each point extracted from it to classify it into ground points and non-ground points.

[0030] The mapping area is divided into grids. Each node of the grid serves as an interpolation point. A search radius is set, and several adjacent ground points are set within the search radius of each interpolation point according to the equal-spacing principle. The distances between each interpolation point and each adjacent ground point are calculated based on the point cloud data, the weights of each adjacent ground point for each interpolation point are calculated, and the elevation of each interpolation point is calculated through weighted averaging.

[0031] The elevations of each interpolation point are organized in the form of a regular grid to form a continuous surface model.

[0032] Preferably, the map production module further includes: superimposing and fusing the generated continuous surface model with the vector data of features containing feature boundary information, connecting the basic attribute table with the vector data of features, associating the basic attribute table with the vector data of features through the unique identifier of the feature, and annotating the attribute information of the feature to form a complete map.

[0033] Compared with the prior art, the present invention has the following beneficial effects: First, the present invention obtains surface remote sensing images of various areas, extracts descriptors of each surface remote sensing image to generate matching pairs, determines the corresponding relationships of the same features between different images, provides a basis for image mosaicking, forms a complete remote sensing image of the mapping area, and trains a correction model to convert the coordinates of ground control points into image coordinates, so that the image has accurate geographic coordinate information.

[0034] Second, the present invention creates a basic attribute table based on the basic attributes corresponding to the types of various features, improves the information content and practicality of the map, combines with the elevations of each interpolation point in the mapping area obtained by interpolation calculation to generate a complete map, enabling the map to not only show the distribution of features but also reflect the terrain undulation, forming a map with rich content. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0036] Figure 1 It is the connection diagram of the system modules of the present invention.

[0037] Figure 2 is Figure 1 the flowchart of the image correction and mosaicking module in

[0038] Figure 3 is Figure 1 the flowchart of the orthorectification module in DETAILED DESCRIPTION OF THE INVENTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0040] Please refer to Figure 1 As shown, an unmanned aerial vehicle (UAV) intelligent mapping system based on remote sensing technology includes a remote sensing image acquisition module, an image correction and stitching module, an orthorectification module, a ground object recognition module, an attribute information entry module, a map making module, and a management database.

[0041] The management database is connected to the orthorectification module, the ground object recognition module, the attribute information entry module, and the map making module. The image correction and stitching module is connected to the remote sensing image acquisition module and the orthorectification module. The attribute information entry module is connected to the ground object recognition module and the map making module.

[0042] The remote sensing image acquisition module is used to set UAV parameters, use the UAV to acquire surface remote sensing images, and perform geometric correction on the surface remote sensing images using each ground control point.

[0043] The UAV parameters include flight altitude and acquisition time interval.

[0044] It should be noted that the specific analysis method for the flight altitude is as follows: Obtain the camera lens focal length , the camera sensor pixel size and the target ground resolution , and substitute them into the formula: Calculate the flight altitude .

[0045] In a specific embodiment, the camera sensor pixel size is 5 microns, the lens focal length is 20 millimeters. If the target image resolution requires a ground resolution of 5 centimeters (i.e., 0.05 meters), substituting into the formula gives the flight altitude meters.

[0046] The image correction and stitching module is used to correct the surface remote sensing images through the timestamp correspondence relationship, extract the descriptors of each surface remote sensing image to generate each matching pair, and form a complete remote sensing image of the mapping area.

[0047] Please refer to Figure 2As shown, the remote sensing image acquisition module further includes: the drone takes surface remote sensing images at preset time intervals, and the ground control point measurement device synchronously records the coordinates of ground control points at the same time intervals, and respectively marks time stamps on the acquired surface remote sensing images and each ground control point; it can accurately associate the coordinates of ground control points with the corresponding images, so that the images have accurate geographic coordinate information, improve the geographic positioning accuracy of the images, and meet the requirements of high-precision surveying and mapping.

[0048] Match the time stamps of each surface remote sensing image with the time stamps of the ground control points. Through the corresponding relationship of the time stamps, accurately mark the coordinates of each ground control point on the corresponding surface remote sensing image; ensure the accurate association between the ground control points and the images, so that the features on the images can accurately correspond to the actual geographic coordinates, improve the accuracy of image geographic positioning, and meet the requirements of high-precision surveying and mapping and geographic information analysis.

[0049] Perform geometric correction on the corresponding surface remote sensing image through the coordinates of the ground control points; after correction, the geometric accuracy of the image is improved, and the shape, size and position of the features are more in line with the actual situation, which is beneficial to subsequent image interpretation, information extraction and analysis.

[0050] The specific analysis method of the image correction and mosaicking module is as follows: extract feature points from each surface remote sensing image and generate descriptors. For the descriptors of each surface remote sensing image, calculate the distance between its descriptor and the descriptors of other surface remote sensing images, find the descriptor with the smallest distance from the descriptors of other surface remote sensing images to the descriptor of this surface remote sensing image, mark this pair of descriptors as a matching pair to obtain each matching pair, set a distance threshold, filter out the matching pairs with a distance greater than this threshold, use the filtered matching pairs to estimate the geometric transformation relationship between each surface remote sensing image, and splice the geometrically corrected surface remote sensing images according to their positions and directions in the surveying area to form a complete remote sensing image of the surveying area; feature points and descriptors have a certain invariance to the geometric and radiometric changes of the image, and can accurately find corresponding points under different perspectives and lighting conditions, improving the accuracy and reliability of matching.

[0051] It should be noted that the feature points are points with significant features in the surface remote sensing image, such as corner points and edge points. These feature points can still remain relatively stable when the image rotates, scales, and changes in brightness, etc. They are important bases for image matching and geometric correction. For example, in the image of high-rise buildings in the city center, the points at the edges and corners of the high-rise buildings are feature points.

[0052] The descriptor is a quantitative representation of the local area features around the feature point, used to describe the characteristics of the feature point, which is convenient for subsequent calculation of the similarity between feature points. For example, the SIFT descriptor generates a 128-dimensional vector to represent the feature point by calculating the gradient direction histogram in the neighborhood of the feature point.

[0053] The matching pair consists of a pair of two feature points from different surface remote sensing images with the smallest descriptor distance, which are considered corresponding points of the same ground object in different images and are used to establish the association between images.

[0054] In a specific embodiment, there is a project for surveying and mapping a certain urban area. 100 surface remote sensing images are obtained. The SIFT algorithm is selected to extract feature points from these images. On average, about 500 extreme points per image are used as feature points. Based on the extracted feature points, the gradient direction histogram is calculated in its neighborhood, and a 128-dimensional vector is obtained as the descriptor of the feature point. Since the SIFT descriptor, which is a floating-point vector, is used, the Euclidean distance is used to calculate the distance between the descriptors of each surface remote sensing image and the descriptors of all other surface remote sensing images. In the distance matrix, for each descriptor, the descriptor with the smallest distance from it among the descriptors of the remaining surface remote sensing images is found, and the feature points corresponding to these two descriptors are marked as a matching pair. By traversing all descriptors, about 20,000 matching pairs are obtained in total. A distance threshold of 0.6 is set. By traversing all the generated 20,000 matching pairs and checking the distance between each pair of descriptors, the matching pairs with a distance greater than 0.6 are filtered out, and about 15,000 matching pairs after screening are obtained.

[0055] The number of iterations is set to 500 times, and the minimum number of matching pairs is determined to be 4 pairs. Each time an iteration is performed, 4 pairs of matching pairs are randomly selected from 15,000 matching pairs. Using these 4 pairs of matching pairs, the transformation matrix parameters are calculated according to the homography transformation model formula. By solving the linear equations, a 3×3 homography transformation matrix is obtained. The remaining 14,996 matching pairs are used to test this transformation matrix. For each remaining matching pair, its coordinates in one image are mapped to the other image through the transformation matrix, and the error between the mapped coordinates and the actual matching point coordinates is calculated. The error threshold is set to 2 pixels. If the error is within the error threshold, then this matching pair is considered an inlier. The number of inliers in this iteration is counted. If the number of inliers in this iteration is more than the previously recorded maximum number of inliers, then the optimal homography transformation matrix and its number of inliers are updated.

[0056] After 500 iterations, the homography transformation matrix obtained in the iteration with the largest number of inliers is selected. Through this homography transformation matrix, the points on each surface remote sensing image are mapped to specific positions on the remote sensing image of the surveying area, so as to establish the corresponding relationship between points. Furthermore, the geometrically corrected surface remote sensing images are stitched together according to their positions and directions in the surveying area to form a complete remote sensing image of the surveying area.

[0057] An orthorectification module is used to train a correction model to convert ground control point coordinates into image coordinates, and use the K-fold cross-validation method to adjust the model parameters within the error tolerance range.

[0058] Please refer to Figure 3 As shown, the specific analysis method of the orthorectification module is as follows: Obtain the surface remote sensing images of each ground control point's coordinates. Select a correction model according to the type of the surface to which the remote sensing image of the mapping area belongs. Import the coordinates of each ground control point into the correction model. For any given ground control point coordinate, substitute it into the selected transformation model equation to obtain the corresponding image coordinate; accurate coordinate transformation provides a reliable basis for subsequent image mosaicking, information extraction, etc., making the processing more accurate and effective.

[0059] The transformation model equation is a specific mathematical expression contained in the correction model. The specific mathematical expression contains the parameters of the model. Substitute the coordinates of the ground control point into the transformation model equation, and use the mapping rule defined by this equation to perform mathematical operations in the form of the model equation to obtain the image coordinate corresponding to this ground control point.

[0060] It should be noted that the type of the surface to which it belongs is the category presented by the natural form or material composition of the earth's surface, mainly divided based on natural elements such as topography, geological structure, soil type, etc., including mountains, plains, plateaus, and basins.

[0061] Calculate the error between the ground control point coordinate and its corresponding image coordinate. If the error is greater than the set error tolerance range, divide the ground control point coordinates and their corresponding image coordinates into K subsets. Each subset is used as a validation set in a certain iteration, and the remaining K - 1 subsets are used as training sets. Input the training sets into the correction model until each subset is used as a validation set once, train the model and record the performance metrics on the validation set until the model performance of the correction model meets the preset requirements, and the training is completed to obtain the correction model with adjusted parameters; Using the cross-validation method to divide subsets for training and validation can more effectively evaluate the performance of the model on different data subsets, avoid overfitting of the model, and improve the stability and reliability of the model.

[0062] It should be noted that the specific analysis method of the error between the ground control point coordinate and its corresponding image coordinate is: Denote the ground control point coordinate and its corresponding image coordinate as , indicating the number of the th ground control point, , and obtain the error between the ground control point coordinate and its corresponding image coordinate through the formula , Indicates the number of ground control points.

[0063] For planar coordinates, The Euclidean distance formula between is the straight-line distance between two points in the planar space. Here, it is the distance error between the coordinates of a single ground control point and its corresponding image coordinates. To measure the overall error situation, all ground control points need to be considered. First, calculate the sum of the squares of the errors for each control point , and then calculate the average value, that is, divide by the number of control points , to obtain . To make the final error value consistent with the Euclidean distance of a single point in terms of dimension, take the square root of the average value to obtain the formula .

[0064] Use the corrected model with adjusted parameters to perform coordinate transformation again and calculate the transformation error again until the error is within the set allowable error range; this can ensure that the final coordinate transformation result has high reliability and provides more accurate data support for subsequent image processing and analysis.

[0065] The allowable error range is set according to the surface type corresponding to the remote sensing image of the survey area. >2 pixels in the plain area and >3 pixels in the mountainous area.

[0066] The ground object recognition module is used to construct a recognition model based on a convolutional neural network to identify the types of various ground objects in the survey area and output and delineate the edge contours of known types of ground objects.

[0067] The specific analysis method of the ground object recognition module is: construct a recognition model based on a convolutional neural network, obtain a large amount of image data and divide it into a training set and a test set, and input the training set into the recognition model for training until the loss function is stable; this avoids the cumbersome process and subjectivity of manually designing features, reduces the influence of human factors on feature extraction, and improves the efficiency and accuracy of feature extraction.

[0068] It should be noted that the convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives image data; the convolutional layer slides different-sized convolutional kernels on the image for convolutional operations to extract local features in the image; the pooling layer downsamples the feature map to reduce the amount of data and computational complexity, while retaining the main features to prevent overfitting; the fully connected layer unfolds the feature map after convolution and pooling into a one-dimensional vector, and then performs feature fusion and classification decision-making through multiple fully connected layers; the output layer: is set according to the specific recognition task.

[0069] It should be noted that the specific operation process until the loss function becomes stable is as follows: In each training iteration, a batch of image data in the training set is input into the recognition model. The model makes predictions based on the current parameters to obtain the prediction results. The loss value between the prediction results and the true labels is calculated through the loss function. Based on the calculated loss value, the gradient of the loss function with respect to the model parameters is calculated, and the model parameters are adjusted according to the direction and magnitude of the gradient. As the training progresses, the value of the loss function will continuously change. When the change amplitude of the loss function value meets the requirements in consecutive multiple training iterations, for example, the difference between the loss values of two adjacent iterations is less than a preset threshold, or the loss value fluctuates within a certain range and no longer significantly decreases, it can be considered that the loss function has reached a stable state.

[0070] Input the test set into the recognition model until the model parameters of the recognition model meet the preset requirements, and the training is completed to obtain a trained recognition model; through repeated verification and adjustment, continuously optimize the parameters and structure of the model to make the model reach the best state in terms of accuracy, stability, etc., and improve the effect of ground object recognition.

[0071] Input the remote sensing image of the survey area into the trained recognition model for recognition to recognize each ground object in the survey area and output the types of each ground object in the survey area; compared with manual visual interpretation of remote sensing images, using a trained model for ground object recognition has higher efficiency and speed. A large amount of image data can be processed in a short time, improving work efficiency.

[0072] It should be noted that the ground objects are various natural and artificial fixed objects on the earth's surface, including lakes, forests, buildings, roads, and farmlands.

[0073] The specific analysis method for delineating the edge contours of each known type of ground object is as follows: Perform grayscale processing on the remote sensing image of the survey area, select sample areas of each known ground object type from it, obtain the grayscale values of each pixel in the sample areas of each known ground object type, and thus obtain the grayscale value range of the pixels in the sample areas of each known ground object type. Select the intermediate value between the minimum and maximum grayscale values as the grayscale value threshold for the sample areas of each known ground object type; By selecting sample areas and calculating the grayscale value range and threshold, unique grayscale features of different ground object types can be extracted, providing a basis for accurately identifying and distinguishing ground objects.

[0074] According to the gray value thresholds of the sample areas of each known ground object type set, the remote sensing images of the survey area are binarized respectively, and the pixels in the remote sensing images of the survey area are divided into pixels of each known type of ground object and background pixels. The edge contours of each known type of ground object are extracted by using edge detection operators. Edge contour extraction can accurately obtain the shape information of the ground object, providing an important basis for the further identification and classification of the ground object.

[0075] The attribute information input module is used to create a basic attribute table, associate the basic attributes with the ground objects in the remote sensing images of the survey area, and input the basic attributes corresponding to the types of each ground object.

[0076] The specific analysis method of the attribute information input module is as follows: convert the types of each ground object in the obtained survey area into pixel values, obtain the basic attributes of the types of each ground object, create a basic attribute table, calculate the spatial distance between the coordinates of each ground object in the basic attribute table and the coordinates of each ground control point in the remote sensing images of the survey area, establish the mapping relationship between the ground control points and the ground objects, perform vectorization processing on the remote sensing images of the survey area to obtain the ground object vector boundary data containing the ground object boundary information, use the vector boundary data as the target layer and the basic attribute table as the connection layer, and perform matching according to the mapping relationship between the ground control points and the ground objects to input the basic attributes corresponding to the types of each ground object. The unified pixel value representation makes the data structure more regular. When performing data processing and analysis, it can reduce the processing complexity and improve the calculation efficiency. With the help of the high-precision coordinates of the ground control points, it can improve the positioning accuracy of the ground objects in the survey area and make the position information of the ground objects more accurate and reliable.

[0077] It should be noted that the vector boundary data is composed of three basic geometric elements: points, lines, and surfaces. Points are used to determine key positions. Lines are formed by connecting a series of continuous points and are used to represent linear ground objects (such as roads, rivers) or the boundaries of planar ground objects. Multiple line elements are closed to form a surface, representing planar ground objects (such as lakes, building areas).

[0078] It should be noted that the coordinates of the ground objects in the basic attribute table and the coordinates of each ground control point in the remote sensing images of the survey area are respectively denoted as , represents the number of the th ground object, , represents the number of the th ground control point, , and the calculation formula for the spatial distance between the coordinates of each ground object and the coordinates of each ground control point in the remote sensing images of the survey area is .

[0079] The derivation of the formula for the spatial distance is based on the extension of the Pythagorean theorem. In a two-dimensional plane, the distance formula between two points is , which is a direct application of the Pythagorean theorem. The horizontal and vertical distances between two points are used to obtain the straight-line distance by taking the square root of the sum of squares. When extended to three-dimensional space, the z - dimension coordinate is added. Suppose there are two points , the distance between the projection points of these two points on the x - y plane can be calculated first . Then, considering the distance in the z - direction and using the Pythagorean theorem according to the geometric relationship in three - dimensional space, the distance formula between these two points in three - dimensional space can be obtained .

[0080] It should be noted that in a specific embodiment, it is stipulated that the pixel value corresponding to vegetation - related features is 1, the pixel value corresponding to water bodies is 2, and the pixel value corresponding to buildings is 3. The type of each feature is judged, and then according to the set conversion standard, each feature type is replaced with the corresponding pixel value.

[0081] Specifically, referring to Table 1, a part of the basic attributes with practical significance are listed in Table 1.

[0082] Table 1. Basic Attribute Table

[0083]

[0084] It should be noted that the feature ID is a code used to uniquely identify each feature object, that is, the unique identifier of the feature.

[0085] The map - making module generates point - cloud data containing three - dimensional coordinate information through a stereo - matching algorithm, calculates the elevation of each interpolation point in the surveyed area through interpolation, forms a continuous surface model, and generates a complete map in combination with feature vector data.

[0086] The specific operation method of the map - making module is as follows: through a stereo - matching algorithm, for image pairs with overlapping areas in the remote - sensing images of the surveyed area, homologous feature points are found to generate point - cloud data containing three - dimensional coordinate information. Using the elevation information of each point extracted from it, they are classified into ground points and non - ground points. Accurate matching of homologous feature points can ensure that the calculated three - dimensional coordinates of features are more accurate, thus improving the three - dimensional reconstruction accuracy of the entire surveyed area and making the generated map more realistically reflect the actual terrain and the spatial position of features.

[0087] It should be noted that the specific analysis method of the stereo - matching algorithm is as follows: the remote - sensing images of the surveyed area with overlapping areas are split into two independent images. In one of the images, a small window centered on a certain pixel point is selected, and within the set range of the other image, the window with the highest similarity to this window is found, and the central pixel point of the window with the highest similarity is used as the homologous feature point corresponding to the window central pixel point in the first image.

[0088] The specific analysis method of the similarity is as follows: Let the window selected from the first image be , and its pixel values , represent the coordinates of the pixels within the window. The window within the search range in the second image is , with pixel values . Substitute them into the formula to obtain the similarity of the two windows.

[0089] It should be noted that the similarity calculation formula is based on comparing the pixel value differences of the corresponding pixel points in the two windows, calculating their differences . To avoid the situation of positive and negative differences canceling each other out and to highlight the influence of larger differences, we perform a square operation on the differences, that is . Summing up the squared differences of all pixel points within the window, we obtain .

[0090] The surveying and mapping area is divided into a grid. Taking the nodes of each grid as the interpolation points, setting a search radius, and setting several adjacent ground points within the search radius of each interpolation point according to the equal-spacing principle. Calculate the distances between each interpolation point and each adjacent ground point based on the point cloud data, calculate the weights of each interpolation point and each adjacent ground point, and calculate the elevation of each interpolation point through weighted average.

[0091] It should be noted that the specific analysis method of the elevation of each interpolation point is as follows: Calculate the weights of each interpolation point and each adjacent ground point: , where represents the number of the th interpolation point, , represents the number of the th adjacent ground point, , represents the distance between the th interpolation point and its th adjacent ground point. Calculate the elevation of each interpolation point through weighted average , where represents the elevation of the th interpolation point and its th adjacent ground point.

[0092] It should be noted that the weight formula for each interpolation point and each adjacent ground point uses the reciprocal of the square of the distance as the weight. The larger the distance , the smaller its reciprocal of the square , which means that the influence of this adjacent ground point on the elevation of the interpolation point is smaller. On the contrary, the closer the distance, the greater the weight, and the greater the influence on the elevation of the interpolation point; in the formula for the elevation of each interpolation point, the numerator is the elevation of each adjacent ground point The corresponding weight The multiplication and summation reflects the weighted contribution of each adjacent point to the interpolation point elevation. is the sum of the weights of all nearby ground points, which plays a normalization role and ensures that the calculated is a reasonable weighted average. In this way, the interpolation points Elevation The elevation of each nearby ground point and their distance relationship to the interpolation point are comprehensively considered.

[0093] The elevations of each interpolation point are organized in the form of a regular grid to form a continuous surface model.

[0094] The map making module also includes: superimposing and fusing the generated continuous surface model with the feature vector data containing feature boundary information, connecting the basic attribute table with the feature vector data, associating the basic attribute table with the feature vector data through the unique identifier of the feature, marking the feature attribute information to form a complete map; connecting the basic attribute table with the feature vector data to facilitate users to query and analyze the feature attribute information, thereby improving the interactivity and practicality of the map.

[0095] It should be noted that the feature vector data is a data type that represents features in geographic space in vector form, and it accurately describes the location, shape, boundary and area of features through geometric elements such as points, lines and surfaces.

[0096] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention and they are still covered by the protection scope of the present invention.

Claims

1. An intelligent UAV mapping system based on remote sensing technology, characterized in that, The system specifically includes the following modules: A remote sensing image acquisition module, which is used to set the parameters of the unmanned aerial vehicle (UAV), use the UAV to acquire surface remote sensing images, and perform geometric correction on each surface remote sensing image using each ground control point; An image correction and stitching module, which is used to correct the surface remote sensing images through the timestamp correspondence relationship, extract descriptors of each surface remote sensing image to generate each matching pair, and form a complete remote sensing image of the survey area; An orthorectification module, which is used to train a correction model to convert the ground control point coordinates into image coordinates, and use the K-fold cross-validation method to adjust the model parameters within the allowable error range; A ground object recognition module, which is used to construct a recognition model based on a convolutional neural network, identify and output the types of each ground object in the survey area, and delimit the edge contours of each known type of ground object; An attribute information entry module, which is used to create a basic attribute table, associate the basic attributes with the ground objects in the remote sensing image of the survey area, and enter the basic attributes corresponding to the types of each ground object; A map production module, which generates point cloud data containing three-dimensional coordinate information through a stereo matching algorithm, obtains the elevation of each interpolation point in the survey area through interpolation calculation to form a continuous surface model, and combines the ground object vector data to generate a complete map; A management database, which is used to store surface remote sensing image data; The remote sensing image acquisition module further includes: The UAV takes surface remote sensing images at a preset time interval, and the ground control point measurement device synchronously records the ground control point coordinates at the same time interval, and marks timestamps for each acquired surface remote sensing image and each ground control point respectively; Match the timestamps of each surface remote sensing image with the timestamps of the ground control points, and accurately mark the coordinates of each ground control point on the corresponding surface remote sensing image through the timestamp correspondence relationship; Perform geometric correction on the corresponding surface remote sensing image through the coordinates of the ground control points.

2. The drone intelligent mapping system based on remote sensing technology according to claim 1, wherein The UAV parameters include flight altitude and acquisition time interval.

3. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 1, characterized in that, The specific analysis method of the image correction and stitching module is as follows: Extract feature points from each surface remote sensing image and generate descriptors. For the descriptors of each surface remote sensing image, calculate the distance between it and the descriptors of other surface remote sensing images, find the descriptor with the smallest distance from the descriptors of other surface remote sensing images to the descriptor of this surface remote sensing image, mark this pair of descriptors as a matching pair to obtain each matching pair, set a distance threshold, filter out the matching pairs with a distance greater than this threshold, use the filtered matching pairs to estimate the geometric transformation relationship between each surface remote sensing image, and stitch the geometrically corrected surface remote sensing images according to their positions and directions in the survey area to form a complete remote sensing image of the survey area.

4. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 1, characterized in that, The specific analysis method of the orthorectification module is as follows: Obtain the surface remote sensing images into which the coordinates of each ground control point are imported, select a correction model according to the type of the surface corresponding to the remote sensing image of the survey area, import the coordinates of each ground control point into the correction model, and for any given ground control point coordinates, substitute them into the selected conversion model equation to obtain the corresponding image coordinates; Calculate the error between the coordinates of the ground control points and their corresponding image coordinates. If the error is greater than the set error tolerance range, divide the coordinates of each ground control point and its corresponding image coordinates into K subsets. Each subset is used as a validation set in a certain iteration, and the remaining K - 1 subsets are used as training sets. Input the training sets into the correction model until each subset is used as a validation set once. Train the model and record the performance metrics on the validation set until the model performance of the correction model meets the preset requirements. Training is completed to obtain the correction model with adjusted parameters; Use the correction model with adjusted parameters to perform coordinate transformation again and calculate the transformation error again until the error is within the set error tolerance range.

5. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 1, characterized in that, The specific analysis method of the feature recognition module is as follows: Build a recognition model based on a convolutional neural network. Obtain a large amount of image data and divide it into a training set and a test set. Input the training set into the recognition model for training until the loss function is stable; Input the test set into the recognition model until the model parameters of the recognition model meet the preset requirements. Training is completed to obtain the trained recognition model; Input the remote sensing image of the survey area into the trained recognition model for recognition to recognize each feature in the survey area and output the types of each feature in the survey area.

6. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 5, characterized in that, The specific analysis method for delineating the edge contours of each known type of feature is as follows: Perform grayscale processing on the remote sensing image of the survey area. Select sample areas of each known feature type from it, calculate the grayscale values of the pixels in the sample areas of each known feature type, obtain the range of grayscale values of the pixels in each sample area, and select the intermediate value between the minimum and maximum grayscale values as the grayscale value threshold for the sample areas of each known feature type; According to the set grayscale value thresholds of the sample areas of each known feature type, perform binary processing on the remote sensing image of the survey area respectively, divide the pixels in the remote sensing image of the survey area into pixels of each known type of feature and background pixels, and use an edge detection operator to extract the edge contours of each known type of feature.

7. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 1, characterized in that, The specific analysis method of the attribute information entry module is as follows: Convert the types of each feature in the survey area obtained into pixel values, obtain the basic attributes of the types of each feature, create a basic attribute table, calculate the spatial distance between the coordinates of each feature in the basic attribute table and the coordinates of each ground control point in the remote sensing image of the survey area, establish a mapping relationship between the ground control points and the features, perform vectorization processing on the remote sensing image of the survey area to obtain the vector boundary data of the features containing feature boundary information, use the vector boundary data as the target layer and the basic attribute table as the connection layer, and perform matching according to the mapping relationship between the ground control points and the features to enter the basic attributes corresponding to the types of each feature.

8. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 1, characterized in that, The specific operation method of the map production module is as follows: Through a stereo matching algorithm, for image pairs with overlapping areas in the remote sensing image of the survey area, find homologous feature points to generate point cloud data containing three-dimensional coordinate information, and use the elevation information of each point extracted from it to classify it into ground points and non-ground points; The surveying and mapping area is divided into grids. Taking each node of the grid as each interpolation point, a search radius is set. According to the equal-spacing principle, a number of adjacent ground points are set within the search radius of each interpolation point. The distances between each interpolation point and each adjacent ground point are calculated based on the point cloud data, the weights of each interpolation point and each adjacent ground point are calculated, and the elevation of each interpolation point is calculated through weighted averaging. The elevations of each interpolation point are organized in the form of a regular grid to form a continuous surface model.

9. An unmanned aerial vehicle intelligent mapping system based on remote sensing technology according to claim 1, characterized in that, The map production module further includes: The generated continuous surface model is superimposed and fused with the feature vector data containing feature boundary information. The basic attribute table is connected to the feature vector data, and the basic attribute table is associated with the feature vector data through the unique identifier of the feature. The attribute information of the feature is labeled to form a complete map.

Images