Mangrove forest ecological information extraction method and system based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of image processing, and discloses a mangrove forest ecological information extraction method and system based on an unmanned aerial vehicle. The method comprises the following steps: acquiring overlooking and side-looking photos through an unmanned aerial vehicle; correcting the overlook picture by using the coordinate information and superposing the overlook picture to the remote sensing image; generating a photographing direction pattern layer based on the side view photo; establishing a mangrove forest species interpretation mark library by using a random forest algorithm; classifying selected samples to generate a species distribution diagram; and finally evaluating the mangrove forest ecological function to form ecological monitoring data. According to the method, multi-angle and multi-level high-precision ecological information extraction of the mangrove forest is realized, so that a more accurate and comprehensive scientific basis is provided for species classification, ecological system function evaluation and protection management of the mangrove forest.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular, to a method and system for extracting mangrove ecological information based on an unmanned aerial vehicle (UAV). Background Art

[0002] Mangroves are woody plant communities growing in the intertidal zones of tropical and subtropical coasts or river estuaries, and are important marine ecosystems, which have important indicative significance for the global environment and climate change. In recent years, with the development of remote sensing technology, more and more researchers have studied the spatial distribution, area change, population classification, biomass and carbon storage estimation of mangroves based on remote sensing technology. Traditional mangrove monitoring methods mainly rely on ground surveys and the analysis of medium- and low-resolution satellite images. Ground survey methods include quadrat method, belt transect method and standard tree method, etc. By on-site measuring parameters such as the height, diameter at breast height and density of mangroves, the growth status and ecological functions of mangroves can be evaluated; satellite remote sensing methods such as Landsat, SPOT and Sentinel multispectral satellite images are widely used in large-scale mangrove distribution monitoring and change analysis. Combining field verification data and professional interpretation techniques, a macroscopic understanding of the spatial distribution of mangroves can be achieved.

[0003] However, the existing mangrove monitoring methods have obvious deficiencies. Mangroves grow in the intertidal shallows with harsh environmental conditions and dense mangrove vegetation. It is difficult for investigators to enter the interior of the forest area, resulting in low efficiency and certain destructiveness of traditional ground surveys. Since field personnel cannot reach most areas, the distribution of survey data is uneven, affecting the representativeness and reliability of the data. And due to the limited spatial resolution and less spectral information of medium- and low-resolution satellite images, there are limitations in the fine classification of mangrove species, and it is difficult to accurately distinguish different mangrove species with similar morphologies. In addition, the existing technology lacks effective means to capture the multi-level structural characteristics of mangroves, resulting in an incomplete and inaccurate assessment of the ecological system functions of mangroves. Especially in the mangrove habitats with frequent tidal changes, it is difficult for traditional methods to achieve high-frequency and high-precision monitoring of mangroves, which severely restricts the protection and scientific management of mangrove resources. Summary of the Invention

[0004] This application provides a method and system for extracting mangrove ecological information based on an unmanned aerial vehicle (UAV), which can break through the limitations such as difficult access for personnel, uneven data distribution and low satellite image resolution in traditional mangrove surveys, and achieve high-precision ecological information extraction of mangroves from multiple angles and at multiple levels, so as to provide a more accurate and comprehensive scientific basis for the species classification, ecological system function assessment and protection management of mangroves.

[0005] First aspect, the present application provides a method for extracting mangrove ecological information based on an unmanned aerial vehicle. The method for extracting mangrove ecological information based on an unmanned aerial vehicle includes: taking low-altitude photos of the mangrove from different angles by the unmanned aerial vehicle to obtain a top-down photo with a camera lens tilt angle of approximately 90° and a side-view photo with a camera lens tilt angle of approximately 30°, so as to obtain mangrove multi-angle image data; according to the mangrove multi-angle image data, calculating projection coordinates by using position coordinates, photo length and width, and flight height information, and performing geometric correction processing on the top-down photo to obtain a corrected image superimposed on a remote sensing image; based on the position and flight angle information of the side-view photo, generating a vector point layer and performing symbolization processing to obtain a photographing direction layer superimposed on the remote sensing image; based on the corrected image, the photographing direction layer, and the remote sensing image, extracting features and establishing an association relationship through a random forest algorithm to form a mangrove species interpretation signature library; according to the mangrove species interpretation signature library, selecting training samples and verification samples on the remote sensing image, performing supervised classification processing of mangrove species, and generating a mangrove species distribution map; based on the mangrove species distribution map, combining spectral characteristics and spatial distribution information, evaluating and analyzing the functions of the mangrove ecosystem to obtain mangrove ecological monitoring data.

[0006] Second aspect, the present application provides a system for extracting mangrove ecological information based on an unmanned aerial vehicle. The system for extracting mangrove ecological information based on an unmanned aerial vehicle includes: An acquisition module, configured to take low-altitude photos of the mangrove from different angles by the unmanned aerial vehicle to obtain a top-down photo with a camera lens tilt angle of approximately 90° and a side-view photo with a camera lens tilt angle of approximately 30°, so as to obtain mangrove multi-angle image data; A calculation module, configured to calculate projection coordinates by using the mangrove multi-angle image data, position coordinates, photo length and width, and flight height information, and perform geometric correction processing on the top-down photo to obtain a corrected image superimposed on a remote sensing image; A processing module, configured to generate a vector point layer and perform symbolization processing based on the position and flight angle information of the side-view photo to obtain a photographing direction layer superimposed on the remote sensing image; An establishment module, configured to extract features and establish an association relationship through a random forest algorithm based on the corrected image, the photographing direction layer, and the remote sensing image to form a mangrove species interpretation signature library; A selection module, configured to select training samples and verification samples on the remote sensing image according to the mangrove species interpretation signature library, perform supervised classification processing of mangrove species, and generate a mangrove species distribution map; An analysis module, configured to evaluate and analyze the functions of the mangrove ecosystem based on the mangrove species distribution map, combining spectral characteristics and spatial distribution information to obtain mangrove ecological monitoring data.

[0007] In a third aspect of the present invention, a computer device is provided, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory to cause the computer device to execute the above-mentioned method for extracting mangrove ecological information based on an unmanned aerial vehicle.

[0008] In a fourth aspect of the present invention, a computer-readable storage medium is provided, in which instructions are stored. When it runs on a computer, it causes the computer to execute the above-mentioned method for extracting mangrove ecological information based on an unmanned aerial vehicle.

[0009] In the technical solution provided by this application, the mangrove is photographed at low altitude from different angles by an unmanned aerial vehicle to obtain a top-down photo with a camera lens tilt angle of about 90° and a side-view photo with a camera lens tilt angle of about 30°, forming multi-angle image data of the mangrove, breaking through the limitation that it is difficult for traditional manual surveys to enter the interior of the mangrove, and greatly improving the spatial coverage and data collection efficiency of mangrove surveys. The projection coordinates are calculated using the position coordinates, photo length and width, and flight altitude information, and the top-down photo is geometrically corrected to obtain a corrected image superimposed on the remote sensing image, solving the problem of precise registration between the unmanned aerial vehicle image and the satellite remote sensing image, and providing a basis for multi-source data fusion analysis. Based on the position and flight angle information of the side-view photo, a vector point layer is generated and symbolized to obtain a photo-taking direction layer superimposed on the remote sensing image, intuitively showing the relationship between the photo-taking direction and position of the side-view photo, and facilitating the understanding of the side structure characteristics of the mangrove. It is particularly worth emphasizing that in this solution, based on the corrected image, the photo-taking direction layer, and the remote sensing image, features are extracted and correlation relationships are established through the random forest algorithm to form a mangrove species interpretation mark library. The random forest algorithm shows excellent performance in feature selection and classification. It can not only process high-dimensional feature data, but also evaluate the importance of features and cope with the correlation between features, providing strong algorithm support for mangrove species identification; the algorithm has strong robustness to noise and outliers, and effectively reduces the risk of overfitting and improves the classification accuracy by constructing a large number of decision trees and adopting a majority voting mechanism. The interpretation mark library established in this way combines the advantages of the top-down, side-view, and remote sensing data sources, realizing high-precision identification of mangrove species. According to the established mangrove species interpretation mark library, training samples and verification samples are selected on the remote sensing image for supervised classification processing of mangrove species to generate a mangrove species distribution map, providing an accurate basis for the study of the spatial distribution of mangroves. Finally, based on the mangrove species distribution map, combined with spectral characteristics and spatial distribution information, the functions of the mangrove ecosystem are evaluated and analyzed to obtain mangrove ecological monitoring data. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0011] Figure 1 Schematic diagram of an embodiment of the method for extracting mangrove ecological information based on an unmanned aerial vehicle in an embodiment of the present application; Figure 2 Schematic diagram of geometric correction of an aerial photo of an unmanned aerial vehicle in an embodiment of the present application; Figure 3 Overlay diagram of the geometrically corrected aerial photo of the unmanned aerial vehicle and the remote sensing image in an embodiment of the present application; Figure 4 Schematic diagram of an embodiment of the system for extracting mangrove ecological information based on an unmanned aerial vehicle in an embodiment of the present application; Figure 5 Schematic block diagram of the structure of a computer device in an embodiment of the present invention. Detailed implementation manners

[0012] The embodiments of the present application provide a method and system for extracting mangrove ecological information based on an unmanned aerial vehicle. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims, and accompanying drawings of the present application are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0013] For ease of understanding, the following describes the specific process of the embodiments of the present application. Please refer to Figure 1 , an embodiment of the method for extracting mangrove ecological information based on an unmanned aerial vehicle in an embodiment of the present application includes: Step S101: Take low-altitude photos of the mangroves from different angles by an unmanned aerial vehicle to obtain an aerial photo with a camera lens tilt angle of about 90° and a side view photo with a camera lens tilt angle of about 30°, and obtain multi-angle image data of the mangroves; Step S102: Based on the multi-angle image data of mangroves, calculate the projection coordinates using the position coordinates, photo length and width, and flight height information, and perform geometric correction processing on the top-down photos to obtain the corrected images superimposed on the remote sensing images; Step S103: Based on the position and flight angle information of the side-looking photos, generate a vector point layer and perform symbolization processing to obtain the photographing direction layer superimposed on the remote sensing image; Step S104: According to the corrected images, the photographing direction layer, and the remote sensing image, extract features and establish correlation relationships through the random forest algorithm to form a mangrove species interpretation symbol library; Step S105: According to the mangrove species interpretation symbol library, select training samples and validation samples on the remote sensing image, perform supervised classification processing of mangrove species, and generate a mangrove species distribution map; Step S106: Based on the mangrove species distribution map, combine spectral features and spatial distribution information to evaluate and analyze the functions of the mangrove ecosystem, and obtain mangrove ecological monitoring data.

[0014] It can be understood that the execution subject of this application can be a mangrove ecological information extraction system based on an unmanned aerial vehicle, or a terminal or a server. Specifically, it is not limited here. This application embodiment is described by taking the server as the execution subject as an example.

[0015] Specifically, the mangroves are photographed at low altitude by an unmanned aerial vehicle. The unmanned aerial vehicle adopts two different shooting angles during flight: one is that the camera lens tilt angle is about 90° (the lens is vertically downward) to obtain top-down photos, the flight height is about 50 meters, and the shooting interval is about 40 meters; the other is that the camera lens tilt angle is about 30° to obtain side-looking photos, the flight height is about 30 meters, and the shooting interval is about 30 meters. This shooting method forms multi-angle image data of mangroves including the top and side views of mangroves, which helps to capture the characteristics of different levels of mangroves. After obtaining the multi-angle image data of mangroves, geometric correction processing is performed on the top-down photos. The specific method is to extract data such as GPS coordinates, camera focal length, photo width and height dimensions, and flight height from the EXIF information of the top-down photos. Based on these parameters, calculate the projection range of the unmanned aerial vehicle's top-down photos on the ground. Among them, the projection half-width W 2 and the projection half-height H 2 are respectively calculated by the formulas W 2 =A×W / (2×f) and H 2It is calculated by =A×H / (2×f), where A is the scale factor, W is the width of the photo, H is the height of the photo, and f is the focal length of the camera. According to the coordinates of the photo center point and the heading angle, the coordinates of the four corners of the photo on the ground are calculated using the four-point projection coordinate formula to form photogrammetric control points. Through these control points, a spatial transformation matrix is established, and an affine transformation and resampling are performed on the top-down photo to finally generate a corrected image in the same coordinate system as the remote sensing image.

[0016] For side-looking photos, the processing method is different from that of top-down photos. The GPS coordinates and flight heading angle information of the side-looking photos are extracted to construct a set of position and angle information of the side-looking photos. These information are filtered to eliminate data points with repeated positions or excessive errors. Then, point features are created using these coordinate information in the ArcGIS software to form a layer of position points of the side-looking photos of mangroves. The heading angle information is then assigned to the point layer as a direction attribute, and the photographing direction is symbolized by an arrow. The size of the arrow is proportional to the flight height. Finally, a photographing direction layer is generated and overlaid on the remote sensing image.

[0017] After having the corrected image and the photographing direction layer, a mangrove species interpretation symbol library can be established. This step first selects representative sample areas of different mangrove species in the overlapping area of the corrected image, the photographing direction layer, and the remote sensing image. For each sample area, texture, color, and shape features are extracted from the corrected image, and spectral and spatial features are extracted from the remote sensing image to form a set of feature vectors. These feature data are processed by the random forest algorithm, where the number of decision trees is set to 500, the minimum number of samples at each node is 5, and the random selection ratio of features is 0.7. Classification rules are extracted from the generated set of decision trees, and these rules are associated with the corresponding photo and image features to form a mangrove species interpretation symbol library with a "side-looking photo - top-down photo - remote sensing image" ternary association.

[0018] Based on the mangrove species interpretation signature library, species classification is carried out on remote sensing images. First, the stratified sampling method is used to select training samples and validation samples. Feature parameters such as the normalized difference vegetation index, red edge position index, and texture complexity index are extracted from the training samples to construct a sample feature matrix. This feature matrix is used to perform supervised classification on the remote sensing images to divide the distribution ranges of different mangrove species. Patch merging and boundary optimization are performed on the preliminary classification results to eliminate isolated pixels and boundary noise. Then, the validation samples are used to evaluate the accuracy of the optimized classification results, calculate the overall accuracy and the classification accuracy of each species, and generate a classification accuracy report. According to the accuracy report, the classification results are graded for credibility and converted into a vector format, and attribute information such as species names and areas is added to obtain a mangrove species distribution map. The area, density, and distribution form of each species are quantified from the mangrove species distribution map to generate a spatial structure parameter table. The normalized difference vegetation index and leaf area index are calculated using the near-infrared and visible light bands of the remote sensing images to construct a growth vitality evaluation index. Combining the species distribution map, spatial structure parameters, and growth vitality index, the mangrove carbon storage is calculated to obtain the quantified data of the carbon sink function. The spatial relationship between the mangroves and the coastline and their wave attenuation and sediment retention capabilities are analyzed to obtain the evaluation results of the coastal protection function. The species diversity index and niche overlap degree are calculated to obtain the data of the biodiversity maintenance function. Finally, these function evaluation data are integrated to form the mangrove ecological monitoring data.

[0019] In the embodiments of the present application, the mangroves are photographed from low altitudes at different angles by an unmanned aerial vehicle (UAV), and top-down photos with a camera lens tilt angle of approximately 90° and side-view photos with a camera lens tilt angle of approximately 30° are obtained, forming multi-angle image data of the mangroves. This breaks through the limitation that it is difficult for traditional manual surveys to enter the interior of the mangroves, greatly improving the spatial coverage and data collection efficiency of mangrove surveys. The projection coordinates are calculated using the position coordinates, photo length and width, and flight altitude information, and geometric correction processing is performed on the top-down photos to obtain corrected images superimposed on the remote sensing images, solving the problem of precise registration between UAV images and satellite remote sensing images and providing a basis for multi-source data fusion analysis. Based on the position and flight angle information of the side-view photos, a vector point layer is generated and symbolized to obtain a photo-taking direction layer superimposed on the remote sensing image, intuitively showing the relationship between the photo-taking direction and position of the side-view photos and facilitating the understanding of the side structural characteristics of the mangroves. It is particularly worth emphasizing that according to the corrected images, photo-taking direction layers, and remote sensing images, features are extracted and correlation relationships are established through the random forest algorithm to form a mangrove species interpretation signature library. The random forest algorithm shows excellent performance in feature selection and classification. It can not only process high-dimensional feature data, but also evaluate the importance of features and cope with the correlation between features, providing strong algorithm support for mangrove species identification. The algorithm has strong robustness to noise and outliers. By constructing a large number of decision trees and adopting a majority voting mechanism, the risk of overfitting is effectively reduced and the classification accuracy is improved. The interpretation signature library established in this way combines the advantages of the top-down, side-view, and remote sensing data sources, achieving high-precision identification of mangrove species. According to the established mangrove species interpretation signature library, training samples and validation samples are selected on the remote sensing image, and supervised classification processing of mangrove species is performed to generate a mangrove species distribution map, providing an accurate basis for the study of the spatial distribution of mangroves. Finally, based on the mangrove species distribution map, combined with spectral characteristics and spatial distribution information, the functions of the mangrove ecosystem are evaluated and analyzed to obtain mangrove ecological monitoring data.

[0020] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Use the UAV to plan a flight route above the mangroves, set the flight altitude for the outbound journey to approximately 50 meters to obtain top-down photos, and set the flight altitude for the return journey to approximately 30 meters to obtain side-view photos, forming a target acquisition flight route; (2) Configure the photo-taking parameters of the UAV. Set the camera lens tilt angle to 90° in the top-down photo-taking mode, with a photo-taking interval of approximately 40 meters, and set the camera lens tilt angle to 30° in the side-view photo-taking mode, with a photo-taking interval of approximately 30 meters; (3) Use the GPS receiver carried by the UAV to record the shooting position coordinates, flight altitude, and heading angle of each photo, generating photo metadata information; (4)Classify and organize the data of the top-down photos and side-view photos according to the photo metadata information, and construct a photo index database according to the shooting time sequence and spatial position relationship; (5)Evaluate the image quality of the top-down photos and side-view photos and perform image screening to form multi-angle mangrove image data; (6)Associate the multi-angle mangrove image data with the shooting position information to establish an image-position correspondence table.

[0021] Specifically, considering the shape, area of the mangrove distribution area and the surrounding terrain conditions, design a closed flight route through the DJI flight planning software, so that the drone can obtain top-down photos at a flight altitude of about 50 meters during the outbound flight, and obtain side-view photos at a flight altitude of about 30 meters during the return flight. This dual-altitude flight route design makes full use of a single flight mission, reduces battery consumption and operation time, and at the same time ensures the acquisition of image data from two different perspectives of the same area. In the photo parameter configuration session, precisely set the camera carried by the drone. In the top-down photo shooting mode, set the tilt angle of the camera lens to 90°, that is, the lens is perpendicular to the ground. This angle can obtain an image of the ground orthographic projection view, which is convenient for subsequent geometric correction and superposition with satellite remote sensing images. At the same time, set the photo shooting interval to about 40 meters to ensure that the images have an appropriate overlap. For the side-view photo shooting mode, adjust the tilt angle of the lens to 30°. This angle can capture the side structural features of the mangroves without being too parallel to the ground, resulting in an overly inclined perspective. The photo shooting interval in the side-view mode is set to about 30 meters, which is denser than the top-down mode to obtain more side detail information.

[0022] During the flight shooting process, the GPS receiver carried by the drone will automatically record the precise shooting information of each photo. These information include key parameters such as longitude and latitude coordinates (WGS84 coordinate system), altitude, flight altitude (relative to the take-off point), and heading angle (the angle between the flight direction and the due north direction). These data will be written into the photo file as EXIF information to form photo metadata information. Photo metadata is an important basis for subsequent geometric correction and spatial registration, and directly determines the accuracy of image spatial positioning.

[0023] After obtaining the photos, process and organize the photo metadata. Divide all the photos into a top-down photo group and a side-view photo group according to the photo EXIF information. Then sort each group of photos according to the shooting timestamp to establish a chronological relationship. Next, construct a spatial index based on the GPS coordinate information to determine the spatial position relationship of each photo. Through the organization in two dimensions of time and space, a structured photo index database is formed, which is convenient for subsequent efficient query and processing of photos in a specific area or time period. Photo quality assessment is a key step in data screening. Detect the blur degree, evaluate the exposure, and analyze the cloud coverage for each photo. The blur degree detection calculates the variance value of the Laplacian transform of the image, and the larger the value, the clearer the image; the exposure evaluation judges whether there is overexposure or underexposure by analyzing the distribution characteristics of the image histogram; the cloud coverage analysis determines by calculating the proportion of high-brightness areas in the image. According to the evaluation results, eliminate the photos with unqualified quality, such as the photos with a blur degree score lower than the threshold, serious exposure problems, or a cloud coverage rate too high, and finally screen out the high-quality mangrove multi-angle image data. Associate the screened mangrove multi-angle image data with its shooting location information to establish an image-location correspondence table. This correspondence table contains the unique identifier, file path, shooting time, GPS coordinates, altitude information, heading angle information, and photo type (top-down or side-view) of each image. This structured data association lays a foundation for subsequent geometric correction, spatial analysis, and feature extraction.

[0024] For example, in a mangrove ecological survey, a drone was used to collect data in a mangrove area of about 2 square kilometers. The flight path was planned in a zigzag layout with a flight path spacing of 80 meters and a total length of about 15 kilometers. During the flight, the altitude was 50 meters on the outbound journey, and 362 top-down photos were obtained; the altitude was 30 meters on the return journey, and 410 side-view photos were obtained. After extracting the metadata from the photo EXIF information, it was found that some photos had quality problems: 25 photos had a blur degree score lower than the set threshold of 200 (calculated based on the Laplacian transform variance), 18 photos had abnormal exposure (the histogram was concentrated in the gray value range of 0-50 or 205-255), and 15 photos had a cloud coverage rate exceeding 30%. After screening, finally, 328 top-down photos and 386 side-view photos were retained, constituting the high-quality mangrove multi-angle image data. Through the image-location correspondence table, each photo is associated with its spatial position.

[0025] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Extract the EXIF information of the top-down photos from the mangrove multi-angle image data to obtain a set of photo parameters including the photo shooting position coordinates, camera focal length, photo pixel width and height, and relative flight altitude; (2) Calculate the projected half-width and projected half-height of the overhead photo on the ground based on the camera focal length, photo pixel width and height, and relative flight height in the photo parameter set, and form ground projection size data; (3) Based on the shooting position coordinates and heading angle of the overhead photo, combined with the ground projection size data, calculate the ground projection coordinates of the four corner points of the overhead photo, and establish a photogrammetric control point coordinate table; (4) Use the photogrammetric control point coordinate table and the pixel coordinates of the overhead photo for correlation analysis, calculate the six-parameter geometric correction parameter set, and construct a spatial transformation matrix; (5) Apply the spatial transformation matrix to the overhead photo, perform affine transformation and resampling processing, and generate a geographically corrected overhead photo; (6) Import the geographically corrected overhead photo into ArcGIS software, perform geographic coordinate registration and overlay operations with the remote sensing image, and obtain the corrected image overlaid on the remote sensing image.

[0026] Specifically, extract the EXIF information of the overhead photo from the mangrove multi-angle image data. These information are metadata automatically recorded by the digital camera during shooting and embedded in the image file. These EXIF information can be read through the exifread library in Python or professional image processing software to obtain a photo parameter set containing the shooting position coordinates (longitude, latitude) of the photo, the camera focal length (in millimeters), the photo pixel width and height (number of pixels), and the relative flight height (height relative to the take-off point, in meters). These parameters are the basic data for subsequent geometric correction. Based on the camera focal length, photo pixel width and height, and relative flight height in the extracted photo parameter set, calculate the projection range of the overhead photo on the ground. According to the formula in the disclosure, the projected half-width and projected half-height are calculated as follows: ; Among them, represents the actual scale factor, which is related to the flight height; represents the photo pixel width; represents the photo pixel height; represents the camera focal length. Through these formulas, the camera parameters are converted into the ground coverage range, forming the ground projection size data. This data represents the actual ground range that a single overhead photo taken by the UAV can cover.

[0027] The third step is to calculate the ground projection coordinates of the four corner points of the overhead photo based on the shooting position coordinates and heading angle of the overhead photo, combined with the ground projection size data. The heading angle refers to the angle between the UAV flight direction and the due north direction. The coordinates of the four corner points are calculated as follows: ; Among them, is the ground coordinate of the center point of the photo, are the ground coordinates of the upper left corner, upper right corner, lower right corner, and lower left corner respectively. Through these calculations, a photogrammetric control point coordinate table is established. This table contains the exact positions of the four corners of the image in the ground coordinate system, as shown in Figure 2 which is the schematic diagram of geometric correction of the UAV overhead photo in the embodiment of the present application; Establish a corresponding relationship between the photogrammetric control point coordinate table and the pixel coordinates of the overhead photo. The photo pixel coordinates are a two-dimensional coordinate system with pixels as the unit. The origin is usually at the upper left corner of the photo, the X-axis is to the right, and the Y-axis is downward. Calculate the geometric correction parameters through the six-parameter model: ; Among them, are the pixel width and height of the photo respectively, A, B, D, and E are scaling and rotation parameters, and C and F are translation parameters. These six parameters form a spatial transformation matrix, which defines the mapping relationship from the photo pixel coordinates to the ground coordinates.

[0028] Apply the calculated spatial transformation matrix to the overhead photo for affine transformation and resampling processing. Affine transformation is a mapping transformation that preserves points, lines, and planes, and it is achieved through matrix multiplication. For each pixel point (i, j) in the photo, calculate its corresponding ground coordinate through the following formula : ; Through inverse mapping and resampling, convert the distorted original photo into a corrected image that conforms to the ground coordinate system. Resampling methods include nearest neighbor interpolation, bilinear interpolation, and cubic convolution interpolation, etc. Select a suitable method according to the accuracy requirements. Generate a geographically corrected overhead photo, and this photo is consistent with the actual ground position in terms of spatial position. Import the geographically corrected overhead photo into the ArcGIS software for georeferencing and overlay operations with the remote sensing image. ArcGIS is a professional geographic information system software with powerful spatial analysis and visualization functions. In ArcGIS, load the existing remote sensing image as the base map, and then import the corrected overhead photo to ensure that both use the same coordinate system (such as WGS84 or UTM). By adjusting display parameters such as transparency and contrast, make the overhead photo and the remote sensing image blend naturally to form a corrected image superimposed on the remote sensing image, as shown in Figure 3 which is the overlay diagram of the UAV overhead photo after geometric correction and the remote sensing image in the embodiment of the present application.

[0029] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Extract GPS coordinates, flight heading angle, and shooting timestamp from the EXIF information of the side-view photos to form a side-view photo position and angle information set; Perform spatial filtering on the side-view photo position and angle information set, and eliminate data points with repeated positions or excessive errors to obtain an optimized position and angle data set; Use the GPS coordinate information in the optimized position and angle data set to create point features in ArcGIS software and construct a side-view photo position point layer of the mangrove forest; According to the flight heading angle data in the optimized position and angle data set, assign direction attributes to the side-view photo position point layer of the mangrove forest to generate a point layer with direction attributes; Based on the point layer with direction attributes, perform arrow symbolization design, where the arrow direction represents the shooting direction, and the arrow size is proportional to the flight height, to form a side-view photo direction symbol layer of the mangrove forest; Spatially overlay the side-view photo direction symbol layer of the mangrove forest and the remote sensing image in the same coordinate system to obtain a shooting direction layer overlaid on the remote sensing image.

[0030] Specifically, different from the geometric correction of the top-view photos, the side-view photos are mainly used to represent the shooting position and direction, providing information support for the interpretation of mangrove species from a side perspective. Extract key data from the EXIF information of the side-view photos. The EXIF information of the side-view photos contains rich metadata. Through professional EXIF reading tools (such as ExifTool or the exifread library in Python), GPS coordinates (longitude, latitude), flight heading angle (representing the angle between the UAV flight direction and the true north direction, in degrees), and shooting timestamp (in the format of YYYY:MM:DD HH:MM:SS) can be extracted. These data are combined to form a side-view photo position and angle information set, which is a data set representing when, where, and at what angle each side-view photo was taken.

[0031] Spatial filtering of the extracted lateral photo position and angle information set is a necessary step to ensure data quality. Due to GPS positioning errors, or the drone may take multiple photos at certain positions in a short period of time, there may be data points with duplicate positions or excessive positioning errors in the data set. Spatial filtering adopts two main strategies: one is duplicate removal of adjacent points, by calculating the spherical distance between every two GPS points, removing duplicate points with a distance less than the distance threshold (usually set to 5 - 10 meters); the other is outlier detection, by calculating the average distance from each point to its K nearest neighbors (usually K = 5), comparing this value with the deviation of the overall distribution, and removing outliers with a deviation degree exceeding the set deviation threshold (usually set to 3 times the standard deviation). After these two steps of processing, an optimized position and angle data set is obtained, which contains higher-quality and more evenly distributed lateral photo position and angle information.

[0032] Using the GPS coordinate information in the optimized position and angle data set, the next step is to create point features in ArcGIS software. ArcGIS is a professional geographic information system software that provides powerful geographic data processing and visualization functions. Through the "XY Data Import" function of ArcGIS, import the GPS coordinate (longitude, latitude) data in CSV or Excel format, set an appropriate coordinate system (usually WGS84), and create an initial point layer. Then, through the "Feature Transformation" tool, convert the temporary layer into a permanent geodatabase feature class to form a lateral photo position point layer of mangroves. This layer accurately represents the shooting position of each lateral photo in the geospatial space.

[0033] According to the flight heading angle data in the optimized position and angle data set, it is necessary to assign a direction attribute to the lateral photo position point layer of mangroves. In ArcGIS, first, through the "Add Field" function, create a new field named "Direction" in the attribute table, with a data type of floating-point. Then use the "Calculate Field" function to assign the flight heading angle data to this field. For lateral photos with a camera lens tilt angle of about 30°, there is a certain angle difference between the shooting direction and the flight direction, and the angle needs to be adjusted according to the flight orientation of the drone and the camera installation position. The adjusted angle value The calculation formula is: ; where, is the flight heading angle, is the camera installation offset angle (usually 90° for cameras installed on the right side and -90° for those installed on the left side), and mod represents the modulo operation to ensure that the angle value is within the range of 0 - 360°. Through this processing, a point layer with direction attributes is generated, which not only contains position information but also the accurate shooting direction of each shooting point.

[0034] Based on the point layer with direction attributes, the arrow symbolization design is carried out next. In ArcGIS, select the point layer, open the "Symbol Properties" dialog box, and select the "Arrow Marker" symbol type. Set the color, line width, and head style of the arrow, and set the rotation angle field of the arrow to the previously calculated "Direction" field to ensure that the arrow orientation is consistent with the photographing direction. To make the symbolization effect more intuitive, the arrow size can also be adjusted according to the flight altitude. The calculation formula is: ; Wherein, is the final arrow size, is the reference arrow size (usually set to 2 - 5 points), is the actual flight altitude, is the reference flight altitude (usually taken as 30 meters). Through such a design, a mangrove side-view photo direction symbol layer is formed. This layer intuitively shows the shooting position and direction of the side-view photo, and the arrow size also reflects the flight altitude information.

[0035] The last step is to perform a spatial overlay of the mangrove side-view photo direction symbol layer and the remote sensing image in the same coordinate system. First, load the existing remote sensing image as the base map in ArcGIS, ensure that its coordinate system is consistent with the direction symbol layer, and use the "Projection" tool for coordinate system conversion if necessary. Then load the mangrove side-view photo direction symbol layer and adjust the layer order to ensure that the symbol layer is above the remote sensing image. By setting appropriate display parameters such as transparency and contrast, the two layers are naturally integrated to form an intuitive visual effect. Finally, a photographing direction layer overlaid on the remote sensing image is obtained, which clearly shows the spatial relationship between the shooting position, direction of the side-view photo, and the distribution of mangroves.

[0036] Taking a specific mangrove survey example to illustrate: In the UAV survey conducted in a certain mangrove reserve, 285 side-view photos were obtained using the UAV. The GPS coordinates, flight heading angle, and shooting timestamp were extracted from the EXIF information to form an initial set of side-view photo position and angle information. Through spatial filtering, it was found that the mutual distance of 23 points was less than the set threshold of 8 meters, which were position duplicate points; another 7 points had a positioning error exceeding 3 times the standard deviation and were identified as abnormal points. After removing these points, 255 effective data points were obtained. These GPS coordinates were imported into ArcGIS to create a layer of side-view photo position points of the mangroves. Considering that the camera was installed on the right side of the UAV, a 90° offset needed to be added to the flight heading angle. For example, for a point with a flight heading angle of 45°, the photo-taking direction angle was (45° + 90°) = 135°, indicating that the camera was facing the southeast direction. According to the calculated direction angles and flight altitudes (ranging from 25 to 35 meters), arrow symbols of different sizes and directions were designed to visually display the photo-taking directions of each shooting point. Finally, the symbol layer was overlaid with the Sentinel-2 satellite image, clearly showing the spatial relationship between the side-view photo shooting points and the mangrove distribution, providing important side-view information support for subsequent mangrove species interpretation.

[0037] In a specific embodiment, the process of performing step S104 may specifically include the following steps: (1) In the overlapping area of the corrected image, the photo-taking direction layer, and the remote sensing image, representative sample areas of different mangrove species are selected to create a set of mangrove species sample points; (2) For each sample point in the set of mangrove species sample points, texture features, color features, and shape features are extracted from the corrected image, and spectral features and spatial features are extracted from the remote sensing image to form a set of mangrove species feature vectors; (3) Through recursive feature elimination, importance ranking is performed on the set of mangrove species feature vectors, key features with high contribution are retained, and an optimized feature subset is constructed; (4) Based on the optimized feature subset, a mangrove species classifier is trained using the random forest algorithm, where the number of decision trees is set to 500, the minimum number of samples for each node is 5, and the random selection ratio of features is 0.7; (5) Classification rules are extracted from the set of decision trees generated by the random forest algorithm, including the splitting thresholds of each feature and the species discrimination path, to generate a set of mangrove species discrimination rules; (6) An association mapping relationship is established between the set of mangrove species discrimination rules and the corresponding features of the side-view photos, top-view photos, and remote sensing images to form a mangrove species interpretation symbol library.

[0038] Specifically, representative sample areas of different mangrove species are selected in the overlapping area of the corrected image, the photographing direction layer, and the remote sensing image. Considering the spatial distribution of the three layers comprehensively, their common coverage area is determined, which is generally achieved by using the "layer overlay analysis" function of GIS software. In the overlapping area, according to expert knowledge or existing ground survey data, typical sample areas are selected for the main mangrove species (such as Kandelia obovata, Rhizophora stylosa, Avicennia marina, etc.). Each sample area should meet the requirements of moderate area (usually 5×5 meters to 10×10 meters), high species purity, and uniform spatial distribution. After determining the location of the sample area, a point sampling tool is used to set multiple sampling points in each sample area to form a mangrove species sample point set, and the coordinate position and the category of the species to which each point belongs are recorded.

[0039] For each sample point in the mangrove species sample point set, multi-source features need to be extracted. The features extracted from the corrected image mainly include three categories: Texture features are statistics that describe the pixel gray-scale distribution pattern in a local area of the image. Commonly used ones are gray-level co-occurrence matrix (GLCM) features, including contrast, correlation, entropy, etc.; Color features describe the color information of the area around the sample point, which is usually obtained by calculating statistics such as the mean, variance, and skewness of each channel in the RGB or HSV color space; Shape features describe the geometric characteristics of the mangrove canopy, such as area, perimeter, compactness, etc. The features extracted from the remote sensing image focus on spectral features and spatial features: Spectral features include the reflectance of each band and their combinations. Commonly used ones are the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and soil-adjusted vegetation index (SAVI), etc.; Spatial features include topographic features (such as elevation, slope) and surrounding environment features (such as the distance from water bodies). For each sample point, these features are organized into a multi-dimensional vector to form a mangrove species feature vector set.

[0040] Since the number of extracted features is large, some features may be redundant or have a low correlation with the target species classification, and feature selection is required. Recursive feature elimination (RFE) is an effective feature selection method. Its basic idea is to iteratively train a base classifier (such as a random forest or a support vector machine), and each time the least important feature is deleted until the preset number of features is reached. In mangrove feature selection, RFE first trains a random forest model using all features and calculates the importance score of each feature (usually based on the Gini index decrease or information gain). Then, the features are removed in ascending order of importance. After removing one or a group of features each time, the model is retrained and the performance is evaluated. By comparing the classification performance (such as accuracy, F1 score) of different feature subsets, the optimal number of features is determined, and the key features with higher importance rankings are retained to construct an optimized feature subset.

[0041] Based on the optimized feature subset, a mangrove species classifier is trained using the random forest algorithm. Random forest is an ensemble learning method composed of multiple decision trees, which conducts classification prediction through majority voting. During the training process, the number of decision trees is set to 500, which is large enough to ensure the stability of the model; the minimum number of samples per node is set to 5 to prevent overfitting while maintaining sensitivity to minority classes; the proportion of randomly selected features is set to 0.7, that is, 70% of the features are randomly selected to participate in the decision-making when each node is split. The Bootstrap sampling method is adopted during training, that is, each decision tree uses a training set obtained by sampling with replacement, and approximately one-third of the samples are not selected. These samples are called "Out-of-Bag samples" and are used to evaluate the model performance.

[0042] Extracting classification rules from the trained random forest algorithm is a key step in establishing an interpretable interpretation mark library. Although the random forest is regarded as a "black box" model, rules can be extracted by analyzing its internal decision tree structure. The specific approach is to traverse each decision tree and record the complete path from the root node to the leaf node. Each path represents a classification rule. For example, for a path judged as "Kandelia obovata", it may contain conditions such as "NDVI > 0.65 AND texture entropy < 1.2 AND red band reflectance < 0.15". These conditions include feature names (such as NDVI), comparison operators (>、<、=), and splitting thresholds (0.65). By counting the rules in all decision trees, rules with high occurrence frequencies and high accuracies are selected to form a mangrove species discrimination rule set.

[0043] An association mapping relationship is established between the mangrove species discrimination rule set and multi-source image features. For each discrimination rule, it is necessary to trace back which data sources (side-looking photos, overhead photos, or remote sensing images) the features it uses come from, and record the feature extraction methods and parameter settings. For example, for texture features, record its window size, calculation method, etc.; for spectral indices, record its band combination formula, etc. This association mapping enables the rule to not only tell "what it is", but also tell "why" and "how to judge". In this way, a ternary association mangrove species interpretation mark library containing species feature descriptions, discrimination rules, and image instances is formed. This mark library combines expert experience with machine learning, having both strict quantitative criteria and intuitive visual interpretations.

[0044] Illustrate this process with a specific case: In the study of a certain mangrove reserve, 120 representative sample plots were selected within the overlapping area of the corrected image, the photographing direction layer, and the remote sensing image, including Kandelia obovata (40), Rhizophora stylosa (35), Avicennia marina (30), and other species (15). For the 324 sample points set in each sample plot, 18 texture features (such as GLCM contrast, homogeneity, correlation, etc.), 12 color features (such as the mean and standard deviation of each RGB channel, etc.), and 8 shape features (such as crown width, perimeter ratio, etc.) were extracted from the corrected image; 10 spectral features (such as NDVI, EVI, red edge position, etc.) and 7 spatial features (such as elevation, distance to water body, etc.) were extracted from the remote sensing image, totaling 55 feature dimensions. Through the recursive feature elimination method, features were removed in sequence according to the feature importance scores, and finally 22 optimal features were determined, including NDVI, texture contrast, NIR band reflectance, etc. The random forest classifier (500 decision trees, minimum node sample number 5, feature selection ratio 0.7) constructed based on these features achieved an accuracy of 94.2% on the training set. More than 1000 initial rules were extracted from the decision trees. After frequency screening and accuracy verification, 178 high-quality rules were retained, forming a detailed mangrove species discrimination rule set. These rules established a clear correlation with multi-source image features. For example, rules like "If NDVI > 0.72 and texture contrast < 1.5 and NIR reflectance > 0.45, then it is determined to be Kandelia obovata" associated both the spectral features of the remote sensing image and the texture features of the corrected image, forming a complete mangrove species interpretation symbol library. This symbol library not only provides high-accuracy automatic classification ability but also maintains good interpretability, facilitating experts to understand and verify the classification results.

[0045] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) According to the species feature descriptions in the mangrove species interpretation symbol library, use stratified sampling on the remote sensing image to select representative areas of each mangrove species, and generate a mangrove training sample set and a validation sample set; (2) For each sample in the mangrove training sample set, extract the normalized difference vegetation index, red edge position index, and texture complexity index as mangrove feature parameters, and construct a mangrove sample feature matrix; (3) Perform mangrove species supervised classification on the remote sensing image through the mangrove sample feature matrix, divide the distribution ranges of different mangrove species, and form a preliminary mangrove classification result; (4) Utilize the spectral similarity and spatial continuity features in the preliminary mangrove classification result to perform mangrove species patch merging and boundary optimization processing, eliminate isolated pixels and boundary noise, and obtain an optimized mangrove classification result; (5) Evaluate the accuracy of the optimized mangrove classification results based on the mangrove verification sample set, calculate the overall accuracy, Kappa coefficient, and the classification accuracy of each species, and generate a mangrove classification accuracy report; (6) According to the accuracy indicators in the mangrove classification accuracy report, classify the credibility of the optimized mangrove classification results and convert them into vector format, add species name, area, distribution characteristic attribute information, and generate a mangrove species distribution map.

[0046] Specifically, sample points are selected on the remote sensing image. The stratified sampling method is adopted instead of simple random sampling, mainly considering the obvious environmental gradient and community characteristics in the spatial distribution of mangrove species. Stratified sampling divides the entire study area into several relatively homogeneous strata according to factors such as different species types, growth environments, and altitudes, and then random sampling is carried out within each stratum to ensure that the samples can represent the mangrove characteristics under various environmental conditions. Usually, the total sample size is divided into a training sample set and a verification sample set in a ratio of 7:3. The training samples are used to construct the classification model, and the verification samples are used to evaluate the classification accuracy. For the selected mangrove training sample set, key characteristic parameters need to be extracted to characterize the spectral and textural properties of different mangrove species. The normalized difference vegetation index (NDVI) is one of the most commonly used vegetation indices, which is calculated by the reflectance difference between the near-infrared band and the red band, and reflects the biomass and vitality of vegetation. The red edge position index (REP) is the inflection wavelength position of the rapid upward region from the red light to the near-infrared in the vegetation spectral curve, which is sensitive to the chlorophyll content and physiological state of vegetation and can effectively distinguish different mangrove species. The texture complexity index is a texture feature calculated based on the gray-level co-occurrence matrix (GLCM), which quantifies the complexity of pixel value changes in the local area of the image and reflects the complexity of the mangrove canopy structure. These three types of indices combine spectral information and spatial structure information, calculate for each sample point and organize them into a feature matrix. The rows of the matrix represent different sample points, and the columns represent different characteristic parameters.

[0047] Using the established mangrove sample feature matrix, supervised classification of mangrove species is carried out on the entire remote sensing image. In the classification process, algorithms such as Maximum Likelihood Classification, Support Vector Machine (SVM), or Random Forest are used to compare each pixel in the remote sensing image with the sample feature matrix, and according to the principle of feature similarity, it is divided into different mangrove species categories. The Maximum Likelihood Method assumes that the feature distribution of each category conforms to a multivariate normal distribution, calculates the probability that a pixel belongs to each category, and selects the category with the highest probability as the classification result of the pixel. The Support Vector Machine realizes classification by constructing the optimal separation hyperplane in the high-dimensional feature space. Each pixel in the remote sensing image is assigned a specific mangrove species category to form the preliminary mangrove classification result. The preliminary classification result usually has problems such as isolated pixels and boundary noise, which need to be post-processed and optimized. The post-processing first analyzes based on spectral similarity, calculates the spectral distance between each pixel and its surrounding pixels. If the spectral distance between the isolated pixel and the surrounding dominant category is less than the set threshold, it is merged into the surrounding dominant category. At the same time, the spatial continuity feature is used for patch merging and boundary smoothing. The Majority Filter is used to perform window filtering on the classification result. The window size is usually set to 3×3 or 5×5 pixels, and the category of the central pixel is determined by the most common category within the window. For isolated patches smaller than the specified area threshold (usually 4 - 9 pixels), they are merged into the largest and most similar surrounding patches. Boundary optimization uses morphological processing methods. Through dilation and erosion operations, the category boundaries are smoothed to generate a more natural distribution law of mangrove optimized classification results.

[0048] The accuracy of the optimized classification result is evaluated based on the reserved mangrove validation sample set. The accuracy evaluation first constructs a Confusion Matrix, where the rows of the matrix represent the classification results and the columns represent the reference truth values (i.e., the actual categories of the validation samples). A series of statistical indicators in the matrix are calculated to evaluate the classification accuracy, including Overall Accuracy, Kappa coefficient, Producer's Accuracy, and User's Accuracy. The Overall Accuracy is the number of correctly classified samples divided by the total number of samples; the Kappa coefficient takes into account the impact of random classification and reflects the degree of consistency between the classification result and the reference data; the Producer's Accuracy represents the proportion of a certain category that is correctly identified; the User's Accuracy represents the proportion of the pixels classified as a certain category that actually belong to that category. These indicators comprehensively form a mangrove classification accuracy report to provide a quantitative assessment of the reliability of the results.

[0049] Optimize and output the results according to the mangrove classification accuracy report. First, classify the credibility of the results based on classification accuracy metrics. Usually, the classification accuracy is divided into three levels: high, medium, and low, with corresponding thresholds of 85%, 70%, and 60% respectively. Then, convert the raster-format classification results into vector format. During the conversion process, set an area threshold to remove too small patches and perform boundary smoothing. For each vector polygon, add attribute information, including species name (based on the classification results), area (obtained by calculating the polygon area), distribution characteristics (such as the average elevation and slope of the distribution area), etc. The resulting mangrove species distribution map contains both spatial distribution information and rich attribute information, providing basic data for subsequent ecosystem function assessment.

[0050] For example: In a study of a mangrove reserve, 430 sample points were selected on the remote sensing image based on the mangrove species interpretation sign library, including 150 points of Kandelia obovata, 120 points of Rhizophora stylosa, 90 points of Avicennia marina, and 70 points of other species. They were divided into 300 training samples and 130 validation samples according to a 7:3 ratio. For each training sample, the NDVI value (reflecting biomass), REP value (reflecting chlorophyll content), and GLCM-based texture complexity index (reflecting canopy structure) were extracted to construct a 300×3 feature matrix. The maximum likelihood method was used to classify the remote sensing image. In the preliminary results, the total mangrove area was approximately 250 hectares, but there were a large number of isolated pixels and jagged boundaries. Through the majority filter with a 3×3 window and patch merging with a minimum area threshold of 4 pixels, the isolated pixels and small patches were eliminated, and the optimized area was 243 hectares, with a more continuous and reasonable distribution. Using 130 validation sample points to evaluate the accuracy, the overall accuracy reached 87.6%, and the Kappa coefficient was 0.84. Among them, the user accuracy of Kandelia obovata was 92.1%, Rhizophora stylosa was 86.3%, and Avicennia marina was 88.7%, indicating that the classification results were reliable. Finally, the classification results were vectorized to generate a distribution map containing 48 mangrove patches, and each patch contained attribute information such as species type and area, providing an accurate spatial information basis for mangrove ecological monitoring and protection management.

[0051] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Based on the mangrove species distribution map, extract the spatial distribution patterns of different mangrove species, quantify the area, density, and distribution form of each species, and generate a mangrove spatial structure parameter table; (2) Use the near-infrared band and visible light band data in the remote sensing image to calculate the normalized difference vegetation index and leaf area index of the mangrove area, and construct a mangrove growth vitality evaluation index; (3) Combine the mangrove species distribution map, the mangrove spatial structure parameter table, and the mangrove growth vitality evaluation indicators, and use the biomass conversion coefficient to calculate the mangrove carbon storage, generating the quantified data of the mangrove carbon sink function; (4) Extract the spatial relationship between the mangroves and the coastline from the mangrove species distribution map and the mangrove spatial structure parameter table, and combine the tidal level and terrain data to analyze the wave attenuation and sediment retention capabilities of the mangroves, forming the evaluation results of the mangrove coastal protection function; (5) Analyze the species composition in the mangrove species distribution map and the distribution pattern in the mangrove spatial structure parameter table, calculate the species diversity index and the niche overlap degree, and obtain the data of the mangrove biodiversity maintenance function; (6) Integrate the quantified data of the mangrove carbon sink function, the evaluation results of the mangrove coastal protection function, and the data of the mangrove biodiversity maintenance function to obtain the mangrove ecological monitoring data.

[0052] Specifically, based on the obtained mangrove species distribution map, quantitatively analyze the spatial distribution patterns of different species. In the GIS environment, conduct spatial statistical analysis on the mangrove species distribution map, and calculate the area (obtained through the vector data area calculation function), density (the number of individuals per unit area), and distribution form (described by the patch shape index, aggregation index, etc.) of each species. For the area, calculate the total area of each mangrove species and its proportion in the entire study area; for the density, use the high-resolution image data obtained by the drone, combined with the canopy recognition algorithm, to calculate the number of tree individuals per unit area; for the distribution form, use landscape pattern indices, such as the patch size variation coefficient, patch shape index, and aggregation index, etc., to quantify the spatial distribution characteristics of the mangroves. These parameters comprehensively form the mangrove spatial structure parameter table, which details the spatial distribution characteristics of each mangrove species.

[0053] Calculate the mangrove growth vitality indicators using the spectral information in the remote sensing images. The Normalized Difference Vegetation Index (NDVI) is an important indicator to characterize the vegetation growth status, and it is calculated from the reflectance of the near-infrared band (NIR) and the red band (R): NDVI = (NIR - R) / (NIR + R). The higher the NDVI value, the higher the vegetation coverage and biomass. The Leaf Area Index (LAI) represents the leaf area per unit ground area and is an important parameter to describe the vegetation canopy structure. LAI can be converted from NDVI through an empirical model or directly calculated after three-dimensional reconstruction using the multi-view images obtained by the drone. These two indicators together constitute the mangrove growth vitality evaluation indicators, reflecting the growth status and photosynthetic ability of the mangroves.

[0054] The third step is the calculation of the mangrove carbon storage, which comprehensively utilizes the three datasets obtained previously: the mangrove species distribution map, the mangrove spatial structure parameter table, and the mangrove growth vitality evaluation indicators. The carbon storage calculation is divided into above-ground and below-ground parts. For the calculation of the above-ground carbon storage, a biomass conversion equation based on species is used. For different mangrove species, the corresponding conversion equation is selected to convert parameters such as tree height and diameter at breast height into biomass, and then multiplied by the carbon content coefficient (usually 0.5) to obtain the carbon storage. The below-ground carbon storage is estimated according to the root-shoot ratio, and different mangrove species have different root-shoot ratio values. In the UAV data processing, the tree height can be extracted from LiDAR data or three-dimensional point cloud models, and the diameter at breast height can be estimated through the combination of tree height and crown width using species-specific correlation equations. In this way, the carbon storage is calculated for each mangrove patch in the study area, generating the quantitative data of the mangrove carbon sink function.

[0055] Next, the spatial relationship between the mangroves and the coastline is extracted from the mangrove species distribution map and the spatial structure parameter table to evaluate their coastal protection function. Among them, the coastline data is obtained (which can be obtained from satellite images or existing geographical databases), and then the distance, width, and continuity between the mangrove belt and the coastline are calculated. The wave attenuation ability of the mangroves mainly depends on the belt width, density, and species composition, and can be calculated through an empirical model: wave energy attenuation rate = T(belt width, tree density, root system complexity). The sediment retention ability is related to the root development degree and the spatial structure of the mangroves, and different mangrove species have different root characteristics and soil fixation abilities. Combining the tide level and terrain data, the wave action scenarios under different tide level conditions are simulated, and the protection efficiency index of the mangroves is calculated to form the evaluation results of the mangrove coastal protection function.

[0056] Based on the species composition in the mangrove species distribution map and the distribution pattern in the spatial structure parameter table, various biodiversity indices are calculated. Commonly used ones include the Shannon-Wiener diversity index, Simpson diversity index, Pielou evenness index, etc. The calculation formula for the Shannon-Wiener index is: H' = -∑(Po × lnPo), where Po is the relative abundance of species o. The niche overlap degree describes the similarity of resource utilization between different species and is calculated through the Pianka index or Morisita index. High species diversity and moderate niche overlap usually indicate the stability of ecosystem functions and strong anti-interference ability. These indicators comprehensively form the data on the mangrove biodiversity maintenance function. Integrate the quantified data of the mangrove carbon sequestration function, the evaluation results of the coastal protection function, and the data on the biodiversity maintenance function to form comprehensive mangrove ecological monitoring data. The integration process uses a multi-criteria evaluation method to standardize different functional indicators, unifying each indicator value within the range of 0-1 for easy comparison. Then set the weights of each function, and obtain the comprehensive evaluation value through weighted summation or the analytic hierarchy process (AHP). This comprehensive data set not only contains the original evaluation data of each function but also includes the overall evaluation of the mangrove ecosystem service function, providing a scientific basis for formulating mangrove protection strategies and determining ecological compensation standards.

[0057] The above describes the method for extracting mangrove ecological information based on drones in the embodiments of the present application. Next, the system for extracting mangrove ecological information based on drones in the embodiments of the present application will be described. Please refer to Figure 4 One embodiment of the system for extracting mangrove ecological information based on drones in the embodiments of the present application includes: An acquisition module, configured to take low-altitude photos of the mangrove from different angles through a drone, obtain a top-down photo with a camera lens tilt angle of about 90° and a side-view photo with a camera lens tilt angle of about 30°, and obtain mangrove multi-angle image data; A calculation module, configured to calculate projection coordinates based on the mangrove multi-angle image data, using position coordinates, photo length and width, and flight height information, and perform geometric correction processing on the top-down photo to obtain a corrected image superimposed on the remote sensing image; A processing module, configured to generate a vector point layer and perform symbolization processing based on the position and flight angle information of the side-view photo to obtain a photo shooting direction layer superimposed on the remote sensing image; A building module, configured to extract features and establish an association relationship through the random forest algorithm based on the corrected image, the photo shooting direction layer, and the remote sensing image to form a mangrove species interpretation mark library; A selection module, configured to select training samples and validation samples on the remote sensing image according to the mangrove species interpretation signature library, perform supervised classification processing of mangrove species, and generate a mangrove species distribution map; An analysis module, configured to evaluate and analyze the functions of the mangrove ecosystem based on the mangrove species distribution map, combined with spectral characteristics and spatial distribution information, to obtain mangrove ecological monitoring data.

[0058] Through the collaborative cooperation of the above-mentioned various components, low-altitude photos of the mangroves are taken from different angles by an unmanned aerial vehicle (UAV), obtaining a top-down photo with a camera lens tilt angle of approximately 90° and a side-view photo with a camera lens tilt angle of approximately 30°, forming multi-angle image data of the mangroves, breaking through the limitation that it is difficult for traditional manual surveys to enter the interior of the mangroves, and greatly improving the spatial coverage and data collection efficiency of mangrove surveys. The projection coordinates are calculated using the position coordinates, photo length and width, and flight altitude information, and geometric correction processing is performed on the top-down photo to obtain a corrected image superimposed on the remote sensing image, solving the problem of accurate registration between UAV images and satellite remote sensing images, and providing a basis for multi-source data fusion analysis. Based on the position and flight angle information of the side-view photo, a vector point layer is generated and symbolized to obtain a photo-taking direction layer superimposed on the remote sensing image, intuitively showing the relationship between the photo-taking direction and position of the side-view photo, and facilitating the understanding of the side structure characteristics of the mangroves. It is particularly worth emphasizing that, based on the corrected image, photo-taking direction layer, and remote sensing image, this solution extracts features and establishes correlation relationships through the random forest algorithm to form a mangrove species interpretation signature library. The random forest algorithm exhibits excellent performance in feature selection and classification. It can not only process high-dimensional feature data, but also evaluate the importance of features and handle the correlation between features, providing strong algorithm support for mangrove species identification; this algorithm has strong robustness to noise and outliers. By constructing a large number of decision trees and adopting a majority voting mechanism, the risk of overfitting is effectively reduced, and the classification accuracy is improved. The interpretation signature library established in this way combines the advantages of the top-down view, side-view, and remote sensing data sources, realizing high-precision identification of mangrove species. According to the established mangrove species interpretation signature library, training samples and validation samples are selected on the remote sensing image, supervised classification processing of mangrove species is performed, and a mangrove species distribution map is generated, providing an accurate basis for the study of the spatial distribution of mangroves. Finally, based on the mangrove species distribution map, combined with spectral characteristics and spatial distribution information, the functions of the mangrove ecosystem are evaluated and analyzed to obtain mangrove ecological monitoring data.

[0059] Refer to Figure 5 , in an embodiment of the present invention, a computer device is further provided. This computer device can be a server, and its internal structure can be as Figure 5As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0060] Those skilled in the art can understand that Figure 5 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0061] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0062] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0063] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, systems, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0064] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0065] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for extracting mangrove ecological information based on drones, characterized in that: The method for extracting mangrove ecological information based on drone includes: Use drones to take low-altitude photos of mangroves at different angles, obtain overhead photos with a camera lens tilt angle of about 90° and side photos with a camera lens tilt angle of about 30°, and obtain multi-angle image data of mangroves; According to the multi-angle image data of the mangrove forest, projection coordinates are calculated using the location coordinates, length and width of the photo and flight altitude information, and geometric correction processing is performed on the overhead photo to obtain a corrected image superimposed on the remote sensing image; Based on the position and flight angle information of the side view photo, a vector point layer is generated and symbolized to obtain a photographing direction layer superimposed on the remote sensing image; Based on the calibrated image, the photographing direction layer and the remote sensing image, features are extracted and association relationships are established through a random forest algorithm to form a mangrove species interpretation sign library; According to the mangrove species interpretation sign library, training samples and verification samples are selected on the remote sensing image to perform supervised classification of mangrove species and generate a mangrove species distribution map; Based on the mangrove species distribution map, combined with spectral characteristics and spatial distribution information, the functions of the mangrove ecosystem are evaluated and analyzed to obtain mangrove ecological monitoring data.

2. The method for extracting mangrove ecological information based on drone according to claim 1 is characterized in that: The drone is used to take low-altitude photos of the mangroves at different angles to obtain overhead photos with a camera lens tilted at an angle of about 90° and side photos with a camera lens tilted at an angle of about 30°, thereby obtaining multi-angle image data of the mangroves, including: Use drones to plan routes above the mangroves, set the outbound flight altitude to about 50 meters to obtain overhead photos, and the return flight altitude to about 30 meters to obtain side-view photos, forming a target collection route; The camera parameters of the drone are configured. In the top-view camera mode, the camera lens tilt angle is set to 90°, and the photo interval is about 40 meters. In the side-view camera mode, the camera lens tilt angle is set to 30°, and the photo interval is about 30 meters. The GPS receiver carried by the drone records the shooting location coordinates, flight altitude and heading angle of each photo, and generates photo metadata information; Classify and sort the overhead view photo and the side view photo according to the photo metadata information, and construct a photo index database according to the shooting time sequence and spatial position relationship; Performing image quality assessment and image screening on the overhead view photo and the side view photo to form multi-angle image data of mangroves; The multi-angle image data of the mangrove forest is associated with the shooting location information to establish an image-location correspondence table.

3. The method for extracting mangrove ecological information based on drone according to claim 1 is characterized in that: The method of calculating projection coordinates based on the multi-angle image data of the mangrove forest using position coordinates, length and width of the photo and flight altitude information, and performing geometric correction processing on the overhead photo to obtain a corrected image superimposed on the remote sensing image includes: Extract the EXIF ​​information of the bird's-eye view photo from the multi-angle image data of the mangrove forest to obtain a photo parameter set including the coordinates of the photo shooting location, the camera focal length, the photo pixel width and height, and the relative flight altitude; Calculate the projection half width and half height of the bird's-eye view photo on the ground according to the camera focal length, the pixel width and height of the photo, and the relative altitude of the flight in the photo parameter set to form ground projection size data; According to the shooting position coordinates and heading angle of the overhead photo, the ground projection coordinates of the four corner points of the overhead photo are calculated in combination with the ground projection size data, and a photogrammetry control point coordinate table is established; The photogrammetry control point coordinate table is used to perform correspondence analysis with the pixel coordinates of the overhead photo, a six-parameter geometric correction parameter set is calculated, and a space transformation matrix is ​​constructed; Applying the spatial transformation matrix to the overhead photograph, performing affine transformation and resampling processing, and generating a geographically corrected overhead photograph; The geographically corrected overhead photograph is imported into ArcGIS software, and geographical coordinate registration and overlay operations are performed with the remote sensing image to obtain a corrected image overlaid on the remote sensing image.

4. The method for extracting mangrove ecological information based on drone according to claim 1, characterized in that: The step of generating a vector point layer based on the position and flight angle information of the side view photo and performing symbolization processing to obtain a photographing direction layer superimposed on the remote sensing image includes: Extracting GPS coordinates, flight heading angle and shooting timestamp from the EXIF ​​information of the side view photo to form a side view photo position angle information set; Performing spatial filtering on the side view photo position angle information set to remove data points with repeated positions or too large errors, thereby obtaining an optimized position angle data set; Using the GPS coordinate information in the optimized position angle data set, point features are created in ArcGIS software to construct a mangrove side view photo location point layer; According to the flight heading angle data in the optimized position angle data set, assigning direction attributes to the mangrove side view photo position point layer to generate a point layer with direction attributes; Based on the point layer with direction attributes, arrow symbolization design is performed, wherein the direction of the arrow indicates the direction of the photo, and the size of the arrow is proportional to the flight altitude, thereby forming a symbol layer for the direction of the mangrove side view photo; The direction symbol layer of the mangrove side view photo and the remote sensing image are spatially superimposed in the same coordinate system to obtain a photographing direction layer superimposed on the remote sensing image.

5. The method for extracting mangrove ecological information based on drone according to claim 1, characterized in that: The method extracts features and establishes associations based on the calibrated image, the photographing direction layer and the remote sensing image through a random forest algorithm to form a mangrove species interpretation sign library, including: In the overlapping area of ​​the calibrated image, the photographing direction layer and the remote sensing image, representative sample areas of different mangrove species are selected to create a mangrove species sample point set; For each sample point in the mangrove species sample point set, extracting texture features, color features and shape features from the calibrated image, and extracting spectral features and spatial features from the remote sensing image, to form a mangrove species feature vector set; The importance of the mangrove species feature vector set is sorted by recursive feature elimination, key features with high contribution are retained, and an optimized feature subset is constructed; Based on the optimized feature subset, a mangrove species classifier is trained using a random forest algorithm, wherein the number of decision trees is set to 500, the minimum number of samples per node is set to 5, and the feature random selection ratio is set to 0.7; Extracting classification rules from the decision tree set generated by the random forest algorithm, including the splitting threshold and species discrimination path of each feature, to generate a mangrove species discrimination rule set; An association mapping relationship is established between the mangrove species identification rule set and the corresponding side view photos, overhead photos and remote sensing image features to form a mangrove species interpretation mark library.

6. The method for extracting mangrove ecological information based on drone according to claim 1, characterized in that: The method of selecting training samples and verification samples on the remote sensing image according to the mangrove species interpretation mark library, performing supervised classification processing on mangrove species, and generating a mangrove species distribution map includes: Based on the species characteristic descriptions in the mangrove species interpretation symbol library, representative areas of each mangrove species are selected on the remote sensing image by using stratified sampling to generate a mangrove training sample set and a verification sample set; For each sample in the mangrove training sample set, the normalized vegetation index, red edge position index and texture complexity index are extracted as mangrove feature parameters to construct the mangrove sample feature matrix. Conduct supervised classification of mangrove species on the remote sensing image through the mangrove sample feature matrix, divide the distribution ranges of different mangrove species, and form a preliminary classification result of mangroves; The spectral similarity and spatial continuity characteristics of the preliminary classification results of mangroves were used to merge mangrove species patches and optimize their boundaries, eliminate isolated pixels and boundary noise, and obtain the optimized classification results of mangroves. Based on the mangrove validation sample set, the accuracy of the mangrove optimization classification results was evaluated, the overall accuracy, Kappa coefficient and classification accuracy of each species were calculated, and a mangrove classification accuracy report was generated; According to the accuracy indicators in the mangrove classification accuracy report, the mangrove optimization classification results are graded for credibility and converted into a vector format, and the species name, area, and distribution characteristic attribute information are added to generate the mangrove species distribution map.

7. The method for extracting mangrove ecological information based on drone according to claim 1, characterized in that: Based on the mangrove species distribution map, combined with spectral characteristics and spatial distribution information, the mangrove ecosystem function is evaluated and analyzed to obtain mangrove ecological monitoring data, including: Based on the mangrove species distribution map, the spatial distribution patterns of different mangrove species are extracted, the area, density and distribution pattern of each species are quantified, and a mangrove spatial structure parameter table is generated; Using the near-infrared band and visible light band data in the remote sensing image, the normalized vegetation index and leaf area index of the mangrove area are calculated to construct an evaluation index for mangrove growth vitality; In combination with the mangrove species distribution map, the mangrove spatial structure parameter table and the mangrove growth vitality evaluation index, the mangrove carbon storage is calculated using the biomass conversion coefficient to generate quantitative data on the mangrove carbon sink function; Extract the spatial relationship between mangroves and coastlines from the mangrove species distribution map and the mangrove spatial structure parameter table, analyze the wave attenuation and sediment retention capacity of mangroves in combination with tide level and topographic data, and form an evaluation result of the coastal protection function of mangroves; Analyze the species composition in the mangrove species distribution map and the distribution pattern in the mangrove spatial structure parameter table, calculate the species diversity index and niche overlap, and obtain the mangrove biodiversity maintenance function data; The quantified data on the carbon sink function of the mangroves, the assessment results of the coastal protection function of the mangroves and the data on the biodiversity maintenance function of the mangroves are integrated to obtain the mangrove ecological monitoring data.

8. A mangrove ecological information extraction system based on drone, used to implement the mangrove ecological information extraction method based on drone as described in any one of claims 1 to 7, characterized in that: The mangrove ecological information extraction system based on drone includes: An acquisition module is used to take low-altitude photos of mangroves at different angles through a drone, obtain overhead photos with a camera lens tilted at an angle of about 90° and side photos with a camera lens tilted at an angle of about 30°, and obtain multi-angle image data of the mangroves; A calculation module is used to calculate projection coordinates based on the multi-angle image data of the mangrove forest using position coordinates, length and width of the photo and flight altitude information, and to perform geometric correction processing on the overhead photo to obtain a corrected image superimposed on the remote sensing image; A processing module, for generating a vector point layer and performing symbolization processing based on the position and flight angle information of the side view photo, to obtain a photographing direction layer superimposed on the remote sensing image; Establishing a module for extracting features and establishing associations through a random forest algorithm based on the calibrated image, the photographing direction layer and the remote sensing image, so as to form a mangrove species interpretation sign library; A selection module is used to select training samples and verification samples on the remote sensing image according to the mangrove species interpretation mark library, perform supervised classification of mangrove species, and generate a mangrove species distribution map; The analysis module is used to evaluate and analyze the functions of the mangrove ecosystem based on the mangrove species distribution map in combination with spectral characteristics and spatial distribution information to obtain mangrove ecological monitoring data.

9. A computer device, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the method for extracting mangrove ecological information based on a drone as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the method for extracting mangrove ecological information based on a drone as claimed in any one of claims 1 to 7.

Images