Vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping

Abstract

The invention discloses a vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping, relates to the technical field of image data processing, and can solve the problems of splicing boundary blurring and detail loss caused by uneven illumination and shadow shielding in batch surveying and mapping of unmanned aerial vehicle hyperspectral images at the present stage. The method comprises the following steps: acquiring a plurality of hyperspectral images and GIS data of a target area; constructing an edge consistency constraint item; according to the plurality of hyperspectral images of the target area, carrying out area division on the target area and distributing an adaptive denoising weight; constructing a low-rank decomposition model according to the edge consistency constraint term and the adaptive denoising weight, and solving according to the low-rank decomposition model to obtain a plurality of denoised hyperspectral images; and carrying out registration and fusion on the plurality of denoised hyperspectral images and GIS data to generate a high-precision orthoimage of the target area, and carrying out inversion calculation on the vegetation coverage based on the orthoimage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing, and particularly relates to a vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, it is widely used in geographic information acquisition, environmental monitoring, disaster warning and other fields. With the advantages of flexible deployment, low cost and high resolution image acquisition, the unmanned aerial vehicle surveying and mapping by carrying remote sensing equipment to obtain remote sensing images and combining with GIS data has become an important technical means for natural resource dynamic supervision, especially in vegetation coverage quantitative analysis, using hyperspectral images to construct high-precision orthoimages is of great significance for natural resource monitoring.

[0003] However, in the process of unmanned aerial vehicle shooting hyperspectral images, large-scale surveying and mapping usually needs to be measured in batches, and the time period is different, which leads to uneven light conditions of the collected images. When splicing, the image splicing boundary is blurred and the edge information is lost. At the same time, due to the large area of shadow shielding of trees, buildings and other high objects, the brightness of the ground vegetation area in the image drops sharply, and its characteristics are covered by soil and other object noise, which affects the construction accuracy of the orthoimage.

[0004] When facing the above-mentioned scene, the traditional low-rank decomposition denoising algorithm smoothes the global noise of the image, which will cause the details to be over-smoothed, resulting in blurred object boundaries, and cannot achieve accurate denoising effect. SUMMARY

[0005] In order to solve the technical problems of blurred splicing boundary and detail loss caused by uneven light and shadow shielding in the batch surveying and mapping of unmanned aerial vehicle hyperspectral images at the present stage, and the over-smoothing of traditional denoising algorithm, the purpose of the present application is to provide a vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping, and the technical scheme adopted is as follows: In a first aspect, the present application provides a vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping, comprising: acquiring a plurality of hyperspectral images and GIS data of a target area; wherein the target area comprises a plurality of sub-areas, each hyperspectral image is used to represent the spectral reflectance distribution and spatial texture information of ground objects in a sub-area, and the GIS data comprises geographic information data used to provide spatial positioning reference for each sub-area; constructing an edge consistency constraint term, which is used to maintain the consistency of radiation brightness and spectral shape at the splicing edge of the current hyperspectral image and the adjacent image; dividing the target area and assigning adaptive denoising weights according to the plurality of hyperspectral images of the target area; wherein the adaptive denoising weight is used to represent the denoising intensity coefficient corresponding to the region with different texture-spectral feature complexity; constructing a low-rank decomposition model according to the edge consistency constraint term and the adaptive denoising weight, and solving the low-rank decomposition model to obtain a plurality of denoised hyperspectral images; registering and fusing the plurality of denoised hyperspectral images with the GIS data to generate a high-precision orthographic image of the target area, and calculating the vegetation coverage based on the orthographic image.

[0006] In a possible implementation, the edge consistency constraint term is constructed, specifically comprising: for the current hyperspectral image, determining all adjacent hyperspectral images according to the geographic coordinates of the sub-area corresponding to the hyperspectral image; wherein the current hyperspectral image is the hyperspectral image to be denoised currently, the adjacent hyperspectral image is the sub-area corresponding to the hyperspectral image to be denoised currently, and there are spatially adjacent hyperspectral images; determining the overlapping area between the current hyperspectral image and each adjacent hyperspectral image; constructing the edge consistency constraint term according to the brightness difference of the corresponding pixel points in the overlapping area and the spectral shape difference in different vegetation feature bands.

[0007] In a possible implementation, the overlapping area between the current hyperspectral image and each adjacent hyperspectral image is determined, specifically comprising: calculating the overlapping area between the current hyperspectral image and each adjacent hyperspectral image based on a preset lateral overlap parameter; wherein the lateral overlap parameter is used to control the overlap ratio between adjacent hyperspectral images.

[0008] In a possible implementation, the edge consistency constraint term is constructed according to the brightness difference of corresponding pixel points in the overlapping region and the spectral shape difference in different vegetation feature bands, and specifically includes: calculating the brightness difference value of the corresponding pixel points in the overlapping region as a brightness consistency component; calculating the spectral shape difference value of the corresponding pixel points in the overlapping region in the preset vegetation feature band as a spectral consistency component; wherein the preset vegetation feature band includes at least one of a near-infrared band, a short-wave infrared band and a red light band sensitive to the reflection characteristics of vegetation; the brightness consistency component and the spectral consistency component are weighted and fused according to a preset weight to construct the edge consistency constraint term.

[0009] In a possible implementation, according to a plurality of hyperspectral images of the target region, the target region is regionally divided and an adaptive denoising weight is assigned, and specifically includes: each hyperspectral image of the target region is divided into a plurality of local image blocks, and a texture-spectrum composite feature value of each local image block is calculated; wherein the texture-spectrum composite feature value is used to represent the comprehensive quantitative index of the texture complexity and the spectral reflection intensity of the pixels in the local image block; according to the distribution of the composite feature value of each local image block, all local image blocks are divided into a plurality of feature region types, and an adaptive denoising weight is assigned to each feature region type; wherein the feature region type is used to represent a set of image blocks with similar texture-spectrum complexity characteristics, and different types of feature regions correspond to different denoising intensity requirements.

[0010] In a possible implementation, according to the distribution of the composite feature value of each local image block, all local image blocks are divided into a plurality of feature region types, and an adaptive denoising weight is assigned to each feature region type, and specifically includes: according to the numerical distribution of the texture-spectrum composite feature value of all local image blocks, a feature threshold interval is determined; based on the feature threshold interval, the local image blocks are divided into three feature region types: feature uniform region, feature transition region and feature complex region; the first level of denoising weight coefficient is assigned to the feature uniform region, the second level of denoising weight coefficient is assigned to the feature transition region, and the third level of denoising weight coefficient is assigned to the feature complex region; wherein the first level is greater than the second level, and the second level is greater than the third level.

[0011] In a possible implementation, a low-rank decomposition model is constructed according to the edge consistency constraint term and the adaptive denoising weight, and specifically includes: a low-rank decomposition target function including a low-rank matrix term, a weighted sparse noise term and an edge consistency constraint term is constructed, and a low-rank decomposition model is constructed according to the low-rank decomposition target function; wherein the low-rank matrix term is used to constrain the low-rank characteristics of the image signal, the weighted sparse noise term is used to represent the sparsity of the noise signal, and the weighted sparse noise term is weighted by the adaptive denoising weight.

[0012] In a possible implementation, the denoised hyperspectral images are obtained according to the low-rank decomposition model, specifically including: iteratively solving the low-rank decomposition target function by using an augmented Lagrange multiplier method; updating the estimated values of the low-rank matrix and the sparse matrix in each iteration process until a convergence condition is met; and outputting the obtained low-rank matrix as the denoised hyperspectral images.

[0013] In a possible implementation, the denoised hyperspectral images are registered and fused with the GIS data to generate a high-precision orthographic image of the target region, and the vegetation coverage is calculated based on the orthographic image, specifically including: performing geometric correction and splicing processing on the denoised hyperspectral images by using a digital surface model in the GIS data to generate an orthographic image containing accurate geographic coordinates; extracting vegetation index features based on the orthographic image, and distinguishing vegetation regions and non-vegetation regions by using a preset classification algorithm; and calculating the vegetation coverage of each pixel by using a pixel bisection model, and statistically obtaining the overall vegetation coverage of the target region.

[0014] In a possible implementation, after obtaining the multiple hyperspectral images and the GIS data of the target region, the method further includes: performing a preprocessing operation on the multiple hyperspectral images and the GIS data of the target region; and the preprocessing operation includes performing atmospheric correction on the multiple hyperspectral images, and performing coordinate unification and topological repair on the GIS data.

[0015] In a second aspect, the present application provides a vegetation coverage quantitative analysis device based on GIS and unmanned aerial vehicle surveying and mapping, comprising: an acquisition unit and a processing unit. The acquisition unit is used to acquire multiple hyperspectral images and GIS data of a target region; wherein the target region comprises multiple sub-regions, each hyperspectral image is used to represent the spectral reflectance distribution and spatial texture information of ground objects in a sub-region, and the GIS data comprises geographic information data used to provide spatial positioning reference for each sub-region; the processing unit is used to construct an edge consistency constraint term, the edge consistency constraint term is used to maintain the consistency of radiation brightness and spectral shape of the current hyperspectral image and the adjacent image at the splicing edge; the processing unit is used to divide the target region and assign adaptive denoising weights according to the multiple hyperspectral images of the target region; wherein the adaptive denoising weights are used to represent the denoising intensity coefficients corresponding to regions with different texture-spectral feature complexity; the processing unit is used to construct a low-rank decomposition model according to the edge consistency constraint term and the adaptive denoising weights, and obtain denoised hyperspectral images according to the low-rank decomposition model; and the processing unit is used to register and fuse the denoised hyperspectral images with the GIS data to generate a high-precision orthographic image of the target region, and calculate the vegetation coverage based on the orthographic image.

[0016] In a third aspect, the present application provides an electronic device, comprising: a processor and a memory; wherein the memory is configured to store one or more programs, the one or more programs comprising computer-executable instructions that, when executed by the electronic device, cause the electronic device to perform the method for quantifying vegetation coverage based on GIS and unmanned aerial surveying as described in the first aspect and any possible implementation of the first aspect.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium storing one or more programs, the one or more programs comprising instructions configured to cause an electronic device of the present application to perform the method for quantifying vegetation coverage based on GIS and unmanned aerial surveying as described in the first aspect and any possible implementation of the first aspect when the instructions are executed by the electronic device.

[0018] In a fifth aspect, the present application provides a computer program product comprising instructions configured to cause an electronic device of the present application to perform the method for quantifying vegetation coverage based on GIS and unmanned aerial surveying as described in the first aspect and any possible implementation of the first aspect when the instructions are executed on a computer.

[0019] In a sixth aspect, the present application provides a chip system applied to a water meter data acquisition device; the chip system comprises one or more interface circuits and one or more processors. The interface circuit and the processor are interconnected through a circuit; the interface circuit is configured to receive a signal from a memory of the water meter data acquisition device and send the signal to the processor, the signal comprising computer instructions stored in the memory. When the processor executes the computer instructions, the water meter data acquisition device performs the method for quantifying vegetation coverage based on GIS and unmanned aerial surveying as described in the first aspect and any possible implementation of the first aspect.

[0020] The present application has the following beneficial effects: by acquiring the hyperspectral image of the target area sub-region and the GIS data containing the spatial positioning reference and preprocessing, a high-quality data foundation is laid for subsequent analysis; then by constructing the edge consistency constraint term, the problem of inconsistent radiation brightness and spectral shape of the image splicing edge collected by the unmanned aerial vehicle in batches is effectively solved, and splicing blur and feature loss are avoided; at the same time, the region is divided according to the feature of ground object texture-spectrum and the adaptive denoising weight is distributed, the defects of noise residue or detail loss caused by traditional low-rank decomposition global smoothing are solved, accurate denoising and complete retention of key features of vegetation are realized; then, the high-precision orthographic image is generated by registering and fusing the denoised image and the GIS data, and the vegetation coverage is retrieved from the image, which can not only accurately distinguish the vegetation and non-vegetation regions and reduce misjudgment, but also ensure that the coverage result has spatial accuracy and numerical accuracy, finally provides reliable quantitative basis for vegetation coverage for natural resource monitoring, ecological environment evaluation and other scenes, and significantly improves the overall precision and practicability of vegetation coverage quantitative analysis. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0022] Figure 1 The architecture schematic diagram of a vegetation coverage quantitative analysis system based on GIS and unmanned aerial vehicle surveying and mapping provided by an embodiment of the present application; Figure 2 The flowchart of a vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping provided by an embodiment of the present application; Figure 3 The flowchart of another vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping provided by an embodiment of the present application; Figure 4 The flowchart of another vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping provided by an embodiment of the present application; Figure 5 The flowchart of another vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping provided by an embodiment of the present application; Figure 6 The flowchart of another vegetation coverage quantitative analysis method based on GIS and unmanned aerial vehicle surveying and mapping provided by an embodiment of the present application. DETAILED DESCRIPTION

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0025] The terms "first" and "second," etc., used in the specification and drawings of this invention are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0026] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0027] It should be noted that in the embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0028] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0029] The technical terms involved in the embodiments of the present invention are explained below: 1. Geographic Information System (GIS) GIS is a technology system that integrates geospatial data acquisition, storage, analysis, and visualization. In this solution, GIS data includes the coordinate information, topographic data, and soil type of the target area. By pairing it with UAV hyperspectral imagery, it provides a geolocation benchmark for the imagery, supporting subsequent geometric correction, orthophoto mosaicking, and vegetation classification-assisted analysis, ensuring the consistency and accuracy of the spatial data.

[0030] 2. Hyperspectral Image (HSI) Hyperspectral Image is a remote sensing image containing continuous multiple bands (usually tens to hundreds), each pixel corresponds to a curve reflecting the spectral characteristics of the ground object. In this scheme, the hyperspectral image collected by the UAV can capture the reflectivity difference of vegetation in the near-infrared (NIR), short-wave infrared (SWIR), red (R) and other bands, and accurately identify vegetation through "red edge effect" and other characteristics. It is the core data to distinguish between vegetation and non-vegetation ground objects.

[0031] 3. Unmanned Aerial Vehicle Surveying and Mapping (UAVSM) UAVSM is a technology that uses UAVs to carry remote sensing equipment (such as hyperspectral cameras) to obtain ground images. Compared with traditional satellite remote sensing, it has the advantages of high resolution (centimeter level), flexible deployment and low cost, and can shoot large target areas in batches to provide fine image data for sub-regions, and realize the correlation of images and geographic coordinates through ground control points.

[0032] 4. Ground Control Point (GCP) GCP is a marked point with precise geographic coordinates (latitude, longitude, and elevation) laid out manually in the surveying area. In this scheme, the coordinate information of GCP is included in GIS data to correct the geometric distortion of UAV hyperspectral images (such as image tilt caused by flight attitude deviation), ensuring the spatial position accuracy of each sub-region image and providing a reference for subsequent image stitching and orthophoto map construction.

[0033] 5. Orthophoto Map (OM) Orthophoto Map is a remote sensing image that has been geometrically corrected to eliminate the effects of terrain undulations and sensor tilt, with consistent scale and direct distance and area measurement characteristics. In this scheme, the orthophoto map is generated by fusing the denoised hyperspectral image and the digital surface model (DSM) simulated by GIS data, and is the basis for vegetation coverage quantitative analysis.

[0034] 6. Normalized Difference Vegetation Index (NDVI) The normalized vegetation index is a vegetation index calculated by the reflectivity of the near-infrared band and the reflectivity of the red light band, and the formula is: NDVI=(near-infrared reflectivity-red light reflectivity) / (near-infrared reflectivity+red light reflectivity). The value range is [-1, 1], and the greater the positive value, the higher the vegetation coverage. It is the core index for retrieving vegetation coverage in the scheme and can effectively distinguish between vegetation and non-vegetation areas.

[0035] The above describes the technical terms related to the embodiments of the present application.

[0036] The specific scheme of the vegetation coverage quantitative analysis method based on GIS and unmanned aerial survey provided by the present application will be described in detail below with reference to the accompanying drawings.

[0037] For example, as shown in Figure 1 , as shown in Figure 1 An architecture diagram of a vegetation coverage quantitative analysis system based on GIS and unmanned aerial survey provided by an embodiment of the present application is shown. The vegetation coverage quantitative analysis system 10 includes a data acquisition module 11, an image denoising module 12, and a coverage calculation module 13. Through the cooperative work of the data acquisition module 11, the image denoising module 12, and the coverage calculation module 13, the vegetation coverage quantitative analysis system 10 can realize high-precision vegetation coverage quantitative analysis. The modules will be introduced in turn as follows: (1) Data acquisition module 11.

[0038] The data acquisition module 11 is the basic input unit of the system 10, responsible for acquiring the original data of the target area and performing preprocessing, providing a high-quality data source for subsequent denoising and analysis.

[0039] Optionally, the data acquisition module 11 is configured to acquire a plurality of hyperspectral images and GIS data of the target area.

[0040] The target area includes a plurality of sub-regions, and each hyperspectral image is used to represent the spectral reflectivity distribution and spatial texture information of the ground objects in a sub-region. The GIS data includes geographic information data for providing spatial positioning reference for each sub-region. In one possible implementation, the data acquisition module 11 acquires the hyperspectral images of the target area through a hyperspectral sensor. Specifically, the data acquisition module 11 carries the hyperspectral sensor on an unmanned aerial vehicle and shoots the target area in batches according to a preset flight plan. The target area is divided into a plurality of sub-regions, and each sub-region corresponds to a collected hyperspectral image. The image contains the spectral reflectivity distribution of the ground objects in the near-infrared, short-wave infrared, red light, and other bands within the sub-region, as well as the spatial texture information of the ground objects (for example, the complex texture of the forest edge and the smooth texture of the bare land). During the collection process, ground control points are arranged in each sub-region to associate the hyperspectral image with the geographic coordinates.

[0041] and a data acquisition module 11, according to the UAV flight plan, placing ground control points in each region, and then pairing the hyperspectral data collected in each region with the positioning information of the ground control points in each region as geographic information system data, facilitating subsequent construction of orthophotos using GIS information and image data.

[0042] Optionally, the data acquisition module 11, after obtaining the plurality of hyperspectral images and GIS data of the target region, further performs a preprocessing operation on the plurality of hyperspectral images and GIS data of the target region. The preprocessing operation includes atmospheric correction of the plurality of hyperspectral images and coordinate unification and topological repair of the GIS data.

[0043] Specifically, the data acquisition module 11 performs atmospheric correction on the hyperspectral images to eliminate the interference of atmospheric scattering on spectral reflectance; performs coordinate system unification, attribute cleaning, and topological repair on the GIS data to ensure that the GIS data of different sub-regions are consistent in space (e.g., closed boundaries, no coordinate deviation). The preprocessed hyperspectral images and GIS data will be input to the image denoising module 12, directly affecting the accuracy of subsequent denoising effect.

[0044] (2) Image denoising module 12.

[0045] The image denoising module 12 is the core processing unit of the system, which, based on the preprocessed data output by the data acquisition module 11, eliminates noise through an improved low-rank decomposition algorithm while preserving key features of vegetation.

[0046] Optionally, the image denoising module 12 is configured to construct an edge consistency constraint term, which is used to maintain the consistency of radiance and spectral shape at the stitching edges of the current hyperspectral image and the adjacent image. Then, the image denoising module 12 performs regional division on the target region and assigns adaptive denoising weights according to the plurality of hyperspectral images of the target region; the adaptive denoising weights are used to represent the denoising intensity coefficients corresponding to regions with different texture-spectral feature complexity. Finally, the image denoising module 12 constructs a low-rank decomposition model according to the edge consistency constraint term and the adaptive denoising weights, and solves the low-rank decomposition model to obtain the plurality of denoised hyperspectral images. Specifically, for the hyperspectral images output by the data acquisition module 11, the image denoising module 12 first determines the adjacent images (e.g., eight adjacent images of a certain sub-region image) of each image according to the geographic coordinates in the GIS data, and then calculates the overlapping region based on a preset lateral overlap parameter. Subsequently, a constraint term is constructed in two dimensions: one is to calculate the brightness difference of the pixels in the overlapping region as the brightness consistency component; the other is to calculate the spectral angle difference of the vegetation feature band as the spectral consistency component. Finally, the two components are weighted and fused to form an edge consistency constraint term, which is used to maintain the radiation brightness and spectral shape consistency of the splicing edge, providing boundary constraints for subsequent denoising.

[0047] Further, the image denoising module 12 receives the preprocessed hyperspectral images, divides each image into 7x7 local image blocks, and calculates the texture-spectrum composite feature value of each image block. According to the threshold interval (e.g., 25% quantile, 75% quantile) of the feature value distribution, the image blocks are divided into feature uniform regions (e.g., bare land), feature transition regions (e.g., forest edge), and feature complex regions (e.g., forest), and are respectively assigned with first to third adaptive denoising weights (weights decreasing in turn), realizing differential processing of strong denoising in smooth regions and detail preservation in complex regions.

[0048] After that, the image denoising module 12 constructs a low-rank decomposition target function based on the edge consistency constraint term determined online and the region division and adaptive denoising weight. The augmented Lagrange multiplier method is used to iteratively solve and update the low-rank matrix (effective signal) and sparse matrix (noise) until convergence. The final output low-rank matrix is the denoised hyperspectral image, which is input into the coverage calculation module 13 as the basis for constructing the orthophoto.

[0049] (3) Coverage calculation module 13.

[0050] The coverage calculation module 13 is the result output unit of the system, responsible for obtaining the original data of the target region and performing preprocessing to provide high-quality data sources for subsequent denoising and analysis.

[0051] Optionally, the coverage calculation module 13 is configured to register and fuse the denoised multiple hyperspectral images with the GIS data to generate a high-precision orthophoto of the target region, and calculate the vegetation coverage based on the orthophoto.

[0052] Specifically, the coverage calculation module 13 receives the denoised hyperspectral images, performs geometric correction using the GIS data output by the data acquisition module 11 to eliminate image distortion caused by terrain undulations and UAV flight attitude. Subsequently, the corrected images are spliced to generate a target region orthophoto with consistent radiation and accurate coordinates, providing a unified base map for vegetation classification and coverage calculation.

[0053] Further, the coverage calculation module 13 extracts vegetation features such as normalized vegetation index based on the orthographic image, and combines auxiliary information such as terrain and soil type in the GIS data to distinguish the vegetation area and non-vegetation area (such as soil, building) by using the random forest classification algorithm. The classification result provides pixel-level classification basis for coverage inversion.

[0054] After that, the coverage calculation module 13 calculates the vegetation coverage of each pixel based on the classification result of the vegetation area and the non-vegetation area output by the foregoing step by using the pixel dichotomy model, and calculates the average coverage of all pixels in the target area, and finally outputs the overall vegetation coverage of the target area, and completes the quantitative analysis.

[0055] The above describes the vegetation coverage quantification system 10 and the modules included therein. In the system, the data acquisition module 11 provides the entire system with "spatial consistency, reliable radiation" raw data through preprocessing, and the output directly determines the processing basis of the image denoising module 12; the image denoising module 12 optimizes the output of high-quality images through edge constraint and adaptive weight, which provides guarantee for the orthographic image construction and classification accuracy of the coverage calculation module 13; and the coverage calculation module 13 realizes the conversion from image to quantitative result based on the results of the previous modules, and finally outputs the vegetation coverage quantitative analysis conclusion.

[0056] For example, refer to Figure 2 which shows a flowchart of a vegetation coverage quantification analysis method based on GIS and unmanned aerial vehicle surveying and mapping according to an embodiment of the present application, including the following steps: S201, obtaining a plurality of hyperspectral images and GIS data of a target area.

[0057] The target area includes a plurality of sub-areas, and each hyperspectral image is used to represent the spectral reflectance distribution and spatial texture information of ground objects in a sub-area. The GIS data includes geographic information data used to provide spatial positioning reference for each sub-area. This step can be performed by the data acquisition module 11 in the foregoing vegetation coverage quantification system 10. The process of the data acquisition module 11 obtaining the plurality of hyperspectral images and GIS data of the target area is described as follows: (1) Obtain a plurality of hyperspectral images of the target area.

[0058] Specifically, the data acquisition module 11 first divides the target region to be analyzed into N continuous sub-regions (N≥2) according to a preset grid, and the spatial overlap ratio of adjacent sub-regions is not less than 70% to meet the lateral overlap degree parameter requirement and ensure that there is enough overlapping area for feature matching when the subsequent image is spliced. For each sub-region, a flight route of the unmanned aerial vehicle is planned, and a flight height is set. For example, the flight height is set to 50-100 meters to ensure that the ground resolution of the hyperspectral image reaches 0.1-0.5 meters / pixel.

[0059] Further, after the sub-regions are divided, the data acquisition module 11 acquires the hyperspectral image. Specifically, the hyperspectral image can be acquired by using an unmanned aerial vehicle carrying a hyperspectral imager to shoot each sub-region according to a preset flight plan. Each hyperspectral image is a three-dimensional data cube, which includes information of the spectral dimension and the spatial dimension: the spatial dimension includes two-dimensional coordinate information of the objects in the sub-region, and the spatial texture information of the objects is represented by the image gray value distribution, such as the dense texture of the forest area and the smooth texture of the bare land area. The spectral dimension includes the reflectivity data of each pixel in the near-infrared waveband, the short-wave infrared waveband, the red waveband and other vegetation-sensitive wavebands, forming a spectral reflectivity distribution curve, wherein the average reflectivity of the vegetation in the NRI waveband is 30%-50% higher than that of the non-vegetation objects, which provides a feature basis for subsequent vegetation identification.

[0060] For example, during the acquisition process, the data acquisition module 11 can arrange 5 ground control points at the four corners and the center of each sub-region to establish the mapping relationship between the hyperspectral image and the geographic space.

[0061] (2) Obtain GIS data.

[0062] For example, the data acquisition module 11 places ground control points in each region according to the flight plan of the unmanned aerial vehicle, and then pairs the hyperspectral data collected in each region with the geographic information system data according to the positioning information of the ground control points in each region, so as to facilitate the construction of the orthographic image by using the GIS information and the image data.

[0063] In one possible implementation, the data acquisition module 11 performs a preprocessing operation on the multiple hyperspectral images and the GIS data of the target region. The preprocessing operation includes atmospheric correction of the multiple hyperspectral images, coordinate unification and topological repair of the GIS data.

[0064] Specifically, the atmospheric correction of the hyperspectral image can use the FLAASH model to eliminate the influence of atmospheric scattering and absorption, so that the spectral reflectivity error is controlled within 5%.

[0065] and, the GIS data is unified in coordinates, and all data is converted to the WGS-84 coordinate system; the GIS data is attribute cleaned, and repeated or invalid control point coordinates are removed; the GIS data is topologically repaired, and the soil type vector boundary is closed.

[0066] Therefore, the preprocessed hyperspectral image and GIS data will be input into the image denoising module 12, wherein the spectral reflectance distribution and spatial texture information of the hyperspectral image provide basic features for the construction of the subsequent edge consistency constraint term, and the geographic coordinate information of the GIS data is used to determine the spatial relationship of the adjacent sub-regions, and provide a spatial reference for the region division and weight allocation in the image denoising module 12.

[0067] S202, constructing an edge consistency constraint term.

[0068] The edge consistency constraint term is used to maintain the consistency of the radiation brightness and spectral shape of the current hyperspectral image and the adjacent image at the splicing edge.

[0069] In a possible implementation, the present step can be performed by the image denoising module 12 in the aforementioned vegetation coverage quantitative analysis system 10, including the following steps: first, the image denoising module 12 determines all adjacent hyperspectral images according to the geographic coordinates of the corresponding sub-region of the hyperspectral image for the current hyperspectral image; then, the image denoising module 12 determines the overlapping region between the current hyperspectral image and each adjacent hyperspectral image; finally, the image denoising module 12 constructs the edge consistency constraint term according to the brightness difference of the corresponding pixel points in the overlapping region and the spectral shape difference in different vegetation feature bands. Wherein, the current hyperspectral image is the current hyperspectral image to be denoised, the adjacent hyperspectral image is the sub-region corresponding to the current hyperspectral image to be denoised, and the hyperspectral images exist spatially adjacent.

[0070] It should be noted that the specific process of the image denoising module 12 constructing the edge consistency constraint term according to the aforementioned three steps can be referred to S301-S303 in the following text, which will not be described here.

[0071] It can be understood that the core of the present step is to construct a targeted constraint mechanism for the inherent pain points of the hyperspectral image collected by the unmanned aerial vehicle in batches, such as the light angle difference caused by shooting at different time periods, the blurred splicing edge, and the vegetation spectral features being covered due to the tree / building shadow shielding, etc., to lay an accuracy foundation for subsequent image denoising and orthographic image construction.

[0072] This step, by constraining the radiance of the overlapping area between the current hyperspectral image and adjacent images, can effectively mitigate problems such as abrupt brightness changes and banding caused by uneven illumination at the stitching points, and avoid blurring of the outlines of features at the stitching edges (such as loss of vegetation boundaries) caused by brightness differences. On the other hand, by constraining the spectral shape of the vegetation-sensitive bands, the inherent spectral characteristics of vegetation can be utilized to maintain the continuity of the vegetation spectral curve even in shaded areas, preventing the vegetation spectral features from being masked by noise from soil and other ground features, and avoiding misjudgments when classifying vegetation and non-vegetation areas in the future.

[0073] Therefore, the edge consistency constraint term constructed by the image denoising module 12 will serve as the key input of the low-rank decomposition model solving submodule 123 in the image denoising module 12, ensuring that random noise is eliminated without destroying the authenticity of ground feature details at the stitching edges during the low-rank decomposition denoising process. At the same time, the stitching edge consistency brought by this constraint term can directly improve the stitching accuracy when generating orthophotos in the subsequent coverage calculation module 13, reduce feature faults in the stitching area of ​​the orthophoto, and ultimately provide a high-quality image foundation for the accurate inversion of vegetation coverage, avoiding coverage quantization deviations caused by stitching edge problems.

[0074] S203. Based on multiple hyperspectral images of the target region, the target region is divided into regions and adaptive denoising weights are assigned.

[0075] Among them, the adaptive denoising weight is used to characterize the denoising intensity coefficient corresponding to regions with different texture-spectral feature complexities.

[0076] In one possible implementation, this step can be specifically performed by the image denoising module 12 in the aforementioned vegetation cover quantification analysis system 10, including the following steps: First, the image denoising module 12 divides each hyperspectral image of the target region into multiple local image blocks and calculates the texture-spectral composite feature value of each local image block. Then, based on the distribution of the composite feature values ​​of each local image block, the image denoising module 12 divides all local image blocks into multiple feature region types and assigns an adaptive denoising weight to each feature region type. The texture-spectral composite feature value is used to characterize the comprehensive quantitative index of pixel texture complexity and spectral reflectance intensity in a preset feature band within a local image block. The feature region type is used to characterize a set of image blocks with similar texture-spectral complexity features, and different types of feature regions correspond to different denoising intensity requirements.

[0077] It should be noted that the specific process of the image denoising module 12 dividing the target region and assigning adaptive denoising weights according to the aforementioned two steps can be found in S401-S402 below, and will not be repeated here.

[0078] It can be understood that, in order to solve the defects that the traditional low-rank decomposition denoising algorithm cannot adapt to the heterogeneity of the feature of the ground object in the hyperspectral image, resulting in noise residue or detail loss, the feature-driven regional division and the demand-matched weight distribution are used to provide differentiated constraint basis for subsequent accurate denoising. Specifically, the target region is divided into different types according to the texture-spectrum feature of the ground object, and the differentiated denoising weight is matched, so as to effectively solve the problem that the traditional global denoising cannot adapt to the heterogeneity of the ground object, realize the accurate processing of strong denoising of uniform regions and detail preservation of complex regions, completely suppress noise and completely preserve key details of vegetation, provide high-quality data support for subsequent orthographic image construction and vegetation classification, and finally guarantee the quantitative accuracy of the vegetation coverage rate.

[0079] S204, constructing a low-rank decomposition model according to the edge consistency constraint term and the adaptive denoising weight, and obtaining the plurality of denoised hyperspectral images according to the low-rank decomposition model.

[0080] In a possible implementation, the step can be specifically performed by the image denoising module 12 in the aforementioned vegetation coverage quantitative analysis system 10, and the construction of the low-rank decomposition model includes: constructing a low-rank decomposition objective function including a low-rank matrix term, a weighted sparse noise term and an edge consistency constraint term, and then defining the low-rank decomposition objective function and a decomposition equation constraint as the low-rank decomposition model; wherein the low-rank matrix term is used to constrain the low-rank characteristics of the image signal, the weighted sparse noise term is used to represent the sparsity of the noise signal, the weighted sparse noise term is weighted by the adaptive denoising weight, and the decomposition equation constraint is used to ensure the conservation of the total amount of image information before and after the denoising process.

[0081] In addition, when the image denoising module 12 obtains the plurality of denoised hyperspectral images according to the low-rank decomposition model, it can include: iteratively solving the low-rank decomposition objective function by using the augmented Lagrange multiplier method; updating the estimated values of the low-rank matrix and the sparse matrix in each iteration process until the convergence condition is met; and outputting the finally obtained low-rank matrix as the denoised hyperspectral image.

[0082] It should be noted that the specific process of the image denoising module 12 constructing the low-rank decomposition model and obtaining the plurality of denoised hyperspectral images according to the above steps can be referred to S501-S502 in the following text, which will not be repeated here.

[0083] It can be understood that in this step, the image denoising module 12 integrates the edge consistency constraint term generated in the previous step and the adaptive denoising weight into the low-rank decomposition model, and realizes the denoising goal of accurate noise suppression and complete preservation of key features through model solving. On the one hand, the edge consistency constraint term plays a role in the model, effectively weakening the problems of sudden change of brightness and spectral feature fault at the edge of the hyperspectral image spliced by the unmanned aerial vehicle in batches, avoiding the blurring or feature mutation of the ground object at the splicing position, and ensuring the continuity of the edge features after splicing the images of different sub-regions.

[0084] On the other hand, the adaptive denoising weight enables the model to dynamically adjust the denoising strength for different texture-spectral feature regions, to realize strong noise suppression for feature uniform regions and to avoid noise residue interference in subsequent classification, to realize weak constraint denoising for feature complex regions, to completely preserve key details such as vegetation texture and object boundary, and to prevent the loss of details caused by excessive smoothing of traditional low-rank decomposition; the hyperspectral image obtained by finally solving has radiation consistency and spatial continuity, and accurately preserves the vegetation spectral-texture features, providing a high-quality data source for high-precision orthoimage generation, vegetation region classification and vegetation coverage inversion in the subsequent coverage calculation module 13, directly reducing the coverage quantization error caused by image quality defects, and ensuring the reliability and accuracy of the quantization result.

[0085] S205, registering and fusing the denoised multiple hyperspectral images with the GIS data to generate a high-precision orthoimage of the target region, and calculating the vegetation coverage based on the orthoimage.

[0086] In one possible implementation, this step can be specifically performed by the coverage calculation module 13 in the aforementioned vegetation coverage quantification analysis system 10, including the following steps: first, the coverage calculation module 13 uses the digital surface model in the GIS data to perform geometric correction and splicing processing on the denoised multiple hyperspectral images to generate an orthoimage containing accurate geographic coordinates; then, the coverage calculation module 13 extracts vegetation index features based on the orthoimage, and distinguishes vegetation regions and non-vegetation regions using a preset classification algorithm; finally, the coverage calculation module 13 calculates the vegetation coverage of each pixel through a pixel bisection model, and statistically obtains the overall vegetation coverage of the target region.

[0087] It should be noted that the specific process of generating a high-precision orthoimage of the target region and calculating the vegetation coverage based on the orthoimage by the coverage calculation module 13 can be referred to as S601-603 in the following text, which will not be described here.

[0088] Understandably, in this step, the vegetation cover calculation module 13 relies on the high-quality data output by the preceding module to achieve the transformation from image data to vegetation cover quantification results through data fusion and quantification inversion. On the one hand, the denoised hyperspectral image, combined with GIS data registration and fusion, can eliminate geometric distortions caused by UAV flight attitude deviations and terrain undulations, and ensure the radiometric consistency of sub-region images. The generated high-precision orthophoto has both accurate geographic coordinates and complete vegetation spectral-texture features, avoiding vegetation area positioning deviations or feature loss, and providing a reliable image base map for vegetation cover inversion. On the other hand, vegetation cover inversion based on this orthophoto can accurately distinguish between vegetated and non-vegetated areas, reduce misjudgments in traditional images, and make the final target area vegetation cover have both spatial accuracy and numerical precision, providing a reliable quantitative basis for natural resource monitoring, ecological environment assessment, etc.

[0089] Based on the above technical solutions, this invention acquires and preprocesses hyperspectral images of paired sub-regions of the target area and GIS data containing spatial positioning references, laying a high-quality data foundation for subsequent analysis. Furthermore, by constructing an edge consistency constraint term, it effectively solves the problem of inconsistent radiance and spectral shape at the edges of images stitched together by batches of UAV acquisition, avoiding blurring and feature loss. Simultaneously, it divides regions according to land cover texture-spectral features and assigns adaptive denoising weights, overcoming the defects of noise residue or detail loss caused by global smoothing in traditional low-rank decomposition, achieving accurate denoising while fully preserving key vegetation features. Subsequently, based on the registration and fusion of the denoised image and GIS data, a high-precision orthophoto is generated. Combined with image inversion to retrieve vegetation cover, it can not only accurately distinguish between vegetated and non-vegetated areas and reduce misjudgments, but also ensure that the coverage results have both spatial accuracy and numerical precision. Ultimately, it provides reliable quantitative evidence of vegetation cover for scenarios such as natural resource monitoring and ecological environment assessment, significantly improving the overall accuracy and practicality of vegetation cover quantitative analysis.

[0090] For example, in combination Figure 2 ,like Figure 3 As shown, this invention provides another method for quantitative analysis of vegetation cover based on GIS and UAV mapping. This method constructs an edge consistency constraint term, specifically including the following steps: S301. For the current hyperspectral image, determine all adjacent hyperspectral images based on the geographic coordinates of the corresponding sub-region of the hyperspectral image.

[0091] Among them, the current hyperspectral image is the hyperspectral image to be denoised, and the adjacent hyperspectral image is the sub-region corresponding to the current hyperspectral image to be denoised, and there are spatially adjacent hyperspectral images.

[0092] Specifically, the image denoising module 12 targets the current hyperspectral image I to be denoised. n(wherein n is the serial number of the current sub-region, and the value range is 1 to N, N is the total number of sub-regions of the target region, each I n corresponds to an independent sub-region in the target region), first, the geographic coordinates of the corresponding sub-region of the hyperspectral image are obtained, and the geographic coordinates are derived from the ground control point coordinates included in the GIS data.

[0093] After that, the image denoising module 12 determines the hyperspectral images corresponding to the sub-regions that have a spatial adjacency relationship with the current sub-region based on the above-mentioned geographic coordinates through spatial topological relationship analysis, and forms an adjacency hyperspectral image set . Among them: if the current sub-region is a non-boundary sub-region of the target region (i.e., there are other sub-regions around the sub-region), then contains 8 adjacent hyperspectral images, respectively corresponding to the east, south, west, north, northeast, southeast, northwest, and southwest eight directions of the adjacent sub-region; if the current sub-region is a boundary sub-region of the target region, that is, part of the boundary of the sub-region is the edge of the target region, and there is no adjacent sub-region, then the number of adjacent images is adjusted according to the actual boundary position, usually contains 3-5 adjacent hyperspectral images, such as only adjacent to the east, south, and northeast direction sub-region, contains 3 images.

[0094] Let any one of the adjacent hyperspectral images be I m (wherein m is the serial number of the adjacent sub-region, m≠n, and I m and I n corresponding sub-regions have a spatial adjacency, and the subsequent steps will be expanded based on the matching relationship between I n and a single I m , and the processing logic of all adjacent images is consistent.

[0095] S302, determine the overlapping region between the current hyperspectral image and each adjacent hyperspectral image.

[0096] Optionally, the image denoising module 12 calculates the overlapping region between the current hyperspectral image and each adjacent hyperspectral image based on a preset lateral overlap parameter. The lateral overlap parameter is used to control the overlap ratio between adjacent hyperspectral images.

[0097] Specifically, in combination with the description in the foregoing S301, the image denoising module 12 extracts the overlapping region between I m and I n based on the lateral overlap parameter preset in the unmanned aerial vehicle flight plan through an image registration algorithm: the local image in I n belonging to the overlapping region is denoted as O n , and the local image in I m belonging to the overlapping region is denoted as O m, O n With O m the same number of pixels, and the corresponding pixels represent the same geographical location in the target area. The number of pixels in the overlapping area is not less than 30% of the total number of pixels in the single hyperspectral image, ensuring sufficient sample size for subsequent calculation of brightness difference and spectral shape difference, and avoiding reduced accuracy of the constraint term due to insufficient samples.

[0098] For example, the above-mentioned overlap value is 30%-50%, which can be set according to the size of the sub-region, the image resolution and the requirement of the splicing accuracy, to ensure that the adjacent image overlapping area is sufficient for feature matching. In actual application, the lateral overlap can be determined according to the requirement, and the embodiment of the present application is not limited specifically.

[0099] S303, constructing an edge consistency constraint term according to the brightness difference of the corresponding pixels in the overlapping area and the spectral shape difference in different vegetation feature bands.

[0100] In a possible implementation, the image denoising module 12 specifically implements constructing an edge consistency constraint term according to the brightness difference of the corresponding pixels in the overlapping area and the spectral shape difference in different vegetation feature bands through the following three sub-steps: (1) calculating the brightness difference value of the corresponding pixels in the overlapping area as a brightness consistency component; For example, the brightness consistency component in this step can be represented by an illumination intensity component.

[0101] Specifically, to quantify the brightness difference between O n and O m , first, the image data of O n and O m is converted from the RGB color space to the HSV color space, and the color space conversion method is a prior art, and the pixel coordinates are kept one-to-one correspondence during the conversion process. The V channel value in the HSV space is extracted as the illumination intensity component, wherein the illumination intensity component of the pixel point (x, y) in O n is denoted as , and the illumination intensity component of the pixel point (x, y) in O m is denoted as It should be pointed out that the RGB subscript here only represents that the illumination intensity component is derived from the RGB band data conversion of the hyperspectral image, and is not directly used as the color value of the RGB band. The purpose is to eliminate the interference of color information and only keep the brightness feature of the surface feature.

[0102] (2) calculating the spectral shape difference value of the corresponding pixels in the preset vegetation feature band in the overlapping area as a spectral consistency component.

[0103] The preset vegetation characteristic waveband includes at least one of a near-infrared waveband, a short-wave infrared waveband and a red light waveband sensitive to vegetation reflection characteristics.

[0104] Exemplarily, the spectral consistency component in the step can be represented by a spectral angle.

[0105] Specifically, to quantify the O n and the O m , the wavebands sensitive to vegetation reflection characteristics, i.e., the near-infrared waveband, the short-wave infrared waveband and the red light waveband, are preferentially selected, and the selected waveband set is denoted as The spectral angle is used to quantify the O n and the O m corresponding to the spectral shape difference of the pixels, and the spectral angle function is denoted as , wherein represents the spectral reflectance value of the pixel point (x, y) in the set b waveband in the O n , and represents the spectral reflectance value of the pixel point (x, y) in the set b waveband in the O m , both of which are directly taken from the preprocessed hyperspectral image waveband data.

[0106] Exemplarily, the spectral angle function is denoted as , and the calculation formula is as follows: , wherein arccos is an inverse cosine function, is an L2 norm, which is used to calculate the module length of the pixel point spectral vector.

[0107] (3) The brightness consistency component and the spectral consistency component are weighted and fused according to a preset weight to construct the edge consistency constraint term.

[0108] Exemplarily, the edge consistency constraint term in the step is denoted as , and the core is to realize the double constraint on the splicing edge by weightedly fusing the brightness difference and the spectral shape difference, and the formula is as follows: Where α represents the weighting coefficient of the brightness consistency component, and β represents the weighting coefficient of the spectral consistency component. In this embodiment, α is used to adjust the contribution of brightness difference in the constraint terms. Since uneven illumination is one of the main causes of blurred stitching edges, it needs to be given a certain weight to weaken abrupt changes in brightness, and is generally set to 0.4. β is used to adjust the contribution of spectral difference in the constraint terms, and is generally set to 0.6. This is because the spectral characteristics of the vegetation area are the core basis for subsequent vegetation classification and cover inversion, and it needs to be given a higher weight to ensure the continuity of the spectral shape. It should be noted that α + β = 1, which can be determined according to the needs in practical applications.

[0109] The edge consistency constraint term constructed through the above steps can simultaneously constrain the brightness continuity and spectral shape consistency of the stitched edges, effectively solving problems such as uneven lighting and shadow occlusion caused by batch shooting by drones. It provides a boundary constraint basis for solving the low-rank decomposition model in the subsequent image denoising module 12, ensuring that the stitched edges of the denoised hyperspectral image are clear and the vegetation features are complete, laying the foundation for the construction of high-precision orthophotos and the quantification of vegetation coverage.

[0110] Based on the above technical solution, the embodiments of the present invention first determine the adjacent images of the current hyperspectral image according to GIS geographic coordinates, then extract sufficient overlapping areas according to a preset lateral overlap degree, and finally construct an edge consistency constraint term by combining the pixel brightness difference of the overlapping area with the spectral angle difference of the vegetation sensitive band. This effectively solves the problems of blurred stitching edges, sudden changes in brightness, and discontinuities in vegetation spectral features caused by uneven lighting and shadow occlusion when taking pictures in batches by UAVs. While ensuring the continuity of stitching edge brightness and consistency with the spectral shape of vegetation, it provides accurate boundary constraints for subsequent low-rank decomposition and noise reduction, thereby laying a high-quality image foundation for the generation of high-precision orthophotos of the target area and quantitative analysis of vegetation coverage, and reducing subsequent quantization errors caused by stitching edge defects.

[0111] For example, combined Figure 2 ,like Figure 4 As shown, this invention provides another method for quantitative analysis of vegetation cover based on GIS and UAV mapping. This method involves dividing the target area into regions and assigning adaptive denoising weights based on multiple hyperspectral images of the target area. Specifically, it includes the following steps: S401. Divide each hyperspectral image of the target region into multiple local image blocks, and calculate the texture-spectral composite feature value of each local image block.

[0112] Among them, the texture-spectral composite feature value is used as a comprehensive quantitative index to characterize the complexity of pixel texture within a local image patch and the intensity of spectral reflection in a preset feature band.

[0113] In one possible implementation, the image denoising module 12 specifically implements S401 through the following two sub-steps: (1) Divide each hyperspectral image of the target region into multiple local image blocks; In this step, the image denoising module 12 uses the preprocessed hyperspectral image as the basis for processing the current hyperspectral image I. n For each object, a fixed-size sliding window is used for block processing. This results in... Image blocks ,in This represents the first image patch. This represents the second image patch. Indicates the first Image blocks.

[0114] For example, the window size of the sliding window can be set to 7×7. The window size of the sliding window can be determined according to the needs in actual applications, and no specific limitation is made here.

[0115] (2) Calculate the texture-spectral composite feature value of each local image patch.

[0116] Regarding the first Image blocks Construct texture-spectral composite eigenvalues This feature value is used to characterize the texture and main spectral bands of different image patches in a hyperspectral image. Specifically, this texture-spectral composite feature value is used to comprehensively quantify the texture complexity of pixels within an image patch and the spectral reflectance intensity in a preset vegetation feature band. A higher value indicates a more complex land cover structure and stronger vegetation spectral activity in that area. For example, It can be determined using the following formula: in, Represents image blocks The total number of pixels in the image (for example, combining the examples in the preceding steps, for a 7×7 image block, =49). This indicates the number of selected characteristic bands. Represents the image patch after grayscale conversion medium pixel The texture feature value is obtained by calculating the gray-level difference between a pixel and its eight neighboring pixels using the gray-level difference statistical method. The weighted average method and the gray-level statistical method are existing technologies, and their specific processes will not be elaborated here. Indicated in image block medium pixel place Spectral intensity characteristic values ​​of the band. Represents image blocks Texture feature values, Represents image blocks Spectral characteristic values.

[0117] In the above formula, the average values ​​of the texture feature components and the spectral intensity feature components are calculated respectively. This is to eliminate the influence of image patch size and the number of selected bands on the feature values, so that the feature values ​​between different image patches are comparable. Optionally, the image denoising module 12 normalizes the calculated values ​​to the [0,1] interval to facilitate subsequent unified thresholding.

[0118] Further calculations were performed on the texture feature values ​​of all image patches. and spectral characteristic values Then, the normalization method was used to... and Normalize to [0,1] to eliminate the interference of texture features and differences in the numerical range of spectral intensity in different bands on the feature values.

[0119] Furthermore, It can be determined using the following formula: and These represent the fusion weight coefficients of texture features and spectral intensity features, respectively, satisfying 0 < <1. By adjusting these two weights, the contribution of texture and spectral intensity features to the texture-spectral composite feature can be adjusted to suit the emphasis of different application scenarios, such as focusing more on ground structure or vegetation spectral information. For example, to balance the contribution of the two features, the following settings can be configured: =0.5.

[0120] S402. Based on the distribution of composite feature values ​​of each local image block, divide all local image blocks into multiple feature region types and assign an adaptive denoising weight to each feature region type.

[0121] Among them, the feature region type is used to characterize a set of image patches with similar texture-spectral complexity features, and different types of feature regions correspond to different denoising intensity requirements.

[0122] In one possible implementation, the image denoising module 12 specifically implements S402 through the following three sub-steps: (1) Determine the feature threshold range based on the numerical distribution of the texture-spectral composite feature values ​​of all local image patches; For example, in this step, the image denoising module 12 uses histogram analysis to statistically analyze the normalized texture-spectral composite feature values, determine the feature threshold, and classify the region types: statistically analyze all... of Distribution, extract the 25th percentile as the first threshold Extract the 75th percentile as the second threshold. .in Characterizing low eigenvalue boundaries, Characterizes the boundary of high eigenvalues ​​and satisfies 0 < < <1.

[0123] (2) Based on the feature threshold range, local image blocks are divided into three feature region types: feature uniform region, feature transition region, and feature complex region; For example, the image denoising module 12 according to and , The size relationship, all Divided into three types of feature regions .

[0124] Where R1 represents the first type of feature region: satisfying < This corresponds to a set of image patches with highly consistent texture-spectral features and minimal variation. Typical land features include large bodies of water and bare land. These areas lack complex vegetation details and require high noise reduction intensity, necessitating complete noise elimination. R2 represents the second type of feature region: satisfying... ≤ This corresponds to a set of image patches with moderate texture or spectral variation and a high number of mixed features. Typical features include forest edges, grasslands, and transition zones between bare land and forests. These areas contain some vegetation detail, requiring a balance between denoising and detail preservation. R3 represents the third type of feature region: satisfying... > This corresponds to a collection of image patches with strong textures, large spectral fluctuations, and mixed ground features. Typical ground features include continuous forest areas, building edges, and areas where vegetation intertwines. These areas have rich vegetation details, low requirements for noise reduction intensity, and need to focus on preserving details.

[0125] (3) Assign an adaptive denoising weight to each feature region type.

[0126] Optionally, a first-level denoising weight coefficient is assigned to the feature uniform region, a second-level denoising weight coefficient is assigned to the feature transition region, and a third-level denoising weight coefficient is assigned to the feature complex region; wherein, the first level is greater than the second level, and the second level is greater than the third level.

[0127] Exemplarily, the image denoising module 12 allocates corresponding adaptive denoising weights for each type of region according to the feature characteristics and denoising requirements of the three types of feature regions The weight size directly represents a denoising intensity coefficient (the greater the weight, the stronger the low-rank decomposition denoising constraint, and the more complete the denoising) The calculation formula of the adaptive denoising weight is as follows: When k takes values 1, 2 and 3 respectively, corresponding to the first level, the second level and the third level respectively.

[0128] It can be understood that the size of different weights determines the strength of the low-rank constraint, thereby affecting the degree of denoising and the fidelity of feature details. For the region, the spatial features are relatively smooth and there are few high-reflectance waveband features of vegetation, and there are few feature regions containing vegetation coverage quantification, so a larger weight can be set to strengthen the low-rank constraint, so that the matrix tends to be low-rank, and random noise is suppressed to the maximum extent to obtain a smoother reconstruction effect. For the region, such as sparse vegetation and light shadow coverage, a medium weight needs to be given to prevent noise residue and avoid excessive smoothing of local details. For the region, the feature characteristics of the features are often rich in texture and complex in vegetation features on the spectral curve and texture features. If a larger low-rank penalty is still applied, the real detail information may be excessively smoothed, resulting in blurred feature boundaries. Therefore, in this type of region, the weight needs to be reduced to preserve complex texture and detailed feature information.

[0129] Based on the above technical solutions, the embodiment of the present application can output the three types of feature region division results and the corresponding adaptive denoising weights, provide differentiated denoising constraint basis for the low-rank decomposition target function, and ensure that the low-rank decomposition can accurately adapt to the characteristics of different feature regions, completely eliminate noise, and completely preserve key vegetation details, thereby providing high-quality image support for subsequent high-precision orthographic image construction and vegetation coverage quantification.

[0130] Exemplarily, in combination with Figure 2 , as shown in Figure 5 , the present application provides another vegetation coverage quantification analysis method based on GIS and unmanned aerial surveying and mapping. In this method, a low-rank decomposition model is constructed according to the edge consistency constraint term and the adaptive denoising weight, and a plurality of denoised hyperspectral images are obtained by solving the low-rank decomposition model, which specifically includes the following steps: S501, constructing a low-rank decomposition model according to the edge consistency constraint term and the adaptive denoising weight.

[0131] In a possible implementation, the image denoising module 12 specifically implements S501 by the following two sub-steps: (1) constructing a low-rank decomposition objective function comprising a low-rank matrix term, a weighted sparse noise term, and an edge consistency constraint term; wherein the low-rank matrix term is used to constrain the low-rank characteristics of the image signal, the weighted sparse noise term is used to represent the sparsity of the noise signal, and the weighted sparse noise term is weighted by an adaptive denoising weight.

[0132] It should be noted that the core of the low-rank decomposition model in S501 is to decompose the noisy hyperspectral image into a low-rank matrix (representing the inherent characteristics of the ground object) and a sparse noise matrix (representing random noise), and to incorporate an edge consistency constraint term and an adaptive denoising weight to construct a low-rank decomposition objective function as follows: wherein, is the low-rank matrix of the region , represents the denoised image signal, is the nuclear norm, is the texture-spectrum feature adaptive weight, is the sparse matrix of the region , represents the noise signal, is the L1 norm, is the edge consistency constraint term of the hyperspectral image . It can be understood that, i.e., the weighted sparse noise term.

[0133] (2) constructing a low-rank decomposition model according to the low-rank decomposition objective function.

[0134] For example, the image denoising module 12 defines the low-rank decomposition objective function and a decomposition equation constraint as a low-rank decomposition model. The decomposition equation constraint is used to ensure the conservation of the total amount of image information before and after denoising.

[0135] S502, solving a plurality of denoised hyperspectral images according to the low-rank decomposition model.

[0136] In a possible implementation, the image denoising module 12 specifically implements S502 by the following three sub-steps: (1) using the augmented Lagrange multiplier method to iteratively solve the low-rank decomposition objective function; Exemplarily, the image denoising module 12 needs to complete the initialization of parameters and variables before solving the low-rank decomposition model to ensure the smooth start of iteration. The total number of iterations is set to 50, which is verified by experiments to balance the solving accuracy and computing efficiency, which can avoid incomplete decomposition caused by insufficient number of iterations, and prevent resource waste caused by excessive iteration. The initial iteration round is counted from 1; the initial value of the low-rank matrix is set to the original hyperspectral image, because the original image already contains the main features of the ground objects, so taking it as the starting point can speed up the convergence process of the low-rank component; the noise matrix is initialized to a zero matrix, assuming that the noise has not been separated from the image in the initial state. The Lagrange multiplier is also initialized to a zero matrix, which is used as a dual variable to constrain the decomposition error in the subsequent iteration; the initial value of the penalty parameter is set to 10, which controls the rising speed of the dual variable, and a small value will lead to slow convergence, and a large value may cause iteration oscillation, and 10 is the optimal initial value verified by tests.

[0137] (2) updating the estimated values of the low-rank matrix and the sparse matrix in each iteration process until the convergence condition is met; Exemplarily, in the tthiteration of the image denoising module 12, the noise matrix and the Lagrange multiplier of the previous round are fixed, and the low-rank matrix is optimized by solving the sub-problem. The objective function of the sub-problem needs to meet three requirements: first, to constrain the decomposition error and ensure that the sum of the low-rank matrix and the noise matrix is as close to the original image as possible; second, to ensure the low-rank property of the low-rank matrix by minimizing the nuclear norm, and to highlight the correlation of the inherent features of the ground objects; third, to maintain the continuity of iteration and avoid drastic fluctuations of the low-rank matrix in iteration. The solving process adopts the singular value threshold algorithm: first, singular value decomposition is performed on the intermediate matrix obtained by fusing the original image, the noise matrix and the Lagrange multiplier information of the previous round to obtain an orthogonal matrix and a singular value diagonal matrix; then, threshold processing is performed on each singular value, and the singular values greater than a certain threshold (representing the main features of the ground objects) are retained, and the singular values less than the threshold (suppressing noise interference) are set to zero; finally, the low-rank matrix optimized in this round is obtained by reconstructing the orthogonal matrix and the processed singular value diagonal matrix, and the accurate iterative update of the low-rank component is realized.

[0138] (3) the final low-rank matrix is output as the denoised hyperspectral image.

[0139] Exemplarily, the image denoising module 12 fixes the low-rank matrix after the current round of optimization and the Lagrange multiplier of the last round, and optimizes the noise matrix through a regional soft threshold algorithm. The objective function of the sub-problem needs to be fused with multiple constraints: first, control the decomposition error to ensure that the noise matrix can accurately represent the difference between the original image and the low-rank matrix; second, combine the adaptive denoising weight to apply differential suppression to noise in different feature regions - high-intensity constraint is used for regions with simple texture-spectrum features, and low-intensity constraint is used for regions with complex features, and the sparse characteristics of L1 norm are used to promote the concentrated distribution of noise; third, the edge consistency constraint is included to ensure that noise processing will not damage the brightness and spectral continuity of the splicing edge. When solving, the corresponding threshold is calculated for each type of feature region, and the intermediate matrix fused with the original image, the low-rank matrix and the Lagrange multiplier information of the current round is subjected to soft threshold operation in each region: elements greater than the threshold are subtracted by the threshold, elements less than the negative threshold are added by the threshold, and the remaining elements are set to zero, which effectively removes noise and avoids damaging the details of the ground object; at the same time, the edge constraint term is optimized by the gradient descent method to ensure that the splicing edge remains continuous and consistent after noise processing, and finally the noise matrix after the current round of optimization is obtained.

[0140] Based on the above technical solution, the embodiment of the present application breaks through the limitation of traditional low-rank decomposition global processing that cannot adapt to the heterogeneity of ground objects and the splicing edge is easy to be lost by constructing a low-rank decomposition model fused with edge consistency constraint and adaptive denoising weight, which not only retains the inherent features of ground objects through low-rank constraint, but also realizes differential denoising in different regions with the help of adaptive weight and edge constraint to ensure consistency at the splicing place; through iterative optimization of the low-rank matrix and the noise matrix, the features of the ground object and the noise are accurately separated - strong denoising is applied to the texture-spectrum simple region to completely suppress the noise, and weak constraint is applied to the complex region to completely retain the texture of the vegetation, and the balance between solution accuracy and efficiency is controlled through iteration; the final denoised hyperspectral image has high definition and spatial consistency, without noise residue or loss of key details of vegetation, and the splicing edge has no brightness sudden change or spectral fault, which provides a reliable data source for subsequent high-precision orthographic image generation and vegetation coverage quantification, and effectively reduces the quantization error caused by image quality defects.

[0141] Exemplarily, in combination with Figure 2 As shown in Figure 6 , the present application provides another vegetation coverage quantification analysis method based on GIS and unmanned aerial vehicle surveying and mapping, which registers and fuses the denoised multiple hyperspectral images with GIS data to generate high-precision orthographic images of the target region, and calculates the vegetation coverage based on the orthographic images, specifically including the following steps: S601, using the digital surface model in the GIS data, performing geometric correction and splicing processing on the denoised multiple hyperspectral images to generate orthographic images containing accurate geographic coordinates.

[0142] The coverage calculation module 13 receives the denoised hyperspectral image, performs geometric correction using the GIS data output by the data acquisition module 11, eliminates image distortion caused by terrain undulations and the flight attitude of the UAV, and then splices the corrected image to generate a target area orthophoto with consistent radiation and accurate coordinates, thereby providing a unified base map for vegetation classification and coverage calculation.

[0143] In S602, vegetation index features are extracted based on the orthophoto, and a preset classification algorithm is used to distinguish between vegetation areas and non-vegetation areas.

[0144] For example, the coverage calculation module 13 extracts vegetation features such as normalized vegetation index based on the orthophoto, and uses random forest classification algorithm to distinguish between vegetation areas and non-vegetation areas in combination with auxiliary information such as terrain and soil type in the GIS data, and the classification result provides a pixel-level classification basis for coverage inversion.

[0145] In S603, the vegetation coverage of each pixel is calculated through a pixel dichotomy model, and the overall vegetation coverage of the target area is obtained by statistics.

[0146] For example, the coverage calculation module 13 first extracts near-infrared band and red band reflectance data sensitive to vegetation features based on the high-precision orthophoto generated from the denoised hyperspectral image. The near-infrared band reflectance can reflect the strong reflection characteristics of vegetation leaves to near-infrared light, and the red band reflectance can reflect the strong absorption characteristics of vegetation leaves to red light, and the combination of the two can effectively distinguish between vegetation and non-vegetation areas.

[0147] Then, the normalized vegetation index NDVI of each pixel is calculated, which is obtained by dividing the difference between the near-infrared band reflectance and the red band reflectance by the sum of the reflectances of the two bands. The NDVI value can directly reflect the vegetation properties of the pixel. The NDVI of the vegetation area is usually high, and the NDVI of the non-vegetation area is usually low. Then, the threshold value required by the pixel dichotomy model is determined. The minimum NDVI value of the pure bare land pixel in the orthophoto is taken as the bare land threshold value (usually not more than 0.1), and the maximum NDVI value of the high-density vegetation area is taken as the vegetation threshold value (usually not less than 0.8).

[0148] Based on the pixel bisection model, it is assumed that each pixel is composed of only two parts of vegetation and non-vegetation: if the pixel NDVI is lower than the bare land threshold, it is determined as completely non-vegetation, and the fractional vegetation cover (FVC) is 0; if the pixel NDVI is higher than the vegetation threshold, it is determined as completely vegetation, and the FVC is 1; if the NDVI is between the two thresholds, the proportion of the vegetation part in the pixel is calculated as the FVC according to the relative relationship of the NDVI and the two thresholds. Finally, the arithmetic mean of the FVC of all pixels of the orthographic image of the target region is taken to obtain the overall vegetation coverage of the region, ensuring that the quantitative result is highly matched with the actual vegetation coverage.

[0149] Based on the above technical solution, the embodiments of the present application accurately register the denoised hyperspectral image based on the ground control points in the GIS data, correct the perspective distortion caused by the tilt photography of the unmanned aerial vehicle in combination with the digital surface model in the GIS data, and then generate a high-precision orthographic image through radiation equalization and multi-band fusion, effectively solving the problem of inaccurate positioning of ground objects caused by the deviation of traditional image registration and perspective distortion; subsequently, based on the orthographic image, the vegetation and non-vegetation areas are accurately distinguished through the normalized difference vegetation index, the vegetation coverage is inverted in combination with the pixel bisection method, and the overall coverage of the target region is counted, avoiding the quantitative error caused by the quality defects of the image, and finally obtaining the vegetation coverage with spatial distribution accuracy and numerical accuracy, which can provide reliable quantitative basis for scenes such as ecological monitoring and resource assessment.

[0150] It should be noted that the above-mentioned embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0151] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

Claims

1. A method for quantitative analysis of vegetation cover based on GIS and UAV mapping, characterized in that, The method includes: Multiple hyperspectral images and GIS data of a target area are acquired; wherein, the target area includes multiple sub-regions, each hyperspectral image is used to characterize the spectral reflectance distribution and spatial texture information of ground objects in a sub-region, and the GIS data includes geographic information data for providing spatial positioning reference for each sub-region; An edge consistency constraint term is constructed, which is used to maintain the consistency of radiance and spectral shape between the current hyperspectral image and adjacent images at the stitching edge; Based on multiple hyperspectral images of the target region, the target region is divided into regions and adaptive denoising weights are assigned; wherein, the adaptive denoising weights are used to characterize the denoising intensity coefficients corresponding to regions with different texture-spectral feature complexities; Based on the edge consistency constraint and the adaptive denoising weight, a low-rank decomposition model is constructed, and multiple denoised hyperspectral images are obtained by solving the low-rank decomposition model. The denoised hyperspectral images are registered and fused with the GIS data to generate a high-precision orthophoto of the target area, and the vegetation coverage is calculated based on the orthophoto.

2. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 1, characterized in that, The construction of edge consistency constraints specifically includes: For the current hyperspectral image, all adjacent hyperspectral images are determined based on the geographic coordinates of the corresponding sub-region of the hyperspectral image; wherein, the current hyperspectral image is the hyperspectral image to be denoised, and the adjacent hyperspectral images are the hyperspectral images that are spatially adjacent to the sub-region corresponding to the current hyperspectral image to be denoised. Determine the overlapping regions between the current hyperspectral image and each of the adjacent hyperspectral images; An edge consistency constraint term is constructed based on the brightness difference of corresponding pixels within the overlapping area and the spectral shape difference in different vegetation feature bands.

3. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 2, characterized in that, The step of determining the overlapping region between the current hyperspectral image and each of the adjacent hyperspectral images specifically includes: Based on a preset lateral overlap parameter, the overlap region between the current hyperspectral image and each of the adjacent hyperspectral images is calculated; wherein, the lateral overlap parameter is used to control the overlap ratio between adjacent hyperspectral images.

4. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 3, characterized in that, The step of constructing an edge consistency constraint term based on the brightness differences of corresponding pixels within the overlapping area and the spectral shape differences in different vegetation feature bands specifically includes: Calculate the brightness difference value of corresponding pixels in the overlapping area as a brightness consistency component; The spectral shape difference value of the corresponding pixel points in the overlapping area on the preset vegetation feature band is calculated as the spectral consistency component; wherein, the preset vegetation feature band includes at least one of the near-infrared band, short-wave infrared band and red light band that are sensitive to vegetation reflectance characteristics. The brightness consistency component and the spectral consistency component are weighted and fused according to a preset weight to construct the edge consistency constraint term.

5. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 1, characterized in that, The step of dividing the target region into regions and assigning adaptive denoising weights based on multiple hyperspectral images of the target region specifically includes: Each hyperspectral image of the target region is divided into multiple local image blocks, and the texture-spectral composite feature value of each local image block is calculated; wherein, the texture-spectral composite feature value is used as a comprehensive quantitative index to characterize the pixel texture complexity and spectral reflectance intensity in a preset feature band within the local image block; Based on the distribution of the composite feature values ​​of each local image patch, all local image patches are divided into multiple feature region types, and an adaptive denoising weight is assigned to each feature region type; wherein, the feature region type is used to characterize a set of image patches with similar texture-spectral complexity features, and different types of feature regions correspond to different denoising intensity requirements.

6. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 5, characterized in that, The step of dividing all local image blocks into multiple feature region types based on the distribution of the composite feature values ​​of each local image block, and assigning an adaptive denoising weight to each feature region type, specifically includes: The feature threshold range is determined based on the numerical distribution of the texture-spectral composite feature values ​​of all local image patches; Based on the aforementioned feature threshold range, local image blocks are divided into three feature region types: feature uniform region, feature transition region, and feature complex region. A first-level denoising weight coefficient is assigned to the uniform feature region, a second-level denoising weight coefficient is assigned to the transitional feature region, and a third-level denoising weight coefficient is assigned to the complex feature region; wherein the first level is greater than the second level, and the second level is greater than the third level.

7. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 6, characterized in that, The step of constructing a low-rank decomposition model based on the edge consistency constraint term and the adaptive denoising weights specifically includes: A low-rank decomposition objective function is constructed, comprising a low-rank matrix term, a weighted sparse noise term, and the edge consistency constraint term. The low-rank decomposition model is then constructed based on the low-rank decomposition objective function. The low-rank matrix term is used to constrain the low-rank characteristics of the image signal, the weighted sparse noise term is used to characterize the sparsity of the noise signal, and the weighted sparse noise term is weighted by the adaptive denoising weights.

8. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 7, characterized in that, The process of obtaining multiple denoised hyperspectral images based on the low-rank decomposition model specifically includes: The low-rank decomposition objective function is solved iteratively using the augmented Lagrange multiplier method. In each iteration, the estimates of the low-rank matrix and the sparse matrix are updated until the convergence condition is met; The resulting low-rank matrix is ​​then output as the denoised hyperspectral image.

9. The vegetation cover quantitative analysis method based on GIS and UAV mapping according to claim 8, characterized in that, The step of registering and fusing the denoised hyperspectral images with the GIS data to generate a high-precision orthophoto of the target area, and calculating the vegetation cover based on the orthophoto, specifically includes: Using the digital surface model in the GIS data, the denoised hyperspectral images are geometrically corrected and stitched together to generate an orthophoto containing accurate geographic coordinates. Based on the orthophoto, vegetation index features are extracted, and a preset classification algorithm is used to distinguish between vegetated areas and non-vegetated areas. The vegetation coverage of each pixel is calculated using a pixel-based binary model, and the overall vegetation coverage of the target area is statistically obtained.

10. The method for quantitative analysis of vegetation cover based on GIS and UAV mapping according to any one of claims 1-9, characterized in that, After acquiring multiple hyperspectral images and GIS data of the target area, the method further includes: Preprocessing operations are performed on multiple hyperspectral images of the target area and the GIS data; wherein, the preprocessing operations include atmospheric correction of the multiple hyperspectral images, coordinate unification and topology repair of the GIS data.

Images