Forest carbon sink estimation system based on unmanned aerial vehicle-mounted hyperspectral remote sensing technology

Abstract

The invention discloses a forest carbon sink estimation system based on an unmanned aerial vehicle-mounted hyperspectral remote sensing technology, and relates to the technical field of remote sensing. After canopy fragments are divided based on LiDAR data, a structure disturbance area is identified through a clustering method, a spatial path attribute matrix is further constructed, and the forest carbon sink estimation method is established by combining the adjacency relation between the canopy fragments. Calculating a canopy disturbance index of the jth canopy segment and performing disturbance correction to solve the problem of error accumulation of the reflectivity value caused by structural disturbance in the spatial propagation process; on the basis, synchronously collected wind speed data is introduced, a path intensity correction coefficient is introduced, the canopy disturbance index of the jth canopy segment is corrected, and a second canopy disturbance index is adopted to reconstruct a canopy thermodynamic diagram after disturbance suppression; and finally, taking the corrected NDVI spatial distribution diagram as input, coupling canopy biomass parameters to carry out forest carbon reserve estimation, and enhancing robustness and adaptability in a dynamic disturbance scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing, in particular to a forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology. BACKGROUND

[0002] In the traditional forest carbon sink estimation method, satellite remote sensing, ground sample survey or aerial image are usually relied on for forest coverage and biomass evaluation. However, in practical application, the forest canopy structure is complex, especially in coniferous forest, mixed forest and other forest types, the canopy undulates unevenly, the spatial heterogeneity is strong, and is easily affected by natural disturbance factors, resulting in the decrease of remote sensing inversion accuracy.

[0003] On the one hand, in the process of using high-spectral remote sensing data for carbon sink estimation, the reflectivity of the canopy is often significantly disturbed by wind speed. The branch and leaf swing caused by wind speed changes the transient form of canopy structure, resulting in reflectivity fluctuation, structure mismatch and other phenomena between multi-time images, and further affecting the stable extraction of vegetation parameters such as normalized difference vegetation index (NDVI), causing estimation deviation.

[0004] On the other hand, the forest canopy segments are not isolated, but are adjacent and connected through vertical or horizontal paths in different directions. In the existing method, the "path coupling disturbance" is not effectively modeled and processed, so that the interference signal of a certain canopy segment may be incorrectly transmitted or diffused to the adjacent area, further affecting the accuracy of the whole area carbon storage estimation, especially in the areas where the local wind field fluctuates violently or the terrain is complex, the problem is more obvious. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology to solve the problems mentioned in the background art.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology, comprising: A high-spectral image acquisition module is mounted on a UAV platform, flies along a preset route in a forest area at low altitude, acquires a multi-time forest high-spectral image sequence in real time, and uses LiDAR technology to acquire DSM and DTM, performs pixel-level difference operation to obtain a canopy height model CHM, and extracts local average height , local standard deviation and height gradient CH(x,y); An identification module is used to identify the local average height , local standard deviation and height gradient CH(x, y) as input feature vectors, using K-means algorithm to cluster the feature vectors of all pixels, denoted as C for each canopy segment j , identify whether adjacent canopy segments exist spatial adjacency relationship, and belong to vertical or horizontal path, mark the path attribute matrix ; The structure disturbance perception module is used to extract canopy structure disturbance features from the hyperspectral image sequence, identify spectral anomaly areas caused by branch and leaf fluctuation, wind speed disturbance and canopy heterogeneity, calculate the canopy disturbance index of the jth canopy segment , combined with the path attribute matrix , introduce the path intensity correction coefficient , and correct the canopy disturbance index of the jth canopy segment , obtain the second canopy disturbance index of the jth adjacent canopy segment after correction , and build a canopy thermal map; The wind speed-disturbance joint modeling module is used to collect the wind speed V(t) of the micro-meteorological sensor carried by the unmanned aerial vehicle and the second canopy disturbance index of the jth adjacent canopy segment after correction After the disturbance-wind speed joint driving model is established and trained by using the multivariate regression model method, the reflectance error of the jth canopy segment is obtained , combined with the initial reflectance of the jth canopy segment , to output the corrected reflectance output value of the jth canopy segment ; The carbon sink estimation module is based on the corrected reflectance output value , uses the NDVI inversion model to reacquire the normalized vegetation index corresponding to the jth corrected canopy segment , and the spatial coverage area , outputs the carbon storage of the jth canopy segment .

[0007] Preferably, the hyperspectral image acquisition module is used to deploy an unmanned aerial vehicle carrying a spectral sensor to collect a multi-temporal forest hyperspectral image sequence passing through the space above the canopy layer. The wavelength range of the spectral sensor is 400-1000 nm, including the red edge and near-infrared channels.

[0008] Preferably, the hyperspectral image acquisition module includes an image acquisition unit, a canopy height model establishment unit, and a feature extraction unit. The image acquisition unit is used to deploy a UAV equipped with a spectral sensor to acquire a series of hyperspectral images of multi-temporal forests passing through the canopy. The hyperspectral image series are preprocessed and corrected, and after registration processing of different spectral bands, the pixel correspondence is unified. The image acquisition timestamp, flight altitude, light intensity, sensor attitude information and synchronous metadata are also labeled. The canopy height model building unit is used to deploy a UAV equipped with a lightweight LiDAR or to generate three-dimensional canopy structure data using multi-angle image reconstruction methods. It acquires the DSM and DTM, performs pixel-level interpolation calculations, and obtains the canopy height model (CHM). Calculate the canopy height H(x,y). Where (x,y) represents the position of a pixel in the image. This indicates that the Digital Surface Model (DSM) includes the surface elevations of tree canopies, buildings, and all other objects. This indicates that the Digital Terrain Model (DTM) only represents ground elevation; The feature extraction unit performs fixed-size sliding window processing on the canopy height model (CHM), including a 5×5 pixel or 3×3 pixel window, to extract local structural features of each center pixel, including: Local average height , This reflects the overall canopy height level in the region; Local standard deviation , This reflects the degree of undulation of the canopy in that region; Height gradient Describes the rate of change of pixel height in the horizontal and vertical directions, calculated using the Sobel operator: ; Where H(i,j) represents the canopy height value at position (i,j) within the window, W(x,y) represents the sliding window centered on the central pixel (x,y), and N represents the number of pixels in the sliding window; The local height standard deviation within a sliding window region centered at pixel (x,y); height gradient. Measuring the rate of height change of a pixel location in the horizontal x and vertical y directions is used to identify regions of structural abrupt change at the canopy edge or boundary. and The gradients of the canopy height model (CHM) in the x and y directions are respectively calculated using the Sobel operator: .

[0009] Preferably, the recognition module comprises a clustering unit and a canopy segment adjacency attribute recognition unit; The clustering unit is used for clustering feature vectors of all pixels based on local mean height , local standard deviation and height gradient CH(x, y) as input feature vectors, and outputting a class label for each pixel, comprising: If the local mean height is greater than 0.7 meters, the local standard deviation is less than 0.15 meters, and GH(x, y) is less than 0.15 meters / pixel, the canopy structure of the region is uniform and consistent in height, and the region is determined as a main canopy area; If the local mean height is lower than 0.4 meters, the local standard deviation is greater than 0.25 meters, and GH(x, y) is greater than 0.35 meters / pixel, it indicates that there is a significant height mutation in the region, which is commonly seen in the transition between the canopy and the ground, and the region is determined as a gap boundary area; If the local mean height is between 0.4 and 0.7 meters, the local standard deviation is greater than 0.2 meters, and GH(x, y) is greater than 0.3 meters / pixel, it indicates that the height difference in the region is large and the structure is chaotic, and the region is determined as a high-low mixed canopy area.

[0010] Preferably, the canopy segment adjacency attribute recognition unit is used for dividing all canopy areas into a plurality of canopy segments. After the structural feature extraction and K-means clustering classification of the pixels in the entire forest area are completed, the entire region is divided into a plurality of spatially continuous canopy segments according to the clustering results, and each canopy segment is recorded as C j , corresponding to a functional canopy of each class; Each canopy class canopy segment C j is taken as a node of a canopy segment adjacency graph, and if the boundary of two adjacent canopy segments C j and C k has a contact point or the distance between adjacent pixels is less than 3 pixel points, an edge is added in the canopy segment adjacency graph, indicating that the two canopy segments C j and C k have a spatial adjacency relationship, so as to establish an adjacency matrix A jk ; if C j and C k are adjacent, A jk =1, otherwise 0; Based on the structure of the adjacency matrix A jk , the spatial structure type and interference path attribute between each pair of adjacent canopy segments are further refined, and a path matrix P with attributes is constructed.jk ; If C j is adjacent to C k , the average height of C j is greater than the average height of C k , and a number of crown layer segments are projected onto a two-dimensional plane to obtain a 2D boundary, if C j and C k overlap in the boundary area of the 2D boundary, it is determined that C j is adjacent to C k , which belongs to a vertical path, and the label C j and C k constitute a vertical path relationship, denoted as: ; When C j and C k do not overlap in the boundary area of the 2D boundary, the horizontal execution distance : ; of the two adjacent crown layer segments C j and C k is calculated, and a spatial adjacency threshold Td is preset, and when is less than the spatial adjacency threshold Td, it is determined that the horizontal adjacency is horizontal, and a horizontal path relationship is constituted, denoted as: ; A path attribute matrix is constructed: .

[0011] Preferably, the structure disturbance perception module includes a spectral stability weight calculation unit, a structure disturbance susceptibility index calculation unit and a canopy transmittance estimation unit. The spectral stability weight calculation unit is used to extract a reflectance sequence at multiple time points for each crown layer segment C j , and calculate a reflectance standard deviation to obtain a spectral stability weight of the jth crown layer segment : ; ; Wherein, represents the reflectance standard deviation of the jth crown layer segment in the spectral main wave band, represents the reflectance of the jth crown layer segment at the tth time point, represents the mean value of the reflectance of all time points of the segment, and m represents the number of time phases. The structure disturbance susceptibility index calculation unit is used to calculate the structure disturbance susceptibility index of the jth crown layer segment : In the formula, represents the void ratio of the jth canopy segment, wherein, represents the number of ground-penetrating and viewpoint numbers in the jth canopy segment, which is obtained based on laser radar (LiDAR) recognition; represents the total ground echo number of the jth canopy segment, including the canopy and the ground, represents the difference between the jth canopy segment and the average height around it, represents the real-time wind speed value at the time of observation; and represents the weight value, , and are normalized values; The canopy transmittance estimation unit is configured to calculate the normalized vegetation index of the jth canopy segment . . NIR represents the reflectivity of the near-infrared band, and RED represents the reflectivity of the red light band. Then, the canopy transmittance of the jth canopy segment is calculated by combining the void ratio of the jth canopy segment and the normalized vegetation index of the jth canopy segment : . The canopy transmittance of the jth canopy segment is used to describe the ability of light to penetrate the canopy segment, which is related to its spectral reflectivity and porosity ratio.

[0012] Preferably, the structure disturbance perception module further comprises an association unit, which is configured to combine the spectral stability weight of the jth canopy segment, the structure disturbance susceptibility index , and the canopy transmittance to calculate the canopy disturbance index of the jth canopy segment by the following association formula: . wherein, , and are weight coefficients, respectively.

[0013] Preferably, the path weight correction unit is configured to, for the jth canopy segment, and for any two adjacent canopy segments C j and C​​k , the path intensity correction coefficient : wherein, , and are the spectral stability weight, the structural disturbance susceptibility index and the canopy transmittance coefficient of the kth canopy segment adjacent to the jth canopy segment, respectively; based on the path intensity correction coefficient , combined with the canopy disturbance index of the jth canopy segment , the canopy disturbance index of the jth canopy segment is corrected for the joint action of the node and the path to obtain the second canopy disturbance index adjacent to the jth canopy segment after correction , specifically: wherein, represents the canopy disturbance index of the kth canopy segment adjacent to the jth canopy segment, represents the index set of all segments adjacent to the jth canopy segment; the meaning of the formula is that, the disturbance transmission influence of adjacent segments is considered, reflecting the current disturbance state of the canopy segment, combined with the disturbance effect of the surrounding segments; and according to the second canopy disturbance index adjacent to the jth canopy segment after correction is divided into four levels, when is not in level I, a correction instruction is triggered, and a canopy thermal map is constructed according to the second canopy disturbance index adjacent to the jth canopy segment after correction .

[0014] Preferably, the wind speed-disturbance joint modeling module comprises a real-time wind speed acquisition unit, a modeling unit and a reflectivity correction unit. The real-time wind speed acquisition unit is used for the unmanned aerial vehicle platform to carry a micro meteorological sensor module, to acquire a wind speed sequence V(t) above the target region in real time during the flight task, and to form a wind speed-canopy segment corresponding mapping by pairing with the image acquisition time through a synchronous time stamp. The modeling unit is used for the second canopy disturbance index adjacent to the jth canopy segment has been constructed, and the wind speed sequence V(t) at the corresponding time point is used as an input variable together, and after a disturbance-wind speed joint driving model is established and trained by using a multivariate regression model method, the reflectivity error of the jth canopy segment is obtained: ; wherein, represents a constant term in a multiple linear regression model, and represents a regression coefficient, represents a residual term. a reflectance correction unit configured to output a corrected reflectance output value of the jth canopy segment based on a reflectance error of the jth canopy segment and an initial reflectance of the jth canopy segment . : .

[0015] Preferably, the carbon sink estimation module comprises a parameter matching unit and a carbon storage estimation unit. The parameter matching unit is configured to receive the corrected reflectance output value of the jth canopy segment , comprising: ; wherein, and represent the corrected near-infrared band reflectance and red band reflectance of the jth canopy segment. The carbon storage estimation unit is configured to reacquire a normalized vegetation index corresponding to the jth canopy segment after correction according to a preset forest type and spectral inversion relationship using an NDVI inversion model and a spatial coverage area to calculate the carbon storage of the jth canopy segment : ; ; ; wherein, represents the biomass value per unit area of the jth segment, and b are empirical model coefficients established in advance according to different Selina types; represents a carbon conversion coefficient, which is set to 0.5; spatially splicing the carbon storage of all canopy segments to output a distribution map of carbon storage per unit area, and the carbon storage can be statistically aggregated according to administrative divisions, forest types or ecological function zones.

[0016] The present application provides a forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology. The system has the following advantages: (1) The forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology first obtains high-spectral images and laser radar point clouds through an unmanned aerial vehicle platform, and extracts crown layer structure data in a target region; based on LiDAR data, the crown layer segments are divided, the average height, standard deviation, height gradient and other parameters of each segment are obtained, the structure disturbance region is identified by a clustering method; a spatial path attribute matrix is further constructed, the path disturbance index is calculated and the disturbance is corrected in combination with the adjacency relationship between the crown layer segments, the error accumulation problem caused by the spatial propagation of the structure disturbance on the reflectivity value is solved; on this basis, the simultaneously collected wind speed data is introduced, and then the spectral stability weight , the structure disturbance susceptibility index and the crown layer transmittance coefficient are combined to obtain the crown layer disturbance index of the jth crown layer segment, and the path strength correction coefficient is introduced in combination with the path attribute matrix , and the crown layer disturbance index of the jth crown layer segment is corrected to obtain the second crown layer disturbance index adjacent to the jth corrected crown layer segment, and the crown layer thermal map after disturbance inhibition is reconstructed; finally, the corrected NDVI spatial distribution map is taken as input, the forest carbon storage is estimated by coupling the crown layer biomass parameters, and the robustness and adaptability of the estimation model under the dynamic disturbance scene are enhanced.

[0017] (2) The forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology takes the local average height, local standard deviation and height gradient GH(x, y) of each pixel as input features, uses the K-means algorithm for unsupervised clustering of the entire region, and then realizes automatic classification of the crown layer spatial structure. In the clustering result, if the local average height of a region is greater than 0.7 meters, and the standard deviation and height gradient are small, the height distribution of the region is stable, the structure is neat, and it is determined as the main crown layer area; if the local average height is lower than 0.4 meters, but accompanied by large standard deviation and sudden gradient, it means that there is ground exposure, branch drop or obvious gap in the region, and it is determined as the gap boundary area; when the local average height is at an intermediate level, but the standard deviation and gradient are large, it reflects that the height of the region fluctuates significantly, which is often the overlap of low vegetation and main crown layer or disturbed patch, and is classified as high-low mixed crown layer area. This classification method automatically partitions the structure features, does not rely on manual determination or external labeling, can effectively divide the potential structure stable region and easy disturbance region, lays a foundation for subsequent crown layer disturbance index calculation, anomaly detection and disturbance identification, and thus improves the pertinence and accuracy of the response of the forest canopy disturbance.

[0018] (3) The forest carbon sink estimation system based on the unmanned aerial high-spectral remote sensing technology, a crown layer segment adjacency attribute identification unit, based on the aforementioned pixel-level clustering result, divides the entire forest area into a plurality of spatially continuous crown layer segments, each segment corresponding to a functional crown layer area in the clustering result, denoted as C j . By extracting the boundary contact relationship between each pair of crown layer segments, an adjacency graph is constructed. In this invention, the adjacency relationship of the crown layer segments is identified to reveal the path mechanism of disturbance propagation in the forest structure. By spatially segmenting and adjacency analyzing the functional crown layer areas after clustering, each crown layer segment is taken as a node in the graph, and its adjacency relationship is judged according to its two-dimensional boundary overlap and three-dimensional elevation difference, and then a path attribute matrix containing "adjacent" and "how adjacent" is constructed. This structure not only distinguishes between vertical propagation paths (such as upper crown layer collapse transmission to lower space) and horizontal propagation paths (such as crown layer continuous lodging area along the wind direction axis), but also clearly indicates the direction and intensity of disturbance propagation along the path. This method can realize the structured identification of disturbance risk propagation channels in the forest without relying on deep models, based only on height and structural features, thereby providing accurate data basis for local disturbance response, intervention priority ranking and spatial connectivity analysis. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The figure is a process schematic diagram of the forest carbon sink estimation system based on the unmanned aerial high-spectral remote sensing technology of the present invention. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present invention.

[0021] Embodiment 1 Please refer to Figure 1 , the present invention provides a forest carbon sink estimation system based on unmanned aerial high-spectral remote sensing technology, comprising: A high-spectral image acquisition module is mounted on a UAV platform, flies at low altitude along the forest area setting route, obtains a multi-temporal forest high-spectral image sequence crossing the forest canopy layer in real time, and obtains DSM and DTM using laser radar LiDAR technology, performs pixel-level difference operation to obtain a crown layer height model CHM, and extracts local average height , local standard deviation and height gradient CH(x, y); An identification module is used to identify the local average height local standard deviation and height gradient CH(x, y) as input feature vectors, using the K-means algorithm to cluster the feature vectors of all pixels, denoted as C j , identify whether adjacent canopy segments have spatial adjacency relationships, and belong to vertical or horizontal paths, and mark the path attribute matrix ; The structure disturbance perception module is used to extract canopy structure disturbance features from the hyperspectral image sequence, identify spectral anomaly areas caused by branch and leaf fluctuation, wind speed disturbance and canopy heterogeneity, calculate the spectral stability weight of the jth canopy segment , structure disturbance susceptibility index and canopy transmittance , and calculate the canopy disturbance index of the jth canopy segment through a weighted correlation formula , combined with the path attribute matrix , introduce the path intensity correction coefficient and correct the canopy disturbance index of the jth canopy segment , obtain the second canopy disturbance index of the jth adjacent canopy segment after correction , and construct a canopy thermal map. The wind speed-disturbance joint modeling module is used to combine the wind speed V(t) collected by the micro-meteorological sensor carried by the unmanned aerial vehicle with the second canopy disturbance index of the jth adjacent canopy segment after correction After the disturbance-wind speed joint driving model is established and trained by using a multivariate regression model method, the reflectance error of the jth canopy segment is obtained , combined with the initial reflectance of the jth canopy segment , to output the corrected reflectance output value of the jth canopy segment ; The carbon sink estimation module is based on the corrected reflectance output value , uses the NDVI inversion model to reacquire the normalized vegetation index corresponding to the jth corrected canopy segment and the spatial coverage area , and outputs the carbon storage of the jth canopy segment .

[0022] In this embodiment, first, the hyperspectral image and the LiDAR point cloud are obtained through the unmanned aerial vehicle platform, and the crown layer structure data in the target area is extracted; the crown layer segments are divided based on the LiDAR data, the average height, standard deviation, height gradient and other parameters of each segment are obtained, the structure disturbance area is identified through the clustering method; further, the spatial path attribute matrix is constructed, the path disturbance index is calculated and the disturbance is corrected combined with the adjacency relationship between the crown layer segments, and the error accumulation problem caused by the structure disturbance in the spatial propagation process is solved; on this basis, the wind speed data collected synchronously is introduced, and then combined with the spectral stability weight , the structure disturbance susceptibility index and the crown layer transmittance , the crown layer disturbance index of the jth crown layer segment is obtained, and combined with the path attribute matrix , the path intensity correction coefficient is introduced, and the crown layer disturbance index of the jth crown layer segment is corrected, and then the second crown layer disturbance index adjacent to the jth corrected crown layer segment is obtained, and the crown layer thermal map after disturbance inhibition is reconstructed; finally, the corrected NDVI spatial distribution map is taken as the input, the forest carbon storage is estimated by coupling the crown layer biomass parameter, and the robustness and adaptability of the estimation model under the dynamic disturbance scene are enhanced. Without relying on a large number of field sample points, the method realizes high-precision carbon calculation of forest carbon sink under multiple disturbance conditions.

[0023] Embodiment 2 This embodiment is an explanation and description in embodiment 1, please refer to Figure 1 , the hyperspectral image acquisition module is used for deploying the unmanned aerial vehicle to carry the spectral sensor, and acquiring the hyperspectral image sequence of the multi-time forest penetrating through the space above the forest canopy layer. The spectral sensor has a wavelength range of 400nm to 1000nm, including a red edge and a near-infrared channel.

[0024] The hyperspectral image acquisition module includes an image acquisition unit, a crown layer height model establishing unit and a feature extraction unit; The image acquisition unit is used for deploying the unmanned aerial vehicle to carry the spectral sensor, and acquiring the hyperspectral image sequence of the multi-time forest penetrating through the space above the forest canopy layer. The hyperspectral image sequence is preprocessed and corrected, and after registration processing of different spectral bands, the pixel correspondence is unified, and the image acquisition timestamp, flight height, light intensity, sensor attitude information and synchronous metadata are labeled; The crown layer height model establishing unit is used for deploying the unmanned aerial vehicle to carry the lightweight LiDAR or using the multi-angle image reconstruction method to generate the crown layer three-dimensional structure data, obtaining the DSM and DTM, and performing pixel-level difference operation to obtain the crown layer height model CHM: Compute canopy height H(x, y), where (x, y) denotes a pixel position in the image, denotes the surface elevation of all objects including tree canopies, buildings in the digital surface model DSM, denotes the surface elevation of all objects including tree canopies, buildings in the digital surface model DSM, a feature extraction unit for performing fixed-size sliding window processing on the canopy height model CHM, including a 5x5 pixel or 3x3 pixel window, extracting local structural features of each center pixel, including: local mean height ,reflecting the overall canopy height level of the region; local standard deviation ,reflecting the degree of fluctuation of the canopy in the region; height gradient , describing the rate of change of height in horizontal and vertical directions of the pixel, calculated using the Sobel operator: ; where H(i, j) denotes the canopy height value at position (i, j) in the window, W(x, y) denotes the sliding window centered at the center pixel (x, y), and N denotes the number of pixels in the sliding window; is the local height standard deviation in the sliding window region centered at the center pixel (x, y); height gradient measures the rate of change of height in the horizontal x and vertical y directions of the pixel position, used to identify canopy edge or interface structure mutation area, and are the gradients of the canopy height model CHM in the x and y directions, respectively, calculated by the Sobel operator: .

[0025] In this embodiment, the method is based on the joint perception principle of hyperspectral and canopy structure. By deploying unmanned aerial vehicles carrying spectral sensors, LiDAR or multi-angle image systems, multi-temporal, multi-band and multi-angle forest canopy information is obtained. The hyperspectral image acquisition module includes an image acquisition unit, a canopy height model establishment unit and a feature extraction unit. The image acquisition unit is responsible for acquiring image sequences with a wavelength range of 400-1000 nm, and performing image registration, timestamp and attitude information synchronization labeling. The canopy height model establishment unit constructs a digital surface model (DSM) and a terrain model (DTM) through LiDAR or image reconstruction technology, and obtains a canopy height model (CHM) by pixel-level difference calculation, which is used to represent the canopy structure. The feature extraction unit performs sliding window operation based on the CHM to extract local average height, standard deviation and horizontal / vertical direction gradient, reflecting the canopy fluctuation state and structural change characteristics. The height gradient is calculated by the Sobel operator to identify the canopy structure boundary and disturbance area. The above structural features and hyperspectral information can be further fused for interference identification or forest dynamic evolution analysis.

[0026] Embodiment 3 This embodiment is an explanation and description in Embodiment 1, please refer to Figure 1 The identification module includes a clustering unit and a canopy segment adjacency attribute identification unit. The clustering unit is used to cluster the feature vectors of all pixels using the K-means algorithm based on the local average height , the local standard deviation and the height gradient CH(x, y) as input feature vectors, and output a class label for each pixel, including: If the local average height is greater than 0.7 meters, the local standard deviation is less than 0.15 meters, and GH(x, y) is less than 0.15 meters / pixel, the canopy structure in this area is uniform and the height is consistent, which is determined as the main canopy area; If the local average height is lower than 0.4 meters, the local standard deviation is greater than 0.25 meters, and GH(x, y) is greater than 0.35 meters / pixel, it indicates that there is a significant height mutation in this area, which is commonly seen in the transition between the canopy and the ground, and is determined as the gap boundary area; If the local average height is between 0.4 and 0.7 meters, and the local standard deviation is greater than 0.2 meters, and GH(x, y) is greater than 0.3 meters / pixel, it indicates that the height difference in this area is large and the structure is chaotic, which is determined as the high-low mixed canopy area.

[0027] In this embodiment, the clustering unit uses the local average height, local standard deviation, and height gradient GH(x,y) of each pixel as input features, and uses the K-means algorithm to perform unsupervised clustering on the entire region, thereby achieving automatic classification of the canopy spatial structure. In the clustering results, if the local average height of a region is greater than 0.7 meters and the standard deviation and height gradient are both small, then the height distribution of the region is stable and the structure is neat, and it is identified as the main canopy area; if the local average height is less than 0.4 meters, but accompanied by a large standard deviation and abrupt gradient, it indicates that there is bare ground, branch drop, or obvious gaps in the region, and it is identified as a gap boundary area; when the local average height is at an intermediate level, but the standard deviation and gradient are large, it reflects that the height of the region is significantly fluctuating, often a patch of low vegetation overlapping or disturbed with the main canopy, and it is classified as a mixed high and low canopy area. This classification method automatically partitions regions based on structural features, without relying on manual judgment or external annotation. It can effectively divide potential structurally stable regions into easily disturbed regions, laying the foundation for subsequent canopy disturbance index calculation, anomaly detection, and disturbance identification, thereby improving the pertinence and accuracy of canopy disturbance response.

[0028] Example 4 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 The canopy segment adjacency attribute identification unit is used to segment all canopy regions into multiple canopy segments. After extracting structural features and performing K-means clustering classification on pixels throughout the forest area, it divides the entire region into multiple spatially continuous canopy segments based on the clustering results. Each canopy segment is denoted as C. j Corresponding to the functional canopy of each type; Canopy fragments C for each canopy category j As a node in the canopy segment adjacency graph, if two adjacent canopy segments C j With C k If there is a contact point at the boundary of two canopy segments, or if the distance between adjacent pixels is less than 3 pixels, then an edge is added to the canopy segment adjacency graph to represent the two canopy segments C. j With C k There are spatial adjacency relationships to establish an "adjacency matrix" A. jk If C j With C k Adjacent, then A jk =1, otherwise 0; Based on the "adjacency matrix" A jk The structure is further refined to determine the spatial structure type and interference path attributes between each pair of adjacent canopy segments, and a path matrix P with attributes is constructed. jk It is not only necessary to know "adjacent", but also to know "how they are adjacent"; for example, the space between the upper canopy area and the lower void. If C j With Ck adjacent, C j has a larger average height than C k projects several canopy segments to a two-dimensional plane to obtain a 2D boundary, that is, the boundary of the ground, if C j and C k overlap in the boundary region of the 2D boundary, it is determined that C j and C k are adjacent to the vertical path, and the label C j and C k constitute a vertical path relationship, denoted as: ; When C j and C k do not overlap in the boundary region of the 2D boundary, the horizontal execution distance j between two adjacent canopy segments C k and C is calculated: ; and a spatial adjacency threshold Td is preset, for example, 1.5 times the canopy radius or 1 meter set empirically, when is less than the spatial adjacency threshold Td, it is determined that the horizontal adjacency is horizontal, and a horizontal path relationship is constituted, denoted as: ; A path attribute matrix is constructed : .

[0029] When a disturbance (such as wind, insect pests, disease spots, etc.) occurs in a forest area, if a canopy segment is damaged, its influence is often transmitted to adjacent segments through a "path". If this transmission path is in the vertical direction (such as the collapse of the upper canopy to the lower layer), it will cause changes in the local shading relationship, exposure of the lower vegetation or bare land, and changes in the shortwave infrared and near-infrared bands of the reflectivity. If the transmission path is in the horizontal direction (such as the inclination of adjacent canopies or the degradation of the connecting area), it will expand the continuous disturbance area, change the canopy density and leaf area of the region, cause overall brightness changes or spectral distortion of the spectrum, and cause tree species replacement, uneven illumination, soil background exposure differences, and other factors, so that different segments in the same category exhibit heterogeneous spectral reflectivity due to different transmission paths, thereby affecting the subsequent classification and recognition accuracy In this embodiment, the canopy segment adjacency attribute recognition unit divides the entire forest area into a plurality of spatially continuous canopy segments based on the aforementioned pixel-level clustering results, each segment corresponding to a functional canopy region in the clustering results, denoted as C jBy extracting the boundary contact relationships between each pair of canopy segments and constructing an adjacency graph, this invention identifies the adjacency relationships of canopy segments, aiming to reveal the possible path mechanisms of disturbance propagation in forest structures. Through spatial segmentation and adjacency analysis of clustered functional canopy regions, each canopy segment is treated as a node in the graph, and its adjacency relationship is determined based on its two-dimensional boundary overlap and three-dimensional elevation differences, thereby constructing a path attribute matrix containing "whether adjacent" and "how adjacent." This structure not only distinguishes vertical propagation paths (such as upper canopy collapse propagating to lower voids) and horizontal propagation paths (such as continuous canopy collapse areas along the wind axis), but also clarifies the direction and intensity of disturbance propagation along the path. This method can achieve structured identification of disturbance risk propagation channels within forest areas based solely on height and structural features without relying on depth models, thus providing a precise data foundation for local disturbance response, intervention priority ranking, and spatial connectivity analysis.

[0030] Example 5 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 The structural disturbance sensing module includes a spectral stability weight calculation unit, a structural disturbance susceptibility index calculation unit, and a canopy permeability coefficient estimation unit. The spectral stability weighting calculation unit is used for each canopy segment C j The reflectance sequences at multiple time points are extracted, and the standard deviation of reflectance is calculated to obtain the spectral stability weight of the j-th canopy segment. : ; ; in, This represents the standard deviation of the reflectance of the j-th canopy segment in the main spectral band. This represents the reflectance of the j-th canopy segment at time t. This represents the mean reflectance of the segment across all moments, where m represents the number of time phases; for stable segments, its... The closer the value is to 1, the smaller it is; The structural disturbance susceptibility index calculation unit is used to calculate the structural disturbance susceptibility index of the j-th canopy segment. : In the formula, This represents the void ratio of the j-th canopy segment. ,in, The number of viewpoints penetrating to the ground within the j-th canopy segment is represented by the number of viewpoints, obtained based on LiDAR identification. This represents the total number of machine echoes across j canopy segments, including both the canopy and the ground. represents the difference between the height of the jth canopy segment and the average height of the surrounding, represents the real-time wind speed value at the time of observation; 、 and represents the weight value, , 、 and are normalized values; 、 and ; the structural disturbance susceptibility index of the jth canopy segment reflects the susceptibility of a segment to disturbance caused by factors such as wind speed and porosity; the canopy transmittance estimation unit is used to calculate the normalized vegetation index of the jth canopy segment ; ; NIR represents the reflectivity of the near-infrared band, and RED represents the reflectivity of the red band; Then, combined with the porosity ratio of the jth canopy segment and the normalized vegetation index of the jth canopy segment , the canopy transmittance of the jth canopy segment is calculated: ; the canopy transmittance of the jth canopy segment is used to describe the ability of light to penetrate the canopy segment, which is related to its spectral reflectivity and porosity ratio.

[0031] In this embodiment, the present application realizes multi-dimensional evaluation of the disturbance risk of the canopy structure by constructing a structural disturbance perception module, which integrates indicators such as spectral stability, structural disturbance susceptibility, and canopy transmittance capability, effectively improving the precision identification capability of the disturbed area of the forest canopy. Especially by quantifying the fluctuation of the reflectivity sequence, combined with the fusion analysis of the porosity and structural difference obtained by the laser radar and the normalized vegetation index (NDVI), the system can dynamically perceive and represent the disturbance sensitive area caused by wind speed, structural disorder degree or canopy porosity change. This not only helps to improve the discrimination accuracy of remote sensing monitoring in dynamic change scenarios such as wind disturbance, disease and insect pests, but also provides basic support for subsequent path optimization, data correction and spectral enhancement processing.

[0032] Embodiment 6 This embodiment is an explanation and description in embodiment 5, please refer to Figure 1 The structural disturbance perception module further comprises an association unit, the association unit is used to combine the spectral stability weight , structural disturbance susceptibility index and canopy transmittance , the canopy disturbance index of the jth canopy segment is calculated by the following associated formula : ; wherein, , and are weight coefficients respectively. , and ; the structure disturbance susceptibility index directly reflects the disturbance sensitivity of the canopy segment caused by physical factors such as wind speed and porosity, and is the most core index for measuring canopy disturbance, so it is given the largest weight of 0.4 to ensure that it plays a leading role in the comprehensive index. The spectral stability weight reflects the stability of the reflectance of the canopy segment over time, and the higher the stability, the smaller the disturbance impact. It has important reference value for the canopy disturbance index, so it is given a weight of 0.3 to reflect its auxiliary but key role. The canopy transmittance represents the ability of light to penetrate the canopy, which is closely related to the structure and reflectance characteristics of the canopy, and also affects the disturbance transmission and reflectance change. Therefore, it is given a weight of 0.3 to ensure a balanced impact on the overall disturbance assessment.

[0033] In this embodiment, the canopy disturbance index is calculated by fusing the spectral stability weight, the structure disturbance susceptibility index and the canopy transmittance, which realizes the comprehensive characterization of the canopy disturbance state and effectively improves the precision and robustness of the disturbance identification. This method not only fully considers the time consistency of spectral information, but also introduces the sensitivity of structural changes and the illumination penetration characteristics, which can more comprehensively reflect the influence of factors such as wind speed change, structural sparseness or local degradation on the canopy state, and provides reliable support for quality control, dynamic change monitoring and path disturbance area correction of high-resolution remote sensing images and other applications.

[0034] Embodiment 7 This embodiment is an explanation and description in embodiment 1, please refer to Figure 1 , the path weight correction unit is used for the jth canopy segment, and for any two adjacent canopy segments C j and C k , a path intensity correction coefficient is introduced: wherein, , and are the spectral stability weight, the structure disturbance susceptibility index and the canopy transmittance of the kth canopy segment adjacent to the jth canopy segment respectively; based on the path intensity correction coefficient , the canopy disturbance index of the jth canopy segment , the canopy disturbance index of the jth canopy segment is modified by the joint action of nodes and paths, to obtain a second modified canopy disturbance index of the jth canopy segment , specifically: wherein, represents the canopy disturbance index of the kth canopy segment adjacent to the jth canopy segment, represents the set of all segment indices adjacent to the jth canopy segment; the meaning of the formula is, considering the disturbance transmission influence of adjacent segments, reflecting the current disturbance state of the canopy segment, combined with the disturbance effect of the surrounding segments; and according to the second modified canopy disturbance index of the jth canopy segment adjacent to the jth canopy segment, is divided into four levels when is not in level I, a modification instruction is triggered, and a canopy thermal map is constructed according to the second modified canopy disturbance index of the jth canopy segment adjacent to the jth canopy segment; and according to the range of the second modified canopy disturbance index of the jth canopy segment adjacent to the jth canopy segment, is divided into four levels, specifically, is divided into four levels to reflect the stability and disturbance degree of the canopy structure in different regions; when is in the range of 0.0000 to 0.2000, it indicates that the canopy structure in this region is extremely stable and the disturbance degree is extremely low, and it is classified as level I; such regions usually exhibit continuous and complete canopy coverage and good spectral stability; when is in the range of 0.2001 to 0.4000, it is classified as level II, representing that the canopy structure in this region has certain slight changes, but the overall structure is still relatively stable and the disturbance effect is weak.

[0035] when is in the range of 0.4001 to 0.6000, it is classified as level III, indicating that the region has experienced moderate disturbance, which may be affected by wind, pests, or human activities, resulting in local damage to the canopy structure.

[0036] if is greater than 0.6001, it is classified as level IV, indicating that the canopy is disturbed strongly and the structural continuity has decreased significantly, with phenomena such as canopy holes, tilting, or fractures.

[0037] The specific example data of embodiments 5-7 are as follows: the spectral stability weight of the jth canopy segment Data example: Suppose the reflectivity of a fragment at 3 time points is , , ; Then ; → indicates that the spectral fluctuation of the fragment is very small, and the stability is extremely high; The structural disturbance susceptibility index of the jth canopy fragment Data example: Suppose the parameters of a fragment are as follows: ; , higher than the field; , real-time wind speed; and after normalization, , , ; The weight is set to 0.4:0.3:0.3; The canopy transmittance coefficient of the jth canopy fragment Example: Normalized difference vegetation index; ; ; Then: =0.0598; The canopy disturbance index of the jth canopy fragment → indicates that the transmittance of the fragment is low, and the density is high.

[0038] The canopy disturbance index of the jth canopy fragment Calculation example: , and are 0.3, 0.4 and 0.3 respectively; Substitute the values in the previous example, ; The canopy disturbance index of the jth canopy fragment is 0.2455; 5 canopy fragments are collected, as shown in the following table 1: The following adjacency network is constructed: ; ; ; ; ; Used for the j-th canopy segment, and for any two adjacent canopy segments C j With C k Introducing a path strength correction factor :calculate , : by For example, combining the canopy disturbance index of the j-th canopy segment The canopy disturbance index of the j-th canopy segment due to the combined effects of nodes and paths. Perform corrections and obtain the second canopy perturbation index of the adjacent segments of the j-th canopy segment after correction. Specifically: ; The same method was used on other fragments as well. and Perform calculations, for example: ; ; And based on the second canopy perturbation index adjacent to the j-th canopy segment Construct a canopy heat map; as shown in Table 2 below: In this embodiment, the path weight correction unit constructs a path intensity correction coefficient by introducing the spectral stability weights of adjacent canopy segments, the structural disturbance susceptibility index, and the canopy permeability coefficient. This achieves dynamic correction of the canopy disturbance index, fully considering the transmission and mutual influence of disturbances between adjacent segments. This method can more accurately reflect the true disturbance state of the canopy structure and effectively improve the spatial continuity and accuracy of disturbance identification. Based on the corrected second canopy disturbance index... Grading allows for a fine-grained classification of canopy structure stability and disturbance levels in different areas, facilitating targeted monitoring and management. Furthermore, constructing canopy heat maps visually displays the distribution of disturbances, aiding in forest health assessments and carbon sequestration estimations, thus improving the accuracy and efficiency of forest ecological monitoring.

[0039] Example 8 This embodiment is an explanation and implementation in embodiment 1, please refer to Figure 1 The wind speed-disturbance joint modeling module includes a real-time wind speed acquisition unit, a modeling unit, and a reflectivity correction unit. The real-time wind speed acquisition unit is used to acquire the wind speed sequence V(t) above the target area in real time during the flight task by the micro meteorological sensor module carried by the UAV platform, and the wind speed-crown layer segment corresponding mapping is formed by pairing the synchronous timestamp with the image acquisition time. The modeling unit is used to construct the second crown layer disturbance index of the jth crown layer segment, and the wind speed sequence V(t) at the corresponding time point is used as an input variable, and after the disturbance-wind speed joint driving model is established and trained by using the multivariate regression model method, the reflectivity error of the jth crown layer segment is obtained : ; Wherein, represents the constant term in the multiple linear regression model, and represent the regression coefficients, and the residual term; β0(intercap): generally close to 0, but specific adjustment according to data offset; and represent the regression coefficients, and the residual term; wherein characterizes the contribution of the crown layer disturbance index to the reflectivity error; reflects the influence of wind speed change on reflectivity error, and after the measured data is verified, ; ; ; the residual term is about ±0.05; The reflectivity correction unit is used to output the corrected reflectivity output value of the jth crown layer segment based on the reflectivity error of the jth crown layer segment and the initial reflectivity of the jth crown layer segment: .

[0040] In this embodiment, the wind speed-disturbance joint modeling module establishes a multivariate regression model driven by wind speed and canopy disturbance by collecting real-time wind speed data of the target area and combining the canopy disturbance index to accurately capture the influence of wind speed on canopy reflectivity. This module can effectively correct the reflectivity fluctuations and deviations caused by wind speed, improving the stability and consistency of multi-temporal hyperspectral data. By dynamically correcting the reflectivity, the accuracy of forest canopy spectral parameters is significantly improved, thereby improving the precision and reliability of carbon sink estimation. This technology effectively solves the problem of remote sensing data quality degradation caused by wind speed disturbance.

[0041] Embodiment 9 This embodiment is an explanation and illustration in Embodiment 8, please refer to Figure 1 The carbon sink estimation module includes a parameter matching unit and a carbon storage estimation unit. The parameter matching unit is configured to receive the corrected reflectivity output value of the jth canopy segment , including: ; wherein, and represent the corrected near-infrared band reflectivity and red band reflectivity of the jth canopy segment. The carbon storage estimation unit is configured to use the NDVI inversion model to reacquire the normalized vegetation index corresponding to the jth canopy segment after correction according to the preset forest type and spectral inversion relationship and spatial coverage area to calculate the carbon storage of the jth canopy segment : ; ; ; wherein, represents the biomass value per unit area of the jth segment, and b are empirical model coefficients established in advance according to different Selina types; set between 10 and 50, b set between 1 and 3, specific reference to the relevant research reports or experimental data of the region or forest type, as shown in the following example Table 3: represents the carbon conversion coefficient, set to 0.5; spatially splicing the carbon storage of all canopy segments to output the carbon storage distribution map per unit area, and can be statistically aggregated according to administrative division, forest type or ecological function zone.

[0042] In this embodiment, the carbon sink estimation module receives the corrected reflectivity data, combines the near-infrared band and red light band reflectivity information, and accurately calculates the normalized vegetation index (NDVI) and the corresponding spatial coverage area by using the pre-established forest type spectral inversion model. According to the empirical model coefficients of different forest types, the module calculates the biomass per unit area and further converts it into carbon storage, realizing accurate carbon sink estimation for each canopy segment. Through spatial splicing of the carbon storage of all canopy segments, a high-resolution carbon storage distribution map per unit area is generated, supporting statistical summary analysis according to administrative division, forest type or ecological function zone. This technology improves the spatial fineness and accuracy of carbon sink estimation.

[0043] The size of the threshold is set for easy comparison. The size of the threshold depends on how much sample data and the number of base set by those skilled in the art for each group of sample data; as long as it does not affect the proportional relationship between the parameters and the quantized values.

[0044] The above formulas are obtained by collecting a large amount of data for software simulation and selecting a formula close to the true value. The coefficients in the formula are set by those skilled in the art according to the actual situation. The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can make equivalent substitutions or changes within the technical scope disclosed in the present application according to the technical solutions and inventive concepts of the present application, which should be covered within the protection scope of the present application.

Claims

1. A forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology, characterized in that, The application relates to a forest carbon sink estimation method based on UAV remote sensing, which comprises the following steps: The hyperspectral image acquisition module is carried on the unmanned aerial vehicle platform, flies at low altitude along the set route in the forest area, acquires a multi-time-phase forest hyperspectral image sequence in real time which passes through the space above the forest canopy, acquires DSM and DTM by using laser radar LiDAR technology, carries out pixel-level difference operation, obtains a canopy height model CHM, and extracts local average height , local standard deviation and height gradient CH(x,y) ; a recognition module for recognizing, based on local mean height , local standard deviation , and height gradient CH(x, y) as input feature vectors, clustering the feature vectors of all pixels using a K-means algorithm, for each crown layer segment C j , recognizing whether adjacent crown layer segments exist in a spatial adjacency relationship, and belonging to a vertical or horizontal path, and marking a path attribute matrix ; a structure disturbance perception module for extracting canopy structure disturbance features from the hyperspectral image sequence, identifying spectral anomaly regions caused by branch and leaf fluctuation, wind speed disturbance and canopy heterogeneity, and calculating a canopy disturbance index of the jth canopy segment , combined with the path attribute matrix , introducing a path intensity correction coefficient , and correcting the canopy disturbance index of the jth canopy segment , obtaining a second canopy disturbance index adjacent to the corrected jth canopy segment , and constructing a canopy thermal map; The wind speed-disturbance combined modeling module is configured to combine the wind speed V(t) collected by the micro-meteorological sensor carried by the unmanned aerial vehicle with the second canopy disturbance index of the jth canopy segment After the disturbance-wind speed combined driving model is established and trained by using the multivariate regression model method, the reflectivity error of the jth canopy segment is obtained After the combination, the initial reflectivity of the jth canopy segment , to output the corrected reflectivity output value of the jth canopy segment ; The carbon sink estimation module is based on the corrected reflectance output value The NDVI inversion model is used to obtain the normalized vegetation index corresponding to the jth canopy segment after correction The spatial coverage area The carbon storage of the jth canopy segment is output . 2.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 1, characterized in that, The high-spectral image acquisition module is used for deploying a UAV to carry a spectral sensor to acquire a high-spectral image sequence of multi-temporal forests penetrating through the space above a forest canopy, wherein the spectral sensor has a wave band range of 400nm to 1000nm, and contains a red edge and a near-infrared channel. 3.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 1, characterized in that, The high-spectral image acquisition module comprises an image acquisition unit, a canopy height model establishing unit and a feature extraction unit. The image acquisition unit is used for deploying a UAV to carry a spectral sensor to acquire a high-spectral image sequence of multi-temporal forests penetrating through the space above a forest canopy, and the high-spectral image sequence is preprocessed and corrected, different spectral wave bands are registered, the pixel correspondence is unified, and image acquisition time stamps, flight heights, light intensities, sensor posture information and synchronous metadata are labeled. The canopy height model establishing unit is used for deploying a UAV to carry a light-weight laser radar (LiDAR) or adopting a multi-angle image reconstruction method to generate canopy three-dimensional structure data, to obtain DSM and DTM, to perform pixel-level difference operation, and to obtain a canopy height model (CHM). computing the canopy height H(x, y), where (x, y) denotes a pixel position in the image, denotes the surface elevation of all objects in the digital surface model DSM including tree canopies, buildings, denotes the surface elevation of all objects in the digital surface model DSM including tree canopies, buildings, The feature extraction unit is used for performing fixed-size sliding window processing on the canopy height model (CHM), including a 5*5 pixel window or a 3*3 pixel window, extracting local structure features of each center pixel, including: local mean height , , which reflects the overall canopy height level of the region; local standard deviation , , which reflects the degree of fluctuation of the canopy of the region; height gradient , which describes the rate of change of the height of the pixel in the horizontal and vertical directions, is calculated using the Sobel operator: ； where H(i, j) represents the canopy height value at position (i, j) in the window, W(x, y) represents the sliding window centered at the center pixel (x, y), and N represents the number of pixels in the sliding window; is the local height standard deviation in the sliding window region centered at the center pixel (x, y); height gradient measures the rate of change of the height of the pixel position in the horizontal x and vertical y directions, and is used to identify the canopy edge or the structure mutation region of the interface, and respectively, the gradients of the canopy height model CHM in the x direction and the y direction, calculated by the Sobel operator: 。 4.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology of claim 1, wherein, The recognition module comprises a clustering unit and a canopy segment adjacency attribute recognition unit. The clustering unit is configured to use a K-means algorithm to cluster the feature vectors of all pixels based on the local average height , the local standard deviation , and the height gradient CH(x, y) as input feature vectors, and output a class label for each pixel, comprising: If the local average height is greater than 0.7 meters, the local standard deviation is less than 0.15 meters, and GH(x,y) is less than 0.15 meters / pixel, the area is determined to be a main canopy zone. If the local average height is less than 0.4 meters, and the local standard deviation is greater than 0.25 meters, and GH(x,y) is greater than 0.35 meters / pixel, it indicates that there is a significant height discontinuity in this region, which is commonly found in the transition between the canopy and the ground, and is determined as a gap boundary. If the local average height is between 0.4 and 0.7 meters, and the local standard deviation is greater than 0.2 meters, and GH(x,y) is greater than 0.3 meters / pixel, it is an indication that there is a large difference in height within the region, and the structure is chaotic, and it is determined as a high-low mixed canopy region. 5.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 4, characterized in that, The crown layer segment adjacency attribute identification unit is configured to divide all crown layer regions into a plurality of crown layer segments, and after completing the extraction of the structural features of the pixels of the entire forest region and the K-means clustering classification, the entire region is divided into a plurality of spatially continuous crown layer segments, each of which is denoted as C j a functional crown layer corresponding to each class; Canopy fragments C for each canopy category j As a node in the canopy segment adjacency graph, if two adjacent canopy segments C j With C k If there is a contact point at the boundary of two canopy segments, or if the distance between adjacent pixels is less than 3 pixels, then an edge is added to the canopy segment adjacency graph to represent the two canopy segments C. j With C k There are spatial adjacency relationships to establish an "adjacency matrix" A. jk If C j With C k Adjacent, then A jk =1, otherwise 0; Based on the structure of the "adjacency matrix" A jk , further refine the spatial structure type and interference path properties between each pair of adjacent crown segment, build a path matrix P with properties jk ; If C j is adjacent to C k , the average height of C j is greater than the average height of C k , and a number of crown layer segments are projected onto a two-dimensional plane to obtain a 2D boundary, if C j and C k overlap in the boundary area of the 2D boundary, it is determined that C j is adjacent to C k , and the label C j and C k constitute a vertical path relationship, denoted as: ; When C j With C k In the absence of overlap in the boundary regions of the 2D boundary, the horizontal execution distance of two adjacent crown segments C j With C k :​ ； And preset space adjacent threshold Td, when Less than the space adjacent threshold Td, is determined as horizontal adjacent, then constitutes horizontal path relationship, recorded as: ; Constructing a path attribute matrix : 。 6.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology of claim 1, wherein, The structure disturbance perception module comprises a spectral stability weight calculation unit, a structure disturbance susceptibility index calculation unit and a canopy transmittance coefficient estimation unit. The spectral stability weight calculation unit is configured to calculate, for each canopy segment C j , extract a reflectance sequence at a plurality of time points, and calculate a reflectance standard deviation to obtain a spectral stability weight of the jth canopy segment : ； ； wherein, Rj, t denotes the reflectance of the jth canopy segment at the main spectral band at time t, Rj, t denotes the reflectance of the jth canopy segment at time t, Rj denotes the mean of the reflectance of the segment over all times, m denotes the number of phases; The structure disturbance susceptibility index calculation unit is configured to calculate a structure disturbance susceptibility index of the jth canopy segment : In the formula, represents the gap ratio of the jth canopy segment, wherein, represents the number of ground-penetrating and viewpoint numbers in the jth canopy segment, obtained based on laser radar (LiDAR) recognition; represents the total number of organ echoes of the jth canopy segment, including the canopy and the ground, represents the difference between the jth canopy segment and the average height around it, represents the real-time wind speed value at the time of observation; , and represents a weight value, , , and are all normalized values; The canopy transpiration coefficient estimation unit is configured to calculate a normalized difference vegetation index of the jth canopy segment ; ； Wherein, NIR represents the reflectivity of the near-infrared wave band, and RED represents the reflectivity of the red light wave band. Then, the gap ratio of the jth canopy segment is combined with the normalized difference vegetation index of the jth canopy segment the normalized difference vegetation index of the jth canopy segment the canopy transmittance of the jth canopy segment is calculated : 。 7.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 6, characterized in that, The structure disturbance perception module further comprises an association unit and a path weight correction unit, the association unit is configured to combine the spectral stability weight of the jth canopy segment , the structure disturbance susceptibility index and the canopy transmittance coefficient , and the canopy disturbance index of the jth canopy segment is calculated by the following associated formula : ； wherein, , and are weight coefficients, respectively. 8.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 7, characterized in that, The path weight correction unit is configured to, for the jth canopy segment, and for any two adjacent canopy segments C j With C k , a path intensity correction coefficient : wherein, , and are the spectral stability weight, the structural perturbation susceptibility index and the canopy transmissivity coefficient of the kth canopy segment adjoining the jth canopy segment, respectively. Based on path intensity correction coefficient The canopy disturbance index of the jth canopy segment in combination with the jth canopy segment The canopy disturbance index of the jth canopy segment in combination with the jth canopy segment The second canopy disturbance index adjacent to the jth canopy segment after correction Specifically: wherein, represents the canopy disturbance index of the kth canopy segment adjacent to the jth canopy segment, represents the set of all segment indices adjacent to the jth canopy segment; the meaning of the formula is, Considering the influence of adjacent segment disturbance transmission, reflecting the current disturbance state of the canopy segment, combining the disturbance effect of the surrounding segments; and according to the second crown layer disturbance index of the jth modified crown layer segment adjacent to the first crown layer segment is divided into four levels, when if not in level I, a modification instruction is triggered, and according to the second crown layer disturbance index of the jth modified crown layer segment adjacent to the first crown layer segment a crown layer thermal map is constructed. 9.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 8, characterized in that, The wind speed-disturbance joint modeling module comprises a real-time wind speed acquisition unit, a modeling unit and a reflectivity correction unit. The real-time wind speed acquisition unit is used for deploying a micro meteorological sensor module on a UAV platform to acquire a wind speed sequence V(t) above a target region in a flight task process, to pair the wind speed sequence V(t) with image acquisition time through synchronous timestamps, and to form a wind speed-canopy segment corresponding mapping. The modeling unit is configured to obtain a second canopy layer disturbance index of a jth constructed canopy layer segment adjacent to the first canopy layer segment , and the wind speed sequence V(t) at the corresponding time point as an input variable, a disturbance-wind speed combined driving model is established and trained by using a multivariate regression model method, and the reflectivity error of the jth canopy layer segment is obtained : ； wherein, represents a constant term in the multiple linear regression model, and represents a regression coefficient, represents a residual term; The reflectivity correction unit is configured to correct the reflectivity error of the jth crown layer segment based on the initial reflectivity of the jth crown layer segment and the initial reflectivity of the jth crown layer segment to output a corrected reflectivity output value of the jth crown layer segment : 。 10.The forest carbon sink estimation system based on unmanned aerial hyperspectral remote sensing technology according to claim 9, characterized in that, The carbon sink estimation module comprises a parameter matching unit and a carbon storage estimation unit. The parameter matching unit is configured to receive the corrected reflectivity output value of the jth canopy segment , comprises: ; wherein, and represent the corrected near-infrared band reflectivity and red light band reflectivity of the jth canopy segment The carbon storage estimation unit is configured to reacquire the normalized vegetation index corresponding to the jth canopy segment after correction according to a preset forest type and spectral inversion relationship by using an NDVI inversion model and a spatial coverage area , and calculate the carbon storage of the jth canopy segment : ； ； ； wherein, Bj represents the biomass value per unit area of the jth segment, and b are empirical model coefficients pre-established according to different Selina types; C represents the carbon conversion coefficient, set to 0.5; Carbon storage for all canopy segments Spatial mosaic is performed, and the carbon storage distribution map per unit area is output, and the carbon storage can be statistically aggregated according to administrative division, forest type or ecological function zone.

Images