Small-region carbon sink calculation method and system based on multiple constraints

Abstract

The invention provides a small-region forest carbon sink accurate estimation method and system based on multiple constraints, and the method comprises the steps: respectively generating a canopy height model, a leaf area index and a chlorophyll content through obtaining unmanned plane laser radar data and multispectral data; generating a digital elevation model by adopting a lowest point filtering technology, realizing individual tree segmentation in combination with a gradient map and a watershed algorithm, and extracting individual tree geometric parameters; and constructing a carbon sink estimation model by using the geometric parameters and the physical and chemical parameters, and comprehensively calculating the carbon sink amount of the small region. The problems that a traditional carbon sink estimation method is insufficient in precision and low in automation level are solved, the single-tree segmentation precision and the carbon sink amount calculation precision are improved through deep fusion of geometric information and spectral information, full-process automation of data acquisition, data processing and carbon sink estimation is achieved, and the method is suitable for large-scale popularization and application. The method is suitable for carbon sink monitoring requirements of various forest types and complex terrain scenes.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and particularly relates to a method and system for calculating small-area carbon sinks based on multiple constraints. Background Art

[0002] Small areas (from a few mu to several square kilometers) are the main bodies of carbon sink trading. However, traditional carbon sink monitoring mainly relies on ground plot surveys, which are not only time-consuming and laborious, but also difficult to be widely used on a large scale. In addition, due to the complexity of forest structure and the accuracy limitation of data sources, the carbon sink estimation results at the small-area scale are inaccurate, have poor stability, and may even show a negative carbon sink accumulation after a certain growth period of the forest. Quickly, accurately, and efficiently obtaining the small-area carbon sink amount is of great significance for understanding the carbon storage dynamics of forest ecosystems, optimizing forest management strategies, and serving carbon sink trading.

[0003] Light Detection and Ranging (LiDAR) and oblique photography technology have significant advantages in small-area carbon sink estimation, especially in terms of high precision, efficiency, and flexibility, providing strong data support for the monitoring and assessment of forest carbon stocks. LiDAR technology can accurately obtain the three-dimensional structure information of forests by emitting laser pulses and measuring the time of the returned signals, which is crucial for individual tree segmentation, crown area calculation, and carbon stock estimation. Specifically, LiDAR data can generate a Canopy Height Model (CHM), and by calculating the difference between the ground elevation and the forest point cloud, the height information of each tree can be accurately obtained. Therefore, individual trees can be accurately identified and segmented, and their growth status and carbon stocks can be analyzed. In addition, LiDAR can accurately measure the crown area of trees through high-density point cloud data, clearly depicting the external contour of the trees, providing key spatial data support for carbon sink estimation. Compared with traditional ground surveys, LiDAR not only has high efficiency but also can quickly cover large areas of forests, especially suitable for complex terrain and dense forest environments. LiDAR is not restricted by crown density and terrain undulation and can penetrate the crown to obtain the spatial information of the ground and trees, significantly improving the monitoring efficiency and accuracy, especially performing well in long-term dynamic monitoring and large-scale forest carbon stock assessment. Popescu (2007) developed a method for estimating the diameter at breast height (DBH) and aboveground biomass of individual trees based on LiDAR measurement parameters such as tree height and crown width by analyzing LiDAR data, and further established linear and nonlinear regression models to estimate the biomass of each component of the tree (such as leaves, coarse roots, trunks, and bark). Finally, LiDAR data can explain more than 90% of the variability in individual tree aboveground biomass (AGB). AGB represents the total amount of organic carbon stored in trees and can directly reflect the forest carbon sink capacity. Estimating carbon stocks through AGB is the basis for measuring the carbon sequestration capacity of forest ecosystems. Oblique photography technology uses unmanned aerial vehicles to collect images from multiple angles, which can provide more abundant tree canopy data. The advantage of oblique photography is that it can capture the three-dimensional morphology of trees from different directions and angles. Especially in complex environments and dense forests, it can more accurately extract the canopy boundary information of trees. This plays an important role in calculating the crown area of trees and conducting individual tree segmentation analysis. Basyuni (2023) studied the use of photogrammetry technology to estimate the aboveground biomass (AGB) of mangroves and their alternative commercial lands (oil palm and coconut plantations) in Indonesia. Through cross-validation, the results showed that photogrammetry technology had high accuracy in estimating AGB, and the determination coefficients (R²) of AGB for mangroves, oil palm, and coconut plantations were all between 0.87 and 0.97, showing the reliability and precision of this technology in different land use types.

[0004] In the estimation of biophysical parameters, multi-spectral or hyperspectral remote sensing technology can provide rich spectral information, which can accurately reflect the growth status, ecological health and environmental adaptability of plants. By analyzing the reflection spectral characteristics of plants, remote sensing technology can effectively invert key biophysical and chemical parameters, such as leaf area index (LAI) and chlorophyll content (P). LAI is an important indicator to measure the growth density of forests, the photosynthetic capacity of plants and the carbon sequestration capacity, and is usually used to describe the total area of leaves per unit land area. LAI directly affects the water, carbon and nutrient cycles of forests and is a key parameter for evaluating the potential of carbon absorption and storage. P represents the photosynthetic efficiency of plants, which is closely related to the growth status, health of plants and their ability to cope with environmental stress. An increase in P usually means that plants have higher photosynthetic capacity and better growth conditions.

[0005] In existing research, individual tree segmentation technology has been widely used in forest resource surveys. Through individual tree segmentation, geometric parameters of individual trees can be extracted from point cloud data, providing input for carbon sequestration models. However, most individual tree segmentation algorithms rely on fixed thresholds or experience-based region segmentation methods, which have low segmentation accuracy when dealing with dense forests or scenes with blurred canopy boundaries. As a segmentation method based on gradient maps, the watershed algorithm has high boundary detection ability and is suitable for the accurate segmentation of individual trees in forests. By combining the canopy height model (CHM) to generate a gradient map and using the watershed algorithm for individual tree segmentation, the deficiencies of traditional segmentation algorithms can be effectively solved and the segmentation accuracy can be improved. Moreover, in the existing technology, parameters such as chlorophyll cannot be calculated based on gray values, etc. The inversion processes of leaf area index LAI and chlorophyll content P do not consider canopy height and canopy projection area, resulting in large measurement errors and unable to accurately measure the carbon sequestration reserve. Summary of the Invention

[0006] The object of the present invention is to solve the problems such as poor inter-annual accuracy, large volatility and easy occurrence of negative values in the estimation results of small-area forest carbon sequestration. The present invention provides a method and system for accurately estimating small-area forest carbon sequestration based on multiple constraints. By obtaining unmanned aerial vehicle lidar data and multi-spectral data, a canopy height model and leaf area index and chlorophyll content are generated respectively; the lowest point filtering technology is used to generate a digital elevation model, and individual tree segmentation is realized by combining the gradient map with the watershed algorithm to extract individual tree geometric parameters; a carbon sequestration estimation model is constructed using geometric parameters and physical and chemical parameters, and the small-area carbon sequestration amount is comprehensively calculated. The present invention solves the problems of insufficient accuracy and low automation level of traditional carbon sequestration estimation methods. By deeply integrating geometric and spectral information, it improves the individual tree segmentation accuracy and carbon sequestration amount calculation accuracy, realizes the full-process automation of data collection, processing and carbon sequestration estimation, and is applicable to the carbon sequestration monitoring needs of various forest types and complex terrain scenarios.

[0007] To achieve the above invention objectives, the present invention provides a small-area carbon sink calculation method based on multiple constraints, including the steps:

[0008] S1: Obtain UAV-Lidar (UAV-LS) data and multispectral data; preprocess the multispectral data;

[0009] S2: Single-tree segmentation. Filter the noise of the UAV-Lidar (UAV-LS) data, remove the ground points, generate a canopy height model, generate a gradient map using the canopy height model, and perform single-tree segmentation on the gradient map using the watershed algorithm to extract the geometric parameters of each tree. The geometric parameters include canopy height and canopy projection area , providing input for subsequent parameter inversion and carbon sink model construction;

[0010] S3: Inversion of leaf area index (LAI) and chlorophyll content (P). Based on the preprocessed spectral data, combined with the single-tree segmentation results, analyze the spectral characteristics within the single-tree segmentation area. According to the Beer-Lambert law, the leaf area index (LAI) is calculated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. The leaf area index (LAI) has a logarithmic relationship with the canopy porosity:

[0011]

[0012] where, is the zenith angle, representing the incident angle of light; is the canopy gap fraction, referring to the proportion of voids in the canopy through which light can pass when observed from the ground upwards; is the canopy light transmittance coefficient, describing the ability of the vegetation canopy to absorb and scatter light; is the canopy height coefficient, is the canopy projection area

[0013]

[0014] is the infrared band of the infrared spectrum, is the near-infrared band of the infrared spectrum, is the first regression coefficient, is the second regression coefficient; is the gray coefficient value, is the average gray value of the canopy projection image;

[0015] S4: Calculate the total carbon sink of the small area as follows:

[0016]

[0017] where, 、 are respectively the canopy projection area and canopy height of the th tree; , are respectively the leaf area index and chlorophyll content of the th tree, is the first carbon storage coefficient, is the second carbon storage coefficient, is the third carbon storage coefficient.

[0018] Preferably, preprocessing the multispectral data includes: removing random noise in the multispectral data through median filtering.

[0019] Preferably, noise filtering is performed on the Unmanned Aerial Vehicle LiDAR (UAV-LS) data to remove ground points, including: generating a digital elevation model through the lowest point filtering technique, calculating the height difference between the point cloud and the DEM, removing the ground points, and only retaining the vegetation points.

[0020] Preferably, generating the canopy height model includes: separating the ground points and non-ground points from the UAV-LS data point cloud, respectively generating a digital elevation model (DEM) and a digital surface model (DSM), and calculating the canopy height model through pixel subtraction;

[0021] Generating a gradient map using the canopy height model includes: calculating the gradients in the horizontal and vertical directions using the Sobel operator, and calculating the square root of the sum of the squares of the horizontal and vertical gradients to generate a canopy height gradient map;

[0022] Using the watershed algorithm to perform single-tree segmentation on the gradient map and extract the geometric parameters of each tree: including extracting the local extreme points as the initial centers of the single-tree segmentation; performing segmentation on the gradient map based on the watershed algorithm to divide the canopy area of each tree.

[0023] Preferably, the steps of generating a digital elevation model through the lowest point filtering technique include:

[0024] Firstly, grid the point cloud, divide the LiDAR point cloud data into regular grids according to a fixed size, i.e., 1 m × 1 m;

[0025] Secondly, screen the lowest points. In each grid, screen out the point with the lowest height value as the ground point, and perform interpolation to complete the grids without points;

[0026] Thirdly, smooth the terrain. Perform Gaussian filtering on the height values of the screened lowest points to remove discrete noise;

[0027] Finally, generate the DEM. Generate a continuous digital elevation model based on the height values of the lowest points within the grids to reflect the surface height of the terrain.

[0028] This application also provides a small-area carbon sink calculation system based on multiple constraints, including:

[0029] An acquisition module that acquires UAV-LiDAR (UAV-LS) data and multispectral data; and preprocesses the multispectral data;

[0030] A single-tree segmentation module that filters noise from the UAV-LiDAR (UAV-LS) data, removes ground points, generates a canopy height model, generates a gradient map using the canopy height model, and performs single-tree segmentation on the gradient map using the watershed algorithm to extract the geometric parameters of each tree. The geometric parameters include canopy height and canopy projected area , providing input for subsequent parameter inversion and carbon sink model construction;

[0031] A leaf area index (LAI) and chlorophyll content (P) inversion module that analyzes the spectral characteristics within the single-tree segmentation area based on the preprocessed spectral data in combination with the single-tree segmentation results. According to the Beer-Lambert law, the leaf area index (LAI) is calculated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. The leaf area index (LAI) has a logarithmic relationship with the canopy porosity:

[0032]

[0033] Wherein, is the zenith angle, representing the incident angle of light; is the canopy gap fraction, which refers to the proportion of gaps in the canopy through which light can pass when observed from the ground upwards; The canopy light transmittance coefficient, which describes the ability of the vegetation canopy to absorb and scatter light; is the canopy height coefficient, is the canopy projected area

[0034]

[0035] is the infrared band of the infrared spectrum, is the near-infrared band of the infrared spectrum, is the first regression coefficient, is the second regression coefficient; is the grayscale coefficient value, is the average grayscale value of the canopy projected image;

[0036] S4: Total carbon sink amount of the small area is calculated as follows:

[0037]

[0038] Wherein, 、 are respectively the The canopy projection area and canopy height of a tree; , respectively the leaf area index and chlorophyll content of the th tree, is the first carbon storage coefficient, is the second carbon storage coefficient, is the third carbon storage coefficient.

[0039] Preferably, the preprocessing of the multispectral data includes: removing random noise in the multispectral data through median filtering.

[0040] Preferably, noise filtering is performed on the Unmanned Aerial Vehicle - LiDAR (UAV - LS) data, and ground points are removed, including: generating a digital elevation model through the lowest point filtering technique, calculating the height difference between the point cloud and the DEM, removing ground points, and only retaining vegetation points.

[0041] Preferably, generating the canopy height model includes: separating ground points and non - ground points from the UAV - LS data point cloud, respectively generating a digital elevation model (DEM) and a digital surface model (DSM), and calculating the canopy height model through pixel subtraction;

[0042] Generating a gradient map using the canopy height model includes: calculating gradients in the horizontal and vertical directions using the Sobel operator, and calculating the square root of the sum of the squares of the horizontal and vertical gradients to generate a canopy height gradient map;

[0043] Using the watershed algorithm to perform single - tree segmentation on the gradient map and extract the geometric parameters of each tree: including extracting local extreme points as the initial centers of single - tree segmentation; performing segmentation on the gradient map based on the watershed algorithm to divide the canopy area of each tree.

[0044] Preferably, the steps of generating a digital elevation model through the lowest point filtering technique include:

[0045] Firstly, point cloud gridding, dividing the LiDAR point cloud data into regular grids according to a fixed size, i.e., 1 m × 1 m;

[0046] Secondly, lowest point screening, in each grid, screening out the point with the lowest height value as the ground point, and interpolating and complementing the grids without points;

[0047] Thirdly, terrain smoothing, performing Gaussian filtering on the height values of the screened lowest points to remove discrete noise;

[0048] Finally, generating the DEM, generating a continuous digital elevation model based on the height values of the lowest points within the grids to reflect the surface height of the terrain.

[0049] The present invention provides a precise estimation method and system for forest carbon sinks in small areas based on multiple constraints. By comprehensively utilizing Unmanned Aerial Vehicle - Light Detection and Ranging (UAV - LS) data and multi - spectral / hyperspectral data, it solves many problems existing in traditional carbon sink estimation methods and has the following remarkable beneficial effects:

[0050] 1. High - precision carbon sink quantity estimation. The present invention uses UAV - LS data to generate a Digital Elevation Model (DEM) and a Canopy Height Model (CHM). Through the extraction of high - precision geometric parameters (such as canopy height and canopy projected area), it provides data support for the precise description of forest structure. Combining the physical and chemical parameters (such as Leaf Area Index LAI and Chlorophyll Content P) inverted from multi - spectral data, it realizes the deep fusion of geometric information and spectral information, effectively improving the accuracy of carbon sink quantity estimation. The accuracy of the inversion of physical and chemical parameters and the applicability of the model are improved. As a characteristic representation of spectral data, the gray value reflects the local optical characteristics of pixels and is closely related to the physiological state and health status of vegetation. Based on the traditional inversion method, the present invention adds the gray value as a correction factor to the parameter calculation formula, realizing the deep fusion of spectral characteristics and spatial distribution characteristics. Specifically, the introduction of the gray value effectively makes up for the limitation of relying solely on the difference or ratio of spectral bands in the traditional method, enabling the inversion model to more sensitively capture the detailed changes in spectra, especially performing more excellently under complex lighting conditions or non - homogeneous canopy structures. At the same time, the inversion method combined with the gray value can enhance the characterization ability of single - tree feature regions, reduce the calculation deviation caused by local anomalies in spectral data, thus significantly improving the estimation accuracy of LAI and P parameters and providing more reliable data support for the construction of carbon sink models. This improvement not only increases the innovation of the technical solution but also expands the diversity of its applicable scenarios.

[0051] 2. Automated processing and high efficiency. The present invention proposes a DEM generation method based on the lowest - point filtering technology, which can quickly remove noise points and non - ground points in the point cloud and realize automatic terrain modeling. At the same time, through the watershed algorithm, the gradient map generated by the canopy height model is used for single - tree segmentation. Without relying on manual operations or empirical parameters, it can efficiently extract the geometric parameters of each tree, greatly reducing the labor cost and time consumption.

[0052] 3. Robustness for adapting to complex scenarios. The single - tree segmentation algorithm of the present invention is based on the canopy height model and the gradient map, combined with the boundary detection ability of the watershed algorithm. When dealing with scenes of dense forests or fuzzy canopy boundaries, it can significantly improve the segmentation accuracy. Even under complex terrains and various forest types, this method can stably extract the geometric features of single trees.

[0053] 4. Innovation in multi-dimensional information fusion. The present invention organically combines the geometric parameters of forests (such as canopy height and canopy projection area) and spectral information (such as leaf area index and chlorophyll content), and establishes a carbon sink estimation model based on multiple constraints. By introducing geometric features and pixel gray values in the calculation process of leaf area index and chlorophyll content, the comprehensive characterization ability of the model for forest ecological characteristics is enhanced, and the applicable range of parameter inversion is expanded. The technical solution proposed by the present invention is applicable to the carbon sink monitoring requirements of various forest types (such as coniferous forests, broad-leaved forests, etc.) and different regions. By adjusting the weight coefficients (such as carbon storage coefficients), it can be flexibly applied to different climate conditions, vegetation types, and management objectives. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flowchart of a method for accurately estimating forest carbon sink in a small area based on multiple constraints according to the present invention;

[0055] Figure 2 is a result diagram of accurately estimating forest carbon sink in a small area based on multiple constraints according to the present invention;

[0056] Figure 3 is an accuracy verification diagram of forest carbon sink in a small area based on multiple constraints according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0057] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0058] In one embodiment, the present invention provides a method for calculating carbon sink in a small area based on multiple constraints, as Figure 1 shown, including the steps:

[0059] S1: Obtain UAV lidar (UAV-LS) data and multispectral data; preprocess the multispectral data. In one embodiment, in a mountain forest area (with an area of 10 hectares and the main tree species being a mixed forest of coniferous and broad-leaved trees), a UAV equipped with a lidar and a multispectral camera is used to collect data in this area to construct a forest carbon sink estimation model. Data acquisition for UAV lidar (UAV-LS): Equipment used: A UAV equipped with a lidar sensor (such as Velodyne VLP-16). Flight parameters: Flight altitude: 120 meters. Flight speed: 5 m / s. Overlap rate: 80% horizontally and 70% vertically for the flight path. Data type: Point cloud data: including three-dimensional coordinates (x, y, z) and reflection intensity. Output file format: LAS format.

[0060] Equipment for multi-spectral data acquisition: An unmanned aerial vehicle (UAV) equipped with a multi-spectral camera (such as MicaSense RedEdge-MX). Band information: Blue, green, red, red edge, near-infrared (NIR). Flight parameters: Flight altitude: 120 meters. Resolution: 5 cm / pixel. Output file format: GeoTIFF format. Preprocessing of multi-spectral data: For the acquired multi-spectral data, perform the following preprocessing steps: Denoising Use median filtering to remove sensor noise. In the red-band image, a 3×3 median filter is used to smooth the noise points. Atmospheric correction method: Use a radiative transfer model (such as the 6S atmospheric correction model). Input parameters: Solar altitude angle (45°), atmospheric model (mid-latitude summer), aerosol type (continental). Output result: Convert the DN value (digital number) of the multi-spectral image to surface reflectance. Geometric correction method: Based on UAV GPS and IMU data, as well as ground control points (GCPs), orthorectify the multi-spectral image. Tool: Use Pix4D Mapper software to complete geometric registration to ensure that the image is aligned with geographical coordinates. Radiometric correction Correct the sensor gain and offset values to standardize the radiometric values of each band's image. Example: Adjust the spectral curve of the multi-spectral image according to the measured reflectance of the radiometric calibration panel. Spectral feature extraction Calculate the Normalized Difference Vegetation Index (NDVI): Generate an NDVI image for subsequent inversion of leaf area index (LAI) and chlorophyll content (P). Output results LiDAR point cloud data: Format: LAS. Content: Three-dimensional point cloud data for generating a digital elevation model (DEM) and a canopy height model (CHM). Multi-spectral data: Format: GeoTIFF. Content: Images of the blue, green, red, red edge, and near-infrared bands after atmospheric correction, as well as the NDVI feature image. Preprocessing quality evaluation: Noise point removal rate: Above 98%. Image geometric error: Less than 5 cm.

[0061] S2: Individual tree segmentation. Filter the noise of the UAV LiDAR (UAV-LS) data, remove the ground points, generate a canopy height model, generate a gradient map using the canopy height model, and perform individual tree segmentation on the gradient map using the watershed algorithm to extract the geometric parameters of each tree. The geometric parameters include canopy height and canopy projected area , providing input for subsequent parameter inversion and carbon sink model construction;

[0062] S3: Inversion of leaf area index (LAI) and chlorophyll content (P). Based on the preprocessed spectral data and combined with the individual tree segmentation results, analyze the spectral characteristics within the individual tree segmentation area. According to the Beer-Lambert law, the leaf area index (LAI) is calculated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. The leaf area index (LAI) has a logarithmic relationship with the canopy porosity:

[0063]

[0064] Among them, is the zenith angle, representing the incident angle of light; is the canopy gap fraction, referring to the proportion of gaps through which light can pass in the canopy when observed from the ground upwards; The canopy light transmittance coefficient describes the ability of the vegetation canopy to absorb and scatter light; is the canopy height coefficient, is the canopy projected area

[0065]

[0066] is the infrared band of the infrared spectrum, is the near-infrared band of the infrared spectrum, is the first regression coefficient, is the second regression coefficient; is the grayscale coefficient value, is the average grayscale of the canopy projected image;

[0067] S4: Total carbon sink in a small area is calculated as follows:

[0068]

[0069] Among them, and are respectively the th tree's canopy projected area and canopy height; and are respectively the th tree's leaf area index and chlorophyll content, is the first carbon storage coefficient, is the second carbon storage coefficient, is the third carbon storage coefficient.

[0070] In one embodiment, a method for accurately estimating the forest carbon sink in a small area based on multiple constraints uses two geometric parameters, namely the canopy height and the canopy projected area, as hard constraint conditions, and two biophysical parameters and physicochemical parameters, namely the leaf area index and the chlorophyll content, as soft constraint conditions to solve problems such as poor annual accuracy, large volatility, and easy occurrence of negative values in the estimation results of the forest carbon sink in a small area. By extracting two geometric parameters, the single-tree canopy height and the canopy projected area, from the high-resolution remote sensing data of drones, and inverting two physicochemical parameters, the leaf area index and the chlorophyll content, a precise estimation model of the small-area carbon sink is constructed, thus meeting the needs of forest carbon sink trading.

[0071] In some embodiments, preprocessing the multispectral data includes: removing random noise in the multispectral data through median filtering.

[0072] In some embodiments, for noise filtering of Unmanned Aerial Vehicle - Light Detection and Ranging (UAV - LS) data and removing ground points, it includes: generating a Digital Elevation Model (DEM) through the lowest - point filtering technique, calculating the height difference between the point cloud and the DEM, removing ground points, and only retaining vegetation points.

[0073] In some embodiments, generating a Canopy Height Model (CHM) includes: separating ground points and non - ground points from the UAV - LS data point cloud, respectively generating a Digital Elevation Model (DEM) and a Digital Surface Model (DSM), and calculating the CHM through pixel subtraction.

[0074] Generating a gradient map using the CHM includes: calculating gradients in the horizontal and vertical directions using the Sobel operator, and calculating the square root of the sum of the squares of the horizontal and vertical gradients to generate a canopy height gradient map.

[0075] Using the watershed algorithm to perform individual tree segmentation on the gradient map and extract geometric parameters of each tree: including extracting local extreme points as the initial centers for individual tree segmentation; performing segmentation on the gradient map based on the watershed algorithm to divide the canopy area of each tree.

[0076] In some embodiments The steps of generating a Digital Elevation Model (DEM) through the lowest - point filtering technique include:

[0077] Firstly, grid the point cloud, divide the LiDAR point cloud data into regular grids according to a fixed size, i.e., 1 meter × 1 meter.

[0078] Secondly, screen the lowest points. In each grid, screen out the point with the lowest height value as the ground point, and perform interpolation to complete the grids without points.

[0079] Thirdly, smooth the terrain. Perform Gaussian filtering on the height values of the screened lowest points to remove discrete noise.

[0080] Finally, generate the DEM. Generate a continuous Digital Elevation Model based on the height values of the lowest points within the grids to reflect the surface height of the terrain.

[0081] In one embodiment, a method for precise estimation of small - area carbon sink based on multiple constraints is provided, including the following steps: (1) Data acquisition: a. Obtain Unmanned Aerial Vehicle - Light Detection and Ranging (UAV - LS) data, multispectral data, hyperspectral data, and measured data; b. According to the dominant tree species in the study area, collect allometric equations in the "Guidelines for Carbon Sink Measurement and Monitoring of Afforestation Projects" of the State Forestry Administration and "Assessment of Biomass and Carbon Storage of Chinese Forest Vegetation".

[0082] (2)Single-tree segmentation, a. Point cloud data preprocessing: Filter noise and remove ground points from UAV data to generate an accurate canopy height model (CHM) for subsequent single-tree segmentation; b. Single-tree segmentation: Perform single-tree segmentation based on CHM and UAV-LS data, and extract geometric parameters related to single trees;

[0083] (3)Inversion of physical and chemical parameters, a. Spectral data preprocessing and correction: Since the spectral reflectance of multi- / hyperspectral image data obtained at different times is different due to different solar elevation angles, correction processing is required. b. Inversion of leaf area index (LAI): LAI is estimated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. According to the Beer-Lambert law, LAI has a logarithmic relationship with canopy gap fraction:

[0084]

[0085] c. Inversion of chlorophyll content (P): Invert P through a semi-empirical regression model;

[0086] (4)Setting of constraint conditions: As long as the forest grows normally, the carbon sink amount between years should be 0. To ensure that the estimated result of the carbon sink of a normal-growing forest is non-negative, it is necessary to set constraint conditions for its evaluation model.

[0087] (5)Construction of carbon sink model for small areas: The single-tree carbon storage models for two years are constructed as follows:

[0088]

[0089]

[0090] The carbon sink amount estimation formula is:

[0091]

[0092] where A is the canopy projection area, H is the canopy height, LAI is the leaf area index, and P is the chlorophyll content.

[0093] The present application also provides a small-area carbon sink calculation system based on multiple constraints. The hardware components of the system include: a drone platform, a lidar sensor with the model of Velodyne VLP-16 or an equivalent model, which is used to collect point cloud data and generate the three-dimensional structure of the forest canopy. A multispectral camera with the model of MicaSense RedEdge-MX or an equivalent model, which is used to collect multispectral images for retrieving the leaf area index (LAI) and chlorophyll content (P). The flight control module includes a GPS module and an IMU module, which record the flight trajectory and attitude information of the drone for subsequent image geometric correction. The ground control station is a high-performance laptop computer configured with an Intel i7 processor or higher, 32GB of memory, and an independent graphics card (NVIDIA GTX 2060 or higher), which is used to receive the data collected by the drone in real time and perform preliminary processing such as point cloud denoising and image stitching. The remote control device is used to control the flight mission of the drone to ensure that the flight path covers the target area. The data processing server is a GPU server configured with dual Intel Xeon processors, 128GB of memory, and an NVIDIA Tesla V100 GPU, which is used to process large-scale point cloud data and multispectral data, including point cloud segmentation and DEM generation, calculation of the canopy height model (CHM), generation of gradient maps and watershed algorithm segmentation. The storage device is configured with a 4TB hard disk array supporting RAID 5, which is used to store point cloud data, multispectral images, CHM, and segmentation results. The field measurement device includes a high-precision GPS surveying instrument with the model of Trimble R10 or an equivalent model, which is used to obtain the accurate coordinates of ground control points (GCPs) for image geometric correction, and a canopy analyzer with the model of LAI-2200 or an equivalent model, which is used to measure the leaf area index (LAI) and chlorophyll content (P) in the field for model calibration. The terminal display device is a desktop workstation configured with 32GB of memory, an independent graphics card, and dual monitors, which is used to visualize the processing results and generate a small-area carbon sink estimation report.

[0094] Hardware connection method: The drone is connected to the ground control station. Communication method: Wireless link (Wi-Fi or LoRa communication module). Transmission content: Point cloud data, multispectral images, and flight trajectory information collected by the drone. The ground control station is connected to the data processing server. Communication method: Ethernet or high-speed wireless network. Transmission content: The original data files (point cloud files and image files) collected by the drone are transferred to the data processing server for subsequent processing. The data processing server is connected to the storage device. Communication method: SAS or NAS network storage protocol. Transmission content: Store the processed point cloud data, CHM, multispectral data, gradient map, and segmentation results. The data processing server is connected to the terminal display device. Communication method: Ethernet connection. Transmission content: The processed carbon sink estimation results, including individual tree segmentation results, physical and chemical parameter inversion results, and total carbon sink calculation results. The field measurement device is connected to the data processing server. Communication method: USB data cable or Bluetooth. Transmission content: GCP data and on-site verification data measured in the field are used for model calibration.

[0095] System working process: The drone flies to collect data. The drone executes a preset flight mission in the target area to collect point cloud data and multispectral images. The data is transmitted to the ground control station through wireless communication. The ground control station preliminarily processes the data: Integrate the original data collected by the drone and transmit it to the data processing server. The data processing server performs in-depth processing: Point cloud data processing: Use the lowest point filtering technology to generate a digital elevation model (DEM). Calculate the canopy height model (CHM). Multispectral data processing: Perform denoising, atmospheric correction, geometric correction, and spectral feature extraction. Individual tree segmentation: Generate a gradient map based on the CHM, and use the watershed algorithm to extract the geometric parameters of individual trees.

[0096] Parameter inversion and carbon sink estimation: Invert the leaf area index (LAI) and chlorophyll content (P), and calculate the carbon sink amount in combination with geometric parameters. Result display and report generation: The processing results are visualized through the terminal display device, and a small-area carbon sink estimation report is generated.

[0097] Acquisition module: Acquire UAV-LS data and multispectral data of the drone; preprocess the multispectral data;

[0098] Individual tree segmentation module: Filter the noise of the UAV-LS data of the drone, remove the ground points, generate a canopy height model, generate a gradient map using the canopy height model, and use the watershed algorithm to perform individual tree segmentation on the gradient map to extract the geometric parameters of each tree. The geometric parameters include canopy height and canopy projected area and provide input for subsequent parameter inversion and carbon sink model construction;

[0099] Leaf Area Index LAI and Chlorophyll Content P Inversion Module. Based on the preprocessed spectral data and combined with the individual tree segmentation results, it analyzes the spectral characteristics within the individual tree segmentation area. According to the Beer-Lambert law, the Leaf Area Index LAI is calculated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. There is a logarithmic relationship between the Leaf Area Index LAI and the canopy porosity:

[0100]

[0101] Among them, is the zenith angle, representing the incident angle of light; is the canopy gap fraction, referring to the proportion of gaps in the canopy through which light can pass when observed from the ground upwards; Canopy light transmittance coefficient, which describes the ability of the vegetation canopy to absorb and scatter light; is the canopy height coefficient, is the canopy projected area

[0102]

[0103] is the infrared band of the infrared spectrum, is the near-infrared band of the infrared spectrum, is the first regression coefficient, is the second regression coefficient; is the gray coefficient value, is the average gray value of the canopy projected image;

[0104] S4: Total Carbon Sink in Small Area It is calculated as follows:

[0105]

[0106] Among them, 、 are respectively the canopy projected area and canopy height of the th tree; 、 are respectively the Leaf Area Index and chlorophyll content of the th tree, is the first carbon storage coefficient, is the second carbon storage coefficient, is the third carbon storage coefficient.

[0107] In some embodiments, preprocessing the multispectral data includes: removing random noise in the multispectral data through median filtering.

[0108] In some embodiments, noise filtering is performed on the UAV-Lidar (UAV-LS) data of the unmanned aerial vehicle, and ground points are removed, including: generating a digital elevation model through the lowest point filtering technique, calculating the height difference between the point cloud and the DEM, removing the ground points, and only retaining the vegetation points.

[0109] In some embodiments, generating a canopy height model includes: separating ground points and non-ground points from the point cloud of the UAV-Lidar (UAV-LS) data, respectively generating a digital elevation model (DEM) and a digital surface model (DSM), and calculating the canopy height model through pixel subtraction;

[0110] Generating a gradient map using the canopy height model includes: calculating the gradients in the horizontal and vertical directions using the Sobel operator, and calculating the square root of the sum of the squares of the gradients in the horizontal and vertical directions to generate a canopy height gradient map;

[0111] Performing individual tree segmentation on the gradient map using the watershed algorithm to extract the geometric parameters of each tree: including extracting local extreme points as the initial centers for individual tree segmentation; performing segmentation on the gradient map based on the watershed algorithm to divide the canopy area of each tree.

[0112] In some embodiments, the steps of generating a digital elevation model through the lowest point filtering technique include:

[0113] First, grid the point cloud, dividing the lidar point cloud data into regular grids according to a fixed size, i.e., 1 meter × 1 meter;

[0114] Second, perform lowest point screening. In each grid, screen out the point with the lowest height value as the ground point, and interpolate and complete the grids without points;

[0115] Third, perform terrain smoothing, performing Gaussian filtering on the height values of the screened lowest points to remove discrete noise;

[0116] Finally, generate the DEM, generating a continuous digital elevation model based on the height values of the lowest points within the grids to reflect the surface height of the terrain.

[0117] This model uses the following research data: UAV data, UAV-LS, and UAV multispectral data come from different sensors. Finally, digital orthophoto maps (DOM), digital surface models (DSM), and digital terrain models (DTM) are generated. Finally, a canopy height model (CHM) is generated based on these processing results, and the coordinate systems of all models are unified. For high-resolution remote sensing data, this invention selects the high-resolution series data. The high-resolution series satellites are the core of China's high-resolution earth observation satellite program, covering from Gaofen-1 to Gaofen-7. This series of satellites provides spatial resolutions ranging from 0.5 meters to 10 meters, covering panchromatic and multispectral images. High-resolution data is widely used in fields such as land use, urban planning, agricultural monitoring, environmental protection, and disaster assessment. The high-resolution series satellites have high spatial resolutions and diverse sensor configurations, capable of providing accurate data support for refined remote sensing applications and facilitating scientific research and decision-making.

[0118] As Figure 2 Shown by the measured data, the measured data collection of this invention was carried out in two years, respectively in July 2022 and August 2023. The measured data for both years were immediately taken after the UAV flights, and the measured sample plots (quadrats) were evenly distributed in the UAV images. In the plots sampled manually on the ground, the tree species names, diameter at breast height (DBH), and tree height of all trees with a DBH ≥ 5 cm in each plot were measured and recorded, and they were numbered manually. In addition, each tree in the sample plot was matched with the UAV images at the measurement site to determine the exact position of each tree in the UAV images, so as to verify the accuracy of single-tree segmentation.

[0119] This invention proposes a precise estimation model for small-area carbon sinks based on multiple constraints to quickly and accurately estimate the small-area carbon sink amount. As shown in the technical flowchart, this model is mainly divided into the following 5 steps:

[0120] 1. Data acquisition: Obtain UAV lidar (UAV-LS) data, UAV multispectral data, UAV hyperspectral data, and measured data; calculate and find the average value according to the allometric equations in the "Guidelines for Carbon Sink Measurement and Monitoring of Afforestation Projects" of the State Forestry Administration and the assessment of forest vegetation biomass and carbon storage in China. For example, this invention takes Pinus sylvestris var. mongolica as an example for mapping, collects the allometric equation of Pinus sylvestris var. mongolica for calculation and finds the average value. The allometric equation is as follows:

[0121]

[0122]

[0123]

[0124] Where Total aboveground biomass, is the biomass of the trunk, the biomass of the branches, the biomass of the leaves, D is the tree diameter at breast height, and H is the tree height.

[0125] 2. Point cloud data preprocessing and individual tree segmentation. For point cloud data preprocessing, when processing UAV-LS data, first improve the data accuracy through noise filtering. Noise filtering identifies and removes abnormal sparse points through statistical methods or clustering algorithms (such as DBSCAN) to ensure the accuracy of the data. Then, classify the ground points and non-ground points in the point cloud to generate a digital terrain model (DTM), and then combine the height information of the canopy points to generate a canopy height model (CHM). CHM can clearly reflect the canopy height and structure of each tree, providing a reliable basis for subsequent individual tree segmentation, tree height estimation, and carbon sink assessment; Individual tree segmentation

[0126] , based on CHM and UAV-LS data, perform individual tree segmentation and extract individual tree-related parameters, including canopy height, crown width, etc.

[0127] 3. Inversion of physical and chemical parameters, spectral data preprocessing and calibration. Due to different solar elevation angles, the spectral reflectance of multi- / hyperspectral image data obtained at different times is different, and calibration processing is required. The spectral data observed at any time is calibrated to a certain specified time through the kernel-driven BRDF model. The BRDF model can describe the reflection characteristics of surface objects and usually needs to consider multiple factors, including the incident angle, reflection angle, surface roughness of light, and the spectral characteristics of objects. Commonly used BRDF models include the Ross kernel and the Lambertian model, and these models can adjust the calculation method of reflectance according to different surface types (such as water bodies, soils, vegetation, etc.). The BRDF model is:

[0128]

[0129] In the process of data calibration, the kernel-driven method plays an important role. Through the kernel function (the Ross kernel is selected in this model), the scattering characteristics of light in the vegetation canopy can be simulated, and the change of reflectance under different lighting conditions can be calculated accordingly. The advantage of the kernel-driven BRDF model is that it can flexibly handle complex surface conditions, such as the multi-layer canopy structure in a forest environment, to ensure that the spectral reflectance can be accurately reflected at a certain specified time. Specifically, this process includes comparing the lighting conditions of the observed data with the conditions at the specified time and calculating the surface reflection value under this standard condition through the BRDF model. The kernel function is the Ross kernel (Ross-thin):

[0130]

[0131] Inversion of Leaf Area Index (LAI)

[0132] LAI is a key biophysical parameter of vegetation structure, which is used to describe the total leaf area per unit ground area in the vegetation canopy. It reflects the density and productivity of vegetation and is widely used in ecological, climatological, and agricultural research. The measurement of LAI is based on the absorption and scattering characteristics of light in the vegetation canopy and is calculated by measuring the proportion of light blocked by leaves when passing through the canopy. Specifically, when light passes through the vegetation canopy, part of the light is reflected or absorbed by the leaves, and the remaining part continues to pass through the vegetation. These light losses can be used to estimate the density and number of leaves. According to the Beer-Lambert law, the relationship between the transmittance and the canopy gap fraction is as follows:

[0133]

[0134] Inversion of Chlorophyll Content (P)

[0135] P is a physical and chemical parameter of plant growth and health status, which plays a key role especially in evaluating photosynthesis efficiency, nitrogen nutrition level, and crop yield. Chlorophyll mainly absorbs red light (650 - 680 nm) and blue light (430 - 460 nm) in the photosynthetically active radiation and reflects near-infrared light (700 - 900 nm). The spectral reflectance of vegetation is low in the red light band and high in the near-infrared band, forming a typical vegetation spectral curve. The underlying mechanism of this phenomenon is that chlorophyll in plant leaves absorbs a large amount of light energy for photosynthesis, and the near-infrared light not absorbed by chlorophyll is mostly reflected and scattered. The change of this spectral reflectance characteristic can be quantitatively analyzed by remote sensing data, thereby inversing the chlorophyll content.

[0136] 4. Setting of constraint conditions. As long as the forest grows normally, the annual cumulative carbon sink should be 0. To ensure that the estimated result of the forest carbon sink is non-negative during normal growth, it is necessary to set constraint conditions for its evaluation model. Setting of hard constraint conditions: Select the optimal CHM percentile statistical values for two years 、 , the crown projection areas for two years 、 ; As a hard constraint condition, it means that under the condition of normal tree growth, the value of the second time the first time, then and ; If there are problems with the segmentation results, resulting in or , then set: and ; Soft constraint setting. The leaf area index (LAI) and chlorophyll content (P), as the biophysical and physicochemical parameters of trees respectively, are positive at a certain specified moment and reflect the annual climate conditions. As the soft constraints of the individual tree carbon sink model, they can effectively limit the rationality of the model prediction results, ensure the accurate reflection of the tree health status and growth vitality during the carbon sink assessment process, and thus improve the robustness and adaptability of the model. The two-year average value is used in the model. ;

[0137] 5. Construction of the individual tree carbon sink estimation method. The methods for the individual tree carbon storage in two years are as follows:

[0138]

[0139]

[0140] Then the carbon sink amount is:

[0141]

[0142] And ensure , Note: where A is the canopy projection area, H is the canopy height, LAI is the leaf area index, and P is the chlorophyll content.

[0143] Optimization of the model method. To balance the soft and hard constraint settings, the objective function is:

[0144]

[0145] Among them, and are the weight coefficients of the hard and soft constraints respectively. By appropriately adjusting these two coefficients, the strict requirements of the hard constraint on the tree growth characteristics and the flexible reflection of the soft constraint on the ecological health status can be balanced in the model, so as to ensure that the change in the carbon sink amount △C predicted by the model is always zero. This adjustment not only helps to enhance the adaptability and robustness of the model, but also can effectively avoid the unreasonable carbon storage assessment caused by data fluctuations or abnormal results, and ensure the scientificity and practicality of the final model.

[0146] Such as Figure 3 shown, this note takes 50 measured Pinus sylvestris var. mongolica trees as an example to construct an individual tree carbon sink model, and the results calculated by the allometric growth equation are cross-validated during the modeling process. The quantitative accuracy evaluation is as Figure 2 shown. We select 4 indicators for accuracy evaluation, which are . Such as Figure 3As shown, we mapped the carbon storage and carbon sink of a 50-Pinus sylvestris var. mongolica plot over two years. It can be clearly seen from the map that there was no negative carbon sink in the plot during the two years, and the accuracy was ensured. Also, it can be seen from the map how the carbon sink levels were distributed in different regions and whether there were obvious high or low differences.

[0147] After small-area mapping using the single-tree carbon sink model constructed based on this model, it helps to identify the carbon sink distribution patterns and growth potential in different regions, especially providing a scientific basis for forest management, carbon sink optimization, and ecosystem carbon storage assessment. Through the mapping results, the carbon sink differences between different tree species or regions can also be visually analyzed, providing data support for the formulation of forest carbon sink policies and promoting the implementation of carbon sink trading and carbon sink projects.

[0148] The present invention provides a method and system for accurately estimating small-area forest carbon sinks based on multiple constraints. By comprehensively using unmanned aerial vehicle lidar (UAV-LS) data and multi-spectral / hyperspectral data, it solves many problems existing in traditional carbon sink estimation methods and has the following remarkable beneficial effects:

[0149] 1. High-precision carbon sink estimation. The present invention uses unmanned aerial vehicle lidar data to generate a digital elevation model (DEM) and a canopy height model (CHM), providing data support for the accurate description of forest structure through the extraction of high-precision geometric parameters (such as canopy height and canopy projection area). Combining the physical and chemical parameters (such as leaf area index LAI and chlorophyll content P) inverted from multi-spectral data realizes the deep integration of geometric information and spectral information, effectively improving the accuracy of carbon sink estimation. The accuracy of the inversion of physical and chemical parameters and the applicability of the model. As a characteristic representation of spectral data, the gray value reflects the local optical characteristics of pixels and is closely related to the physiological state and health status of vegetation. Based on the traditional inversion method, the present invention adds the gray value as a correction factor to the parameter calculation formula, realizing the deep integration of spectral characteristics and spatial distribution characteristics. Specifically, the introduction of the gray value effectively makes up for the limitation of relying solely on the difference or ratio of spectral bands in the traditional method, enabling the inversion model to more sensitively capture the spectral detail changes, especially performing more excellently under complex lighting conditions or heterogeneous canopy structures. At the same time, the inversion method combined with the gray value can enhance the characterization ability of the single-tree characteristic area, reducing the calculation deviation caused by local anomalies in spectral data, thus significantly improving the estimation accuracy of LAI and P parameters and providing more reliable data support for the construction of the carbon sink model. This improvement not only increases the innovation of the technical solution but also expands the diversity of its applicable scenarios.

[0150] 2. Automation processing and high efficiency. The present invention proposes a method for generating DEM based on the lowest point filtering technology, which can quickly remove noise points and non-ground points in the point cloud and realize automatic terrain modeling. At the same time, through the watershed algorithm, the gradient map generated by the canopy height model is segmented into individual trees. Without relying on manual operations or empirical parameters, it can efficiently extract the geometric parameters of each tree, significantly reducing labor costs and time consumption.

[0151] 3. Robustness for adapting to complex scenarios. The individual tree segmentation algorithm of the present invention is based on the canopy height model and the gradient map, combined with the boundary detection ability of the watershed algorithm. When dealing with scenarios of dense forests or blurred canopy boundaries, it can significantly improve the segmentation accuracy. Even in complex terrains and various forest types, this method can stably extract the geometric features of individual trees.

[0152] 4. Innovation of multi-dimensional information fusion. The present invention organically combines the geometric parameters of forests (such as canopy height, canopy projection area) and spectral information (such as leaf area index and chlorophyll content) to establish a carbon sink estimation model based on multiple constraints. By introducing geometric features and pixel gray values in the calculation process of the leaf area index and chlorophyll content, the comprehensive characterization ability of the model for forest ecological characteristics is enhanced, and the applicable range of parameter inversion is expanded. The technical solution proposed by the present invention is applicable to carbon sink monitoring requirements of various forest types (such as coniferous forests, broad-leaved forests, etc.) and different regions. By adjusting the weight coefficients (such as carbon storage coefficients), it can be flexibly applied to different climate conditions, vegetation types, and management objectives.

[0153] In addition to the above embodiments, the present invention may have other embodiments. Any technical solutions formed by equivalent replacement or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A small area carbon sink calculation method based on multiple constraints, characterized in that: Includes steps: S1: Acquire UAV-LS data and multispectral data; pre-process the multispectral data; S2: Single tree segmentation: filter the noise of UAV-LS data, remove ground points, generate a canopy height model, use the canopy height model to generate a gradient map, use the watershed algorithm to segment the gradient map, and extract the geometric parameters of each tree, including the canopy height. and canopy projected area , providing input for subsequent parameter inversion and carbon sink model construction; S3: Leaf area index LAI and chlorophyll content P inversion, based on the preprocessed spectral data, combined with the single tree segmentation results, the spectral characteristics within the single tree segmentation area are analyzed. According to the Beer-Lambert law, the leaf area index LAI is calculated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. The leaf area index LAI is logarithmically related to the canopy void ratio: in, is the zenith angle, which indicates the angle of incidence of the light; is the canopy gap fraction, which refers to the proportion of gaps in the canopy through which light can pass when viewed from the ground; The canopy light transmittance coefficient describes the ability of the vegetation canopy to absorb and scatter light; is the canopy height coefficient, is the canopy projected area Infrared spectrum infrared band, It is the near infrared band of infrared spectrum. is the first regression coefficient, is the second regression coefficient; is the grayscale value, is the grayscale mean of the canopy projection image; S4: Total carbon sink in small area The calculation is as follows: in, , Respectively The projected canopy area and canopy height of each tree; , Respectively Leaf area index and chlorophyll content of trees, is the first carbon storage coefficient, is the second carbon storage coefficient, is the third carbon storage coefficient.

2. A small area carbon sink calculation method based on multiple constraints as claimed in claim 1, characterized in that: Preprocessing the multispectral data includes: removing random noise in the multispectral data by median filtering.

3. A method for calculating carbon sinks in a small area based on multiple constraints as claimed in claim 1, characterized in that: Noise filtering is performed on the UAV-LS data to remove ground points, including: generating a digital elevation model through the lowest point filtering technology, calculating the height difference between the point cloud and the DEM, removing ground points, and retaining only vegetation points.

4. A method for calculating carbon sinks in a small area based on multiple constraints as claimed in claim 1, characterized in that: Generating a canopy height model includes: separating ground points and non-ground points from the UAV-LS data point cloud, generating a digital elevation model DEM and a digital surface model DSM respectively, and calculating the canopy height model by pixel subtraction; Generating a gradient map using a canopy height model includes: using a Sobel operator to calculate horizontal and vertical gradients, and calculating the square root of the sum of the squares of the horizontal and vertical gradients to generate a canopy height gradient map; The watershed algorithm is used to segment the gradient map into individual trees and extract the geometric parameters of each tree: including extracting local extreme points as the initial center of single tree segmentation; segmentation is performed based on the watershed algorithm on the gradient map to divide the canopy area of ​​each tree.

5. A method for calculating carbon sinks in a small area based on multiple constraints as claimed in claim 3, characterized in that: The steps to generate a digital elevation model using the lowest point filtering technique include: First, the point cloud is gridded, and the lidar point cloud data is divided into regular grids of a fixed size, i.e., 1 meter × 1 meter; Secondly, the lowest point screening, in each grid, screen out the point with the lowest height value as the ground point, and interpolate and complete the grid without points; Again, the terrain is smoothed, and the height values ​​of the lowest points screened out are processed by Gaussian filtering to remove discrete noise; Finally, DEM is generated, and a continuous digital elevation model is generated based on the height value of the lowest point in the grid to reflect the surface height of the terrain.

6. A small area carbon sink calculation system based on multiple constraints, characterized in that: include: Acquisition module, which acquires UAV-LS data and multispectral data; Preprocess multispectral data; The single tree segmentation module filters the noise of the UAV-LS laser radar data, removes the ground points, generates a canopy height model, uses the canopy height model to generate a gradient map, uses the watershed algorithm to segment the gradient map, and extracts the geometric parameters of each tree, including the canopy height. and canopy projected area , providing input for subsequent parameter inversion and carbon sink model construction; The leaf area index LAI and chlorophyll content P inversion module analyzes the spectral characteristics within the single tree segmentation area based on the preprocessed spectral data and the single tree segmentation results. According to the Beer-Lambert law, the leaf area index LAI is calculated by measuring the proportion of light blocked by leaves when passing through the vegetation canopy. The leaf area index LAI is logarithmically related to the canopy void ratio: in, is the zenith angle, which indicates the angle of incidence of the light; is the canopy gap fraction, which refers to the proportion of gaps in the canopy through which light can pass when viewed from the ground; The canopy light transmittance coefficient describes the ability of the vegetation canopy to absorb and scatter light; is the canopy height coefficient, is the canopy projected area Infrared spectrum infrared band, It is the near infrared band of infrared spectrum. is the first regression coefficient, is the second regression coefficient; is the grayscale value, is the grayscale mean of the canopy projection image; S4: Total carbon sink in small area The calculation is as follows: in, , Respectively The projected canopy area and canopy height of each tree; , Respectively Leaf area index and chlorophyll content of trees, is the first carbon storage coefficient, is the second carbon storage coefficient, is the third carbon storage coefficient.

7. A small area carbon sink calculation system based on multiple constraints as claimed in claim 6, characterized in that: Preprocessing the multispectral data includes: removing random noise in the multispectral data by median filtering.

8. A small area carbon sink calculation system based on multiple constraints as claimed in claim 5, characterized in that: Noise filtering is performed on the UAV-LS data to remove ground points, including: generating a digital elevation model through the lowest point filtering technology, calculating the height difference between the point cloud and the DEM, removing ground points, and retaining only vegetation points.

9. A small area carbon sink calculation system based on multiple constraints as claimed in claim 6, characterized in that: Generating a canopy height model includes: separating ground points and non-ground points from the UAV-LS data point cloud, generating a digital elevation model DEM and a digital surface model DSM respectively, and calculating the canopy height model by pixel subtraction; Generating a gradient map using a canopy height model includes: using a Sobel operator to calculate horizontal and vertical gradients, and calculating the square root of the sum of the squares of the horizontal and vertical gradients to generate a canopy height gradient map; The watershed algorithm is used to segment the gradient map into individual trees and extract the geometric parameters of each tree: including extracting local extreme points as the initial center of single tree segmentation; segmentation is performed based on the watershed algorithm on the gradient map to divide the canopy area of ​​each tree.

10. A method for calculating carbon sinks in a small area based on multiple constraints as claimed in claim 8, characterized in that: The steps to generate a digital elevation model using the lowest point filtering technique include: First, the point cloud is gridded, and the lidar point cloud data is divided into regular grids of a fixed size, i.e., 1 meter × 1 meter; Secondly, the lowest point screening, in each grid, screen out the point with the lowest height value as the ground point, and interpolate and complete the grid without points; Again, the terrain is smoothed, and the height values ​​of the lowest points screened out are processed by Gaussian filtering to remove discrete noise; Finally, DEM is generated, and a continuous digital elevation model is generated based on the height value of the lowest point in the grid to reflect the surface height of the terrain.

Images