A mangrove carbon sink quantity monitoring and metering method based on a drone, a radar and an AI technology

Abstract

The present application relates to the technical field of ecological environment protection, and more particularly to a mangrove carbon sink monitoring and metering method based on unmanned aerial vehicles, radars and AI technology, comprising: step 1, acquiring laser radar data of the mangrove by means of an unmanned aerial vehicle carrying a laser radar; step 2, acquiring elevation data in the vegetation layer area of the mangrove to obtain point cloud data after topographic error correction; step 3, acquiring multispectral images of the mangrove to obtain a multispectral data matrix; step 4, identifying forest growth data features, constructing a mangrove carbon sink prediction model, and predicting the mangrove carbon sink; and step 5, marking image regions with NDVI values higher than a preset NDVI threshold as leaf dense regions; using a photosynthetic depression factor formula to calculate a light depression factor according to the area of the leaf dense regions and the multispectral data matrix, using a regional carbon sink deviation formula to determine the carbon sink deviation of the leaf dense regions, and obtaining the true carbon sink value of the mangrove according to the predicted carbon sink calculation value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological environment protection, and particularly relates to a mangrove carbon sink monitoring and measuring method based on a UAV, a radar and AI technology. BACKGROUND

[0002] Mangroves, as an important ecological system, not only provide habitats for organisms, but also play a crucial role in global carbon cycling. Due to its strong carbon storage capacity, mangroves are considered as one of the key ecosystems to mitigate climate change. However, with the continuous impact of climate change and human activities, the area of mangroves is decreasing year by year, therefore, accurate assessment and monitoring of mangrove carbon sink is of great significance for environmental protection, carbon emission management and policy making. Currently, the growth characteristics of mangroves, such as tree height, diameter at breast height, crown width, etc., are the key parameters for carbon sink estimation, but the current technology is difficult to accurately obtain these information, especially in high-density mangrove areas, how to extract accurate growth characteristics from complex point cloud data still remains a challenge. At the same time, the carbon storage of mangrove ecosystem is affected by many factors, such as vegetation density, leaf state, etc. The dense area of leaves will cause photosynthesis inhibition phenomenon, which will affect the accurate estimation of carbon sink, but the existing technology is difficult to comprehensively consider these factors, resulting in the deviation of carbon sink prediction results. Multispectral remote sensing technology provides effective data support in vegetation coverage, leaf growth status, etc., but its data processing and analysis process also has many challenges, multispectral image is affected by noise, image distortion and light change, etc., the image quality and processing effect is more dependent on the environmental conditions in the acquisition process. In addition, the existing multispectral image processing technology still has certain technical problems in band alignment, image standardization and feature extraction, etc., which affects the accuracy of multispectral data in mangrove carbon sink estimation. Although in recent years, the application of deep learning technology in remote sensing data processing has made significant progress, but the existing carbon sink prediction model still faces the challenge of data fusion when processing multi-source data. How to effectively fuse laser radar point cloud data, multispectral remote sensing image data and other growth data, and train through a deep learning model to realize accurate carbon sink prediction, is still a problem to be solved. SUMMARY

[0003] The present application provides a mangrove carbon sink monitoring and measuring method based on a UAV, a radar and AI technology to solve the above problems of the prior art, mainly including:

[0004] Step 1, acquiring laser radar data of mangroves by UAV carrying laser radar, acquiring remote sensing image data by satellite remote sensing technology, classifying and filtering the laser radar point cloud data, identifying the ground layer and vegetation layer of the forest, and marking signal class labels for the ground layer and vegetation layer respectively;

[0005] Step 2, according to the vegetation layer data obtained in step 1, obtain elevation data in the mangrove vegetation layer area, and combine the point cloud data, elevation data and signal category label to construct a mangrove point cloud data correction model, correct the obtained point cloud data, and obtain point cloud data after topographic error correction;

[0006] Step 3, a multi-spectral imaging device is carried by the unmanned aerial vehicle to obtain multi-spectral images of the mangrove forest, and the multi-spectral images are denoised, SIFT key feature points are extracted, and affine transformation is performed on each different waveband image to form a multi-spectral data matrix. After standardizing the data of each waveband, the multi-spectral data matrix is obtained.

[0007] Step 4, according to the corrected point cloud data in step 2, identify forest growth data features, use a recurrent neural network to train a model according to the forest growth data features, remote sensing image data obtained in step 1 and multi-spectral data matrix obtained in step 4, and construct a mangrove carbon sink prediction model to predict the amount of carbon sink of the mangrove forest.

[0008] Step 5, according to the NDVI value obtained in step 1, mark the image area with an NDVI value higher than the preset NDVI threshold as a leaf dense area; determine the area of the leaf dense area by calculating the number of pixels in the leaf dense area; calculate the light suppression factor using the photosynthetic suppression factor formula according to the area of the leaf dense area and the multi-spectral data matrix; determine the carbon sink deviation of the leaf dense area using the regional carbon sink deviation formula according to the calculated light suppression factor; and obtain the true carbon sink value of the mangrove forest according to the carbon sink deviation of the leaf dense area and the calculated value of the carbon sink obtained by the mangrove carbon sink prediction model constructed in step 4.

[0009] Further, step 1 includes obtaining laser radar point cloud data of the mangrove forest area according to the setting of the unmanned aerial vehicle carrying the laser radar sensor, and obtaining remote sensing image data using satellite remote sensing technology. The laser radar point cloud data includes three-dimensional coordinate information and signal reflection intensity of each point, and the remote sensing image data includes forest coverage area, NDVI and biomass. According to the obtained original point cloud data, noise removal is performed by using high-low threshold filtering, voxel grid filtering or statistical outlier filtering algorithm. According to the point cloud data after filtering, the height relationship of each point relative to the surrounding points is determined, and the points lower than the set threshold are classified as ground points, and the points higher than the set threshold are classified as vegetation points, to identify the ground layer and the vegetation layer of the mangrove forest.

[0010] Further, step 2 comprises obtaining elevation data in the mangrove vegetation layer region, the elevation data comprising the slope and elevation of the region; performing spatial interpolation on the elevation data in the mangrove vegetation layer region by Kriging or inverse distance weighting method to obtain the elevation value of each pixel position; using the obtained point cloud data, signal category label, elevation data and corrected point cloud data, using the PointNet algorithm for model training, constructing a mangrove point cloud data correction model, correcting the obtained point cloud data to obtain the point cloud data corrected by the terrain error, and the signal category label comprises the ground layer and the vegetation layer.

[0011] Further, step 3 comprises obtaining a multispectral image of the mangrove by a multi-spectral imaging device carried by a UAV, removing noise from the multispectral image using a median filter, and detecting and matching SIFT key feature points in different waveband images using a SIFT feature point matching algorithm, and aligning different waveband images through affine transformation; integrating the reflectivity data of each pixel point in different wavebands into a vector to form a multispectral data matrix, the size of the multispectral data matrix being n x m, where n is the number of pixel points and m is the number of wavebands, each row of the multispectral data matrix representing the reflectance data of a pixel point in all wavebands; standardizing the data of each waveband by subtracting the mean value and dividing by the standard deviation to obtain a standardized multispectral data matrix.

[0012] Further, step 4 comprises identifying forest growth data features from the corrected laser radar data of the mangrove, the forest growth data features comprising tree species distribution, tree height, diameter at breast height, crown width, age structure, using a recurrent neural network to train a model based on the forest growth data features, remote sensing image data and multispectral data matrix, and constructing a mangrove carbon sink prediction model to predict the mangrove carbon sink.

[0013] Further, step 5 comprises calculating the NDVI value from the multispectral image and generating an NDVI distribution image of the mangrove vegetation cover, mapping the NDVI distribution image to a grayscale image, binarizing the grayscale image based on a preset NDVI threshold as a segmentation criterion, and marking the image area with an NDVI value higher than the preset NDVI threshold as a leaf dense area; using a Canny edge detection algorithm to extract the area contour of the leaf dense area, determining the area of the leaf dense area by counting the number of pixels in the leaf dense area; using the photosynthetic depression factor formula to determine the photosynthetic depression factor of the leaf dense area, wherein is the number of pixels in the region, m is the number of wavebands, is the normalized reflectivity of the kth pixel in the jth waveband, , represents the average value of the jth wave band in region i; according to the photosynthetic depression factor of the leaf over-dense region, the regional carbon sink deviation formula is used , to determine the carbon sink deviation of the leaf over-dense region , wherein, is the carbon sink down-regulation coefficient per unit area, which is obtained by fitting historical data, is the area of the leaf over-dense region, is an exponential amplification of the total deviation based on the number of pixels in the region and the shading density, is a fine-tuning coefficient, which is obtained by fitting historical data; according to the carbon sink deviation of the leaf over-dense region and the calculated value of the carbon sink obtained by the step 4, the real carbon sink value of the mangrove forest is obtained.

[0014] The technical scheme provided by the embodiment of the present application can include the following beneficial effects:

[0015] The present application effectively improves the measurement accuracy of the carbon sink capacity of mangrove forest by fusing laser radar point cloud data and multispectral data matrix and using recurrent neural network to construct a mangrove forest carbon sink prediction model. It innovatively realizes the quantitative integration of the photosynthetic depression effect caused by the sub-saturation state of the closed canopy layer of mangrove forest. Through the calculated normalized difference vegetation index value, the spatial range and area of the leaf over-dense region are identified and determined by the preset normalized difference vegetation index threshold. Using the area and multispectral data matrix, the light depression factor is calculated by using the photosynthetic depression factor formula. Combined with the regional carbon sink deviation formula, the carbon sink capacity loss caused by the photosynthetic depression effect in the leaf over-dense region, i.e. the carbon sink deviation, is accurately quantified. Finally, the preliminary carbon sink estimation value output by the mangrove forest carbon sink prediction model is corrected and compensated for this carbon sink deviation, and the real carbon sink value more consistent with the actual physiological and ecological response of mangrove forest is obtained. The present scheme effectively solves the core problem of systematic overestimation of carbon sink capacity caused by ignoring the photosynthetic depression effect of the internal crown layer in the prior art, significantly improves the measurement error caused by the over-dense crown layer in the mature forest area, and provides a more accurate data basis for blue carbon accounting. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is the step flow chart of the present application;

[0017] Figure 2 is the step flow chart of the present application for constructing a mangrove forest carbon sink prediction model;

[0018] Figure 3 is the calculation step flow chart of the real carbon sink value of the mangrove forest of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be described in detail below with reference to the drawings and specific embodiments.

[0020] The present application takes a coastal mangrove wetland in Futian District of Shenzhen as an example. The mangrove area is about 126.09 hectares, belonging to the South Asia tropical marine monsoon climate zone, and having the properties of estuary and bay. The interaction between river and sea, the rich fine sediment deposition and the fertile water quality provide a good geomorphic environment for the development of mangrove. The appearance of the mangrove community in the region is simple, with neat canopy, aggregated distribution and single stand composition, mostly composed of dominant mangrove communities of single or two or three mangrove plants, with the highest tree height of about 15 m. Part of the artificially planted Terminalia mangrove is about 7 m high, and the shrubs growing under the mangrove are about 1.5 m high, often forming dominant mangrove communities with other dominant mangrove plants. The Elaeocarpus fordii is distributed in clusters on the outer edge of the community close to the sea, and also sporadically distributed under the canopy of the mangrove community, with a height of about 2 m.

[0021] Please refer to Figure 1 The present application is a kind of mangrove carbon sink monitoring and metering method based on unmanned aerial vehicle, radar and AI technology, which can specifically include:

[0022] Step 1, acquiring laser radar data of mangrove by unmanned aerial vehicle carrying laser radar, obtaining remote sensing image data by satellite remote sensing technology, classifying and filtering the laser radar point cloud data, identifying the ground layer and vegetation layer of the forest, and marking the signal class label for the ground layer and vegetation layer respectively. Among them, step 1 includes: according to the setting of laser radar sensor carried by unmanned aerial vehicle, acquiring laser radar point cloud data of mangrove area, and obtaining remote sensing image data by satellite remote sensing technology, the laser radar point cloud data includes three-dimensional coordinate information and signal reflection intensity of each point, and the remote sensing image data includes forest coverage area, NDVI and biomass; according to the obtained original point cloud data, high and low threshold filtering, voxel grid filtering or statistical outlier filtering algorithm is adopted for noise removal; according to the point cloud data after filtering processing, the height relationship of each point relative to the surrounding points is judged, the points lower than the set threshold are classified as ground points, and the points higher than the set threshold are classified as vegetation points, and the ground layer and vegetation layer of mangrove are identified.

[0023] Specifically, the unmanned aerial vehicle is equipped with RIEGL VUX-1LR laser radar system, the flight height is set to 150 meters, and the original point cloud data with a point density of 150 points per square meter is obtained. The satellite remote sensing platform Landsat 8 is used to obtain the remote sensing image of the mangrove forest area, and the image data of different bands is collected. The original satellite image is geometrically corrected to ensure that the image coordinate system is consistent with the laser radar point cloud data coordinate system. According to the satellite remote sensing image, the near-infrared and red light band remote sensing image data are used to calculate the NDVI value, and the NDVI threshold is set to 0.2-0.6 to filter out the area with high vegetation coverage. The forest area boundary is extracted to determine the forest coverage area of the mangrove forest area. According to the relationship between NDVI and biomass, a vegetation biomass inversion model such as a spectral feature regression model is used to generate the biomass and biomass distribution map of the entire mangrove forest area combined with the remote sensing image data. For the high-altitude noise and low-altitude noise in the original point cloud data, such as flying birds and tide reflection, first, high and low threshold filtering is applied to remove abnormal points with an elevation value higher than 60 meters and lower than the chart datum-2 meters. Then, voxel grid filtering is used to fuse multiple points in a 0.1m x 0.1m x 0.1m cube into a single point to reduce the amount of redundant data. Finally, statistical outlier filtering is performed to calculate the elevation standard deviation of 15 neighboring points within a radius of 0.5 meters centered on each point, and remove outliers deviating from the mean by more than 1.5 times the standard deviation. After three levels of filtering, the total number of points is reduced from the original 5 million points to 4.2 million effective points. On this basis, the ground and vegetation are separated by height threshold method. The classification threshold is set to 0.3 meters, points with an elevation lower than the surrounding terrain by 0.3 meters are classified as ground points including mudflats and tidal ditches, and points with an elevation greater than 0.3 meters are classified as vegetation points including mangrove prop roots and tree crowns. The average canopy height of the Kandelia obovata community is successfully identified as 8.2 meters.

[0024] Step 2, according to the vegetation layer data obtained in step 1, the elevation data in the vegetation layer area of the mangrove forest is obtained, and the point cloud data correction model of the mangrove forest is constructed combined with the point cloud data, the elevation data and the signal category label. The obtained point cloud data is corrected to obtain the point cloud data after topographic error correction. In step 2, the elevation data in the vegetation layer area of the mangrove forest is obtained, and the elevation data includes the slope and elevation of the area. The elevation data in the vegetation layer area of the mangrove forest is spatially interpolated by Kriging method or inverse distance weighted method to obtain the elevation value of each pixel position. According to the obtained point cloud data, signal category label, elevation data and corrected point cloud data, the PointNet algorithm is used for model training to construct the point cloud data correction model of the mangrove forest. The obtained point cloud data is corrected to obtain the point cloud data after topographic error correction. The signal category label includes the ground layer and the vegetation layer.

[0025] Specifically, to avoid the original vegetation layer point cloud acquired by the laser radar being affected by the gully terrain, systematic elevation deviation occurs in the slope area. To ensure the accuracy of subsequent photosynthetic suppression effect quantification, first, the elevation characteristics of the area are extracted: slope gradient 0.8°-7.5°, altitude interval 1.8-4.3 meters. The kriging interpolation method is used to reconstruct the terrain surface, with a range parameter of 18 meters / block and a gold value of 0.15, to generate a 1-meter resolution elevation grid. The interpolated elevation, original point cloud, and manually labeled vegetation layer / ground layer labels are input into the PointNet model for training, focusing on learning the elevation error pattern caused by the gully edge due to laser incidence angle distortion. After correction, the elevation deviation of the Baiguangzhang pillar root aggregation area decreases from 0.25 meters to 0.03 meters, for example, the crown layer point cloud elevation at a certain location is restored from 3.62 meters before correction to the actual 3.35 meters. This terrain correction directly affects the accuracy of crown structure analysis. If not corrected, the crown height in steep slope areas will be overestimated, leading to misjudgment of the NDVI threshold in dense leaf areas, such as identifying inclined crown as vertical dense canopy area. At the same time, the height error of the pillar root will distort the calculation of the crown thickness parameter in the light suppression factor formula, ultimately affecting the accuracy of the carbon sink deviation. This step lays the foundation for subsequent elimination of terrain interference on crown photosynthetic efficiency evaluation.

[0026] Step 3, acquiring the multispectral image of the mangrove forest by the unmanned aerial vehicle carrying the multispectral imaging device, denoising the multispectral image, extracting SIFT key feature points, and aligning each different waveband image through affine transformation to form a multispectral data matrix. After standardizing the data of each waveband, the multispectral data matrix is obtained. In step 3, the multispectral imaging device is carried by the unmanned aerial vehicle to acquire the multispectral image of the mangrove forest, the median filter is used to remove noise from the multispectral image, and the SIFT feature point matching algorithm is used to detect and match SIFT key feature points in different waveband images, and affine transformation is used to align different waveband images. The reflectivity data of each pixel point in different wavebands is integrated into a vector to form a multispectral data matrix, and the size of the multispectral data matrix is n x m, where n is the number of pixel points and m is the number of wavebands. Each row of the multispectral data matrix represents the reflectance data of a pixel point in all wavebands. The reflectivity data of each waveband is standardized by subtracting the mean value and dividing by the standard deviation to obtain the standardized multispectral data matrix.

[0027] Specifically, the unmanned aerial vehicle carries a RedEdge-MX multispectral camera with wave bands of blue 475 nm, green 560 nm, red 668 nm, and red edge 717 nm / 842 nm to fly and obtain multispectral images of the white bone soil community. To eliminate the image noise caused by the reflection of tides, a 5x5 pixel window median filter is used, such as reducing the abnormal fluctuation of the pixel value of a certain tidal channel area in the 717 nm wave band from ± 25 to ± 8. Due to the influence of wind during the rising tide, the wave bands are offset, and the SIFT algorithm is used to match the feature points. About 1500 key points are extracted from a single image, and it is found that the red edge wave band has an average displacement of 1.8 pixels relative to the blue wave band, about 0.27 meters on the ground. Through the affine transformation matrix, the accurate alignment is completed, and the residual error is less than 0.3 pixels. After alignment, the reflectance of each pixel point in the 5 wave bands is integrated into a vector to form a multispectral data matrix with a pixel resolution of 1024x768, a total of 786432 rowsx5 columns, and the data in the 3rd column of the kth row represents the reflectance of the kth pixel point in the red wave band. To eliminate the influence of different light conditions, each wave band is standardized, and the average reflectance of all pixels in the blue wave band is calculated as 0.21 and the standard deviation is 0.07. The original value 0.35 is converted to (0.35-0.21) / 0.07=2.0. This standardization matrix directly serves the NDVI threshold 717 nm / 842 nm wave band analysis in step 5, calculates the reflectance parameter input in the NDVI and light pressure suppression factor formula, and ensures that the subsequent photosynthetic suppression effect quantification is not disturbed by atmospheric scattering.

[0028] Please refer to Figure 2 , step 4, according to the corrected point cloud data in step 2, identify the forest growth data characteristics, according to the forest growth data characteristics, the remote sensing image data obtained in step 1 and the multispectral data matrix obtained in step 3, use recurrent neural network for model training, build mangrove carbon sink prediction model, predict the amount of mangrove carbon sink. Wherein, step 4 includes: according to the corrected laser radar data of mangrove, identifying forest growth data characteristics, forest growth data characteristics including tree species distribution, tree height, diameter at breast height, crown width, age structure, according to forest growth data characteristics, remote sensing image data and multispectral data matrix, using recurrent neural network for model training, building mangrove carbon sink prediction model, predicting mangrove carbon sink.

[0029] Specifically, based on the point cloud data of step 2, the single tree in the mangrove area is segmented and the structural features are extracted. The DBSCAN algorithm based on density clustering is used to cluster the point cloud, and the spatial neighborhood radius ε is set to 0.45 meters and the minimum neighbor point number MinPts is set to 25 to eliminate excessive connection between support roots. The segmentation results show that the number of single trees in the white bone soil area is 1265, and the number of single trees in the kandelia candel area is 842. On the basis of single tree point cloud, various forest growth data features are calculated, and the height difference between the highest point and the lowest ground point of the single tree is obtained, such as the sample height of kandelia candel 8.34 meters and the sample height of white bone soil 6.97 meters. The point cloud slice at 1.3 meters from the ground is intercepted, and the diameter is calculated by fitting a circular section, for example, the breast diameter of kandelia candel is 0.21 meters, and the breast diameter of white bone soil is 0.17 meters. The crown points are projected onto the horizontal plane, and the average length and short axis of the minimum circumscribed ellipse are calculated, and the average crown width of kandelia candel is 3.62 meters, and the average crown width of white bone soil is 2.95 meters. According to the specific growth curve model of the species, the age distribution of single tree is calculated, and the age of kandelia candel is concentrated in 12-18 years, and the age of white bone soil is concentrated in 9-14 years. The species distribution information is obtained by combining the red edge band characteristics of Landsat8 remote sensing image in step 1 and the multispectral matrix reflectance data in step 3, using random forest pre-classification, and the spatial distribution map of kandelia candel accounts for 54.8%, white bone soil accounts for 41.6%, and wood olive accounts for 3.6%. The extracted forest growth feature vector, including species, tree height, diameter at breast height, crown width, age, vegetation index NDVI of remote sensing image, and 5-band reflectance value of multispectral matrix, are input into the recurrent neural network for model training, and a mangrove carbon sink prediction model is constructed. RNN adopts a three-layer structure, the feature dimension of the input layer = 15, including growth characteristics, spectral characteristics and vegetation index, the number of hidden layer nodes is 64 and 32 respectively, the activation function uses ReLU, and the output layer is the predicted value of carbon sink. The ratio of training set to validation set is 7:3, the optimizer is Adam, the initial learning rate is 0.001, and the loss of validation set converges to 0.018 after 1000 iterations. On the test data set, the R² of the model reaches 0.94, p>0.05, and the unbiased test passes. According to the constructed mangrove carbon sink prediction model, the monthly average carbon sink of mangrove is 255.3 kgC.

[0030] Please refer to Figure 3, step 5, calculating the NDVI value according to the multispectral data matrix obtained in step 3, marking the image area with NDVI value higher than the preset NDVI threshold as the leaf dense area; determining the area of the leaf dense area by calculating the number of pixels in the leaf dense area, calculating the photosynthetic depression factor according to the area of the leaf dense area and the multispectral data matrix using the photosynthetic depression factor formula, determining the carbon sink deviation of the leaf dense area according to the calculated photosynthetic depression factor using the regional carbon sink deviation formula, and obtaining the true carbon sink value of the mangrove according to the carbon sink deviation of the leaf dense area and the calculated carbon sink value obtained by the mangrove carbon sink prediction model constructed in step 4.

[0031] , step 5 includes calculating the NDVI value by multispectral image and generating the NDVI distribution image of the mangrove vegetation coverage, mapping the NDVI distribution image into a gray image, performing binaryzation processing on the gray image based on the preset NDVI threshold as the segmentation standard, marking the image area with NDVI value higher than the preset NDVI threshold as the leaf dense area; using the Canny edge detection algorithm to extract the area contour of the leaf dense area, and determining the area of the leaf dense area by calculating the number of pixels in the leaf dense area; calculating the photosynthetic depression factor of the leaf dense area according to the area of the leaf dense area and the multispectral data matrix using the photosynthetic depression factor formula , wherein, is the number of pixels in the region, and m is the number of bands, is the normalized reflectance of the kth pixel in the jth band, , which represents the average value of the jth band in the region i; determining the carbon sink deviation of the leaf dense area according to the photosynthetic depression factor of the leaf dense area using the regional carbon sink deviation formula , wherein, is the unit area carbon sink down-regulation coefficient, which is obtained by fitting historical data, is the area of the leaf dense area, is an exponential amplification of the total deviation based on the number of pixels in the region and the shading density, is a fine-tuning coefficient, which is obtained by fitting historical data; obtaining the true carbon sink value of the mangrove according to the carbon sink deviation of the leaf dense area and the calculated carbon sink value obtained by the mangrove carbon sink prediction model constructed in step 4.

[0032] Specifically, through NDVI calculation and leaf dense area identification, the 842nm and 668nm band reflectance of the target grid is extracted from the standardized multispectral data matrix generated in step 3, which has a size of 1024x768 pixelsx5 bands. The reflectance of pixel k=42 is: .

[0033] Calculate NDVI:​

[0034] ;

[0035] ;

[0036] Pre-set NDVI threshold = 0.68, which is set based on the critical value of LAI > 4.0. Set the pixel with NDVI ≥ 0.68 as 1, and the rest as 0. Statistics of connected regions, mark out 92 continuous super-threshold pixels including pixel 57. Using Canny algorithm parameters, set Gaussian kernel σ = 1.0, low threshold = 0.1, high threshold = 0.3, output closed polygon area 185 ± 1.5 m², corresponding to 92 1m × 1m pixels.

[0037] According to the area of the leaf dense area and the multi-spectral data matrix, the photosynthetic suppression factor formula of the leaf dense area is determined , wherein, is the number of pixel points in the region, m is the number of wave bands, is the normalized reflectivity of the kth pixel in the jth wave band, , which represents the average value of the jth wave band in the region i.

[0038] Reflectivity mean calculation: according to the formula, the average value of each wave band is calculated for the 5 wave bands of the 92 pixels in the region:

[0039] .

[0040] Single wave band standard deviation calculation:

[0041] Normalized reflectivity of pixel k = 57: ;

[0042] Deviation square: ;

[0043] Region summation: ;

[0044] Standard deviation: ;

[0045] LPSI integration calculation:

[0046]

[0047] Carbon sink deviation Calculation parameter acquisition:

[0048] According to the flux tower data fitting, the average loss rate of the shaded area is 18.2%, α = 0.18, and the Canny contour area The area is 185 m², the laser radar point cloud generates a canopy height model, the surface concave-convex degree is 3.8, γ = 0.003. The total number of regional pixels n_i = 92, the tidal humidity correction term, δ = 0.015.

[0049] Log item calculation:

[0050] ;

[0051] .

[0052] Carbon sink deviation integration:

[0053] .

[0054] Real carbon sink value generation:

[0055] The carbon sink output of step 4 is 255.3 kgC, and the unit conversion gets a surface density of .

[0056] Real carbon sink value = 255.3 - 10.715 = 244.585 kgC.

[0057] Unit conversion:

[0058] .

[0059] Model verification: In the 185m² area, 6 LI-6800 leaf chambers are arranged according to the hexagonal grid, and the photosynthetic parameters of the lower leaves (2.5-3.0 meters away from the canopy top) are as shown in the following table:

[0060] Table 1 Photosynthetic parameter table

[0061]

[0062] Precision comparison:

[0063] This scheme output: 1.322 kgC / m² → deviation + 0.38%

[0064] Uncorrected model: 1.38 kgC / m² → deviation + 4.78%

[0065] Traditional NDVI model: 1.41 kgC / m² → deviation + 7.06%

[0066] The verification can correct and compensate the preliminary carbon sink estimation value output by the mangrove carbon sink prediction model to obtain a real carbon sink value that conforms to the actual physiological and ecological response of the mangrove. The scheme effectively solves the core problem that the existing technology systematically overestimates the carbon sink capacity due to the photosynthetic depression effect caused by the neglect of light competition in the canopy, significantly improves the measurement error caused by the over-dense canopy in the mature forest area, and provides a more accurate data basis for blue carbon accounting.

[0067] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. It should be understood by those skilled in the art that the scope of the application involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and also covers other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the technical solutions formed by replacing the above features with the technical features disclosed in the present application (but not limited to) having similar functions.

Claims

1. A mangrove carbon sink monitoring and measuring method based on unmanned aerial vehicles, radars and AI technology, characterized in that, The method comprises: Step 1, acquiring laser radar data of the mangrove forest by a UAV carrying a laser radar, acquiring remote sensing image data by satellite remote sensing technology, classifying and filtering the laser radar point cloud data, identifying the ground layer and the vegetation layer of the forest, and marking signal category labels for the ground layer and the vegetation layer respectively; Step 2, according to the vegetation layer data obtained in step 1, acquiring elevation data in the vegetation layer region of the mangrove forest, and combining the point cloud data, the elevation data and the signal category labels to construct a mangrove point cloud data correction model, correcting the obtained point cloud data to obtain point cloud data after topographic error correction; Step 3, acquiring multispectral images of the mangrove forest by a UAV carrying a multispectral imaging device, denoising the multispectral images, extracting SIFT key feature points, and aligning the images of different wavebands by affine transformation to form a multispectral data matrix, and obtaining the multispectral data matrix after standardizing the data of each waveband; Step 4, identifying forest growth data features according to the corrected point cloud data in step 2, using recurrent neural network for model training according to the forest growth data features, the remote sensing image data obtained in step 1 and the multispectral data matrix obtained in step 3, constructing a mangrove carbon sink prediction model, and predicting the carbon sink of the mangrove forest; Step 5, according to the NDVI value obtained in step 1, marking the image region with an NDVI value higher than a preset NDVI threshold as a leaf dense region; determining the area of the leaf dense region by calculating the number of pixels in the leaf dense region, calculating the light suppression factor using the photosynthetic suppression factor formula according to the area of the leaf dense region and the multispectral data matrix, determining the carbon sink deviation of the leaf dense region using the regional carbon sink deviation formula according to the calculated light suppression factor, and obtaining the true carbon sink value of the mangrove forest according to the carbon sink deviation of the leaf dense region and the calculated value of the carbon sink obtained by the mangrove carbon sink prediction model constructed in step 4.

2. The method according to claim 1, wherein, Step 1 comprises: acquiring laser radar point cloud data of the mangrove forest region according to the setting of the laser radar sensor carried by the UAV, and acquiring remote sensing image data by satellite remote sensing technology; the laser radar point cloud data includes three-dimensional coordinate information and signal reflection intensity of each point, and the remote sensing image data includes forest coverage area, NDVI and biomass; according to the obtained original point cloud data, noise removal is performed by using high-low threshold filtering, voxel grid filtering or statistical outlier filtering algorithm; according to the point cloud data after filtering, the height relationship of each point relative to the surrounding points is judged, the points lower than the set threshold are classified as ground points, and the points higher than the set threshold are classified as vegetation points, and the ground layer and the vegetation layer of the mangrove forest are identified. 3.The method of claim 1, wherein, Step 2 comprises: acquiring elevation data in the mangrove vegetation layer area, the elevation data including the slope and elevation of the area; performing spatial interpolation on the elevation data in the mangrove vegetation layer area by Kriging method or inverse distance weighting method to obtain the elevation value of each pixel position; using PointNet algorithm to train the model according to the acquired point cloud data, signal category label, elevation data and corrected point cloud data, and constructing a mangrove point cloud data correction model to correct the acquired point cloud data to obtain the point cloud data corrected by the terrain error, and the signal category label includes the ground layer and the vegetation layer.

4. The method of claim 1, wherein the method further comprises: Step 3 comprises: acquiring a multispectral image of the mangrove by a multi-spectral imaging device carried by a UAV, removing noise from the multispectral image using a median filter, and detecting and matching SIFT key feature points in different waveband images using a SIFT feature point matching algorithm, and aligning different waveband images through affine transformation; integrating the reflectivity data of each pixel point in different wavebands into a vector to form a multispectral data matrix, the size of the multispectral data matrix being n x m, wherein n is the number of pixel points and m is the number of wavebands, and each row of the multispectral data matrix represents the reflectivity data of a pixel point in all wavebands; standardizing the data of each waveband by subtracting the mean value and dividing by the standard deviation to obtain a standardized multispectral data matrix.

5. The method according to claim 1, wherein the method is characterized in that: Step 4 comprises: identifying forest growth data features according to the corrected laser radar data of the mangrove, the forest growth data features including tree species distribution, tree height, diameter at breast height, crown width, age structure, using a recurrent neural network to train the model according to the forest growth data features, remote sensing image data and multispectral data matrix, and constructing a mangrove carbon sink prediction model to predict the mangrove carbon sink. 6.The mangrove carbon sink monitoring and measuring method based on UAV, radar and AI technology according to claim 1, wherein, Step 5 comprises generating an NDVI distribution image of the mangrove vegetation coverage area using the NDVI value obtained in step 1, mapping the NDVI distribution image to a grayscale image, performing binaryzation processing on the grayscale image based on a preset NDVI threshold as a segmentation standard, and marking the image area with an NDVI value higher than the preset NDVI threshold as a leaf dense area; using a Canny edge detection algorithm to extract the area contour of the leaf dense area, and determining the area of the leaf dense area by calculating the number of pixels in the leaf dense area; According to the area of the leaf over-dense area and the multi-spectral data matrix, using the photosynthetic depression factor formula , determine the photosynthetic depression factor of the leaf over-dense area, wherein, is the number of pixel points in the area, m is the number of wave bands, is the normalized reflectivity of the kth pixel in the jth wave band, , indicates the average value of the jth wave band in the area i; according to the photosynthetic depression factor of the leaf over-dense area, using the area carbon sink deviation formula , determine the carbon sink deviation of the leaf over-dense area , wherein, is the carbon sink down-regulation coefficient per unit area, which is obtained by fitting historical data, is the area of the leaf over-dense area, is an exponential amplification of the total deviation based on the number of pixel points in the area and the shading density, is a fine-tuning coefficient, which is obtained by fitting historical data; according to the carbon sink deviation of the leaf over-dense area and the calculated value of the carbon sink obtained by the mangrove carbon sink prediction model constructed in step 4, the real carbon sink value of the mangrove is obtained.

Images