Mangrove forest LAI layered remote sensing inversion method under complex weather and related equipment

Abstract

The invention discloses a mangrove forest LAI layered remote sensing inversion method in complex weather. The method mainly comprises the following steps: acquiring unmanned aerial vehicle laser radar point cloud data, a hyperspectral remote sensing image, a mangrove forest actually measured leaf area index and clearance rate data of a mangrove forest region; based on the LiDAR point cloud data and the actually measured clearance rate, adopting a semi-supervised sample generation method to generate a mangrove forest LAI sample; layering cloud coverage and clear sky reflectivity based on a response relationship between mangrove forest LAI samples and hyperspectral remote sensing image reflectivity; identifying an optimal feature for inverting the mangrove forest LAI from the spectral feature data set and the LiDAR feature data set; and constructing mangrove forest LAI layered inversion models suitable for a cloud coverage area and a clear sky area respectively, and further realizing mangrove forest LAI spatial mapping based on the optimal layered inversion model. The method provides a solution for mangrove forest LAI inversion under the alternation of cloud coverage and clear sky, and further realizes mangrove forest LAI accurate and rapid inversion under complex weather conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wetland vegetation parameter remote sensing monitoring, and particularly relates to a mangrove LAI hierarchical remote sensing inversion method under complex weather and related equipment. BACKGROUND

[0002] LAI as an important parameter representing the canopy structure of mangrove, it is closely related to the interaction of vegetation and external material and energy, transpiration, photosynthesis, etc. But most of the low spatial resolution LAI products are difficult to accurately depict the growth status of mangrove, such as CYCLOPES (about 1km), MOD15 (about 500m), GIMMS LAI4g (about 8km), GLASS (about 5km) and the like. At the same time, some of the products do not meet the LAI estimation accuracy requirements set by the Global Climate Observing System (GCOS). Therefore, it is urgent to realize the high spatial resolution LAI inversion of mangrove by using remote sensing technology to monitor the growth status of mangrove, which is of great importance to the protection and restoration of mangrove.

[0003] From 1979 to 2001, 75% of the NOAA HIRS data observed cloud cover, which shows that cloud cover has been a problem for quantitative retrieval of vegetation parameters. The weather in the mangrove area of China changes significantly, and sunny and cloudy weather appears frequently in a day. At the same time, the change of LAI during the growth period of mangrove is a highly dynamic process. This inevitably requires the use of unmanned aerial vehicles to obtain optical images in the sunlit area and the cloud-shaded area to monitor the growth of mangrove under the weather of alternating cloud cover and clear sky. But the obtained images will show a clear separation of bright and dark areas. In this case, the spectral reflectance of dark and bright areas is inconsistent with the LAI response relationship, which leads to the difficulty in accurately obtaining the LAI of mangrove based on spectral images. Therefore, it is urgent to develop an accurate method for retrieving the LAI of mangrove under the frequent alternation of cloud cover and clear sky to realize the LAI retrieval of mangrove under complex weather.

[0004] In addition, the short and variable sampling window and the difficult-to-walk mudflat caused by the tide make it difficult to obtain the mangrove LAI training sample. Although the LiDAR laser point cloud can extract the LAI, the pulse signal is difficult to penetrate the dense mangrove canopy to reach the ground. Therefore, the conventional use of the radar penetration index and the like as a proxy variable of the gap fraction (P (θ)) based on the Beer-Lambert law to extract the LAI will introduce greater uncertainty and error. At the same time, many existing technologies ignore the change of the extinction coefficient G (θ) caused by the scanning angle and the leaf inclination, and G (θ) is usually set to 0.5 or the shape parameter (χ) of the leaf inclination distribution is set to a fixed value (for example, χ = 1 or χ = 2). This will make the LAI inversion have greater variance, because the average aggregation factor is difficult to separate from the data. The above problems make it difficult for the LiDAR point cloud to quickly and accurately obtain a large number of LAI samples.

[0005] Therefore, how to quickly, stably and accurately invert the mangrove LAI under complex weather has become a problem to be solved. SUMMARY

[0006] Based on the above problems, the embodiments of the present application provide a mangrove LAI hierarchical remote sensing inversion method and related equipment under complex weather, which aims to solve the following two key problems of LAI inversion in cloudy areas: (1) the short and variable sampling window of the mangrove (the influence of the tide) and the difficult-to-walk mudflat make it difficult to obtain the mangrove LAI training sample. Although the LiDAR point cloud data can extract the LAI, the laser pulse is difficult to penetrate the dense mangrove canopy, which leads to low LAI extraction accuracy; (2) the weather in the mangrove area is variable, and the response relationship of the spectral reflectance to the LAI disturbance is inconsistent under the alternation of cloud coverage and clear sky, which leads to low inversion accuracy. The present application not only reveals the response mechanism of the mangrove LAI to the spectral reflectance and the LiDAR feature under complex weather, but also uses the response mechanism to develop a mangrove LAI hierarchical inversion model that can effectively explain the disturbance of the reflectance to the LAI, and then realizes the robust inversion of the mangrove LAI. The complex weather of the present application refers to the weather condition of the alternation of cloud coverage and clear sky.

[0007] In a first aspect, the embodiments of the present application provide a mangrove LAI hierarchical remote sensing inversion method under complex weather, which comprises:

[0008] Obtaining unmanned aerial vehicle LiDAR point cloud data, hyperspectral remote sensing images, mangrove leaf area index and gap fraction data of the mangrove area;

[0009] Pretreating the LiDAR point cloud data and the hyperspectral remote sensing images, and respectively generating a LiDAR feature data set and a spectral feature data set;

[0010] Based on the LiDAR point cloud data and the measured gap fraction, a semi-supervised sample generation method is used to generate mangrove LAI samples;

[0011] Based on the response relationship between the mangrove LAI samples and the hyperspectral remote sensing image reflectivity, the cloud coverage and the clear sky reflectivity are layered to identify the cloud coverage area and the clear sky area;

[0012] Based on the response mechanism of LAI disturbance to the features, the optimal features for inverting the mangrove LAI are identified from the spectral feature dataset and the LiDAR feature dataset;

[0013] Based on the optimal features, a mangrove LAI inversion algorithm with high interpretability is developed, and a mangrove LAI layered inversion model suitable for the cloud coverage area and the clear sky area is constructed, and then the spatial mapping of the mangrove LAI is realized based on the optimal layered inversion model.

[0014] In combination with the first aspect and the above implementation manners, the unmanned aerial vehicle laser radar point cloud data and the hyperspectral remote sensing image are obtained by corresponding sensors carried on the unmanned aerial vehicle; the measured leaf area index and the gap fraction of the mangrove are obtained in the sample area by ground measuring instruments while the unmanned aerial vehicle is flying.

[0015] In combination with the first aspect and the above implementation manners, the laser radar point cloud data and the hyperspectral remote sensing image are preprocessed, and LiDAR feature datasets and spectral feature datasets are generated, including:

[0016] The laser radar point cloud data is resampled, denoised, ground point filtered and normalized, and the preprocessing of the hyperspectral remote sensing image includes radiation correction, tiling and denoising;

[0017] LiDAR features including height, intensity and density are extracted from the laser radar point cloud data; the spectral feature dataset is constructed based on the hyperspectral remote sensing image.

[0018] In combination with the first aspect and the above implementation manners, based on the LiDAR point cloud data and the measured gap fraction, a semi-supervised sample generation method is used to generate mangrove LAI samples, including:

[0019] A priori knowledge model between the LiDAR point cloud intensity and the measured gap fraction is constructed;

[0020] The priori knowledge model is fused with the Beer-Lambert law to form a priori knowledge enhanced Beer-Lambert law, and the mangrove LAI samples are generated.

[0021] In combination with the first aspect and the above implementation manners, based on the response relationship between the mangrove LAI sample and the hyperspectral remote sensing image reflectivity, the cloud coverage and the clear sky reflectivity are layered to identify the cloud coverage area and the clear sky area, including:

[0022] A two-dimensional space coordinate system is constructed based on the mangrove LAI sample and the corresponding hyperspectral reflectivity;

[0023] Based on the two-dimensional space coordinate system, a midpoint straight line equation representing the difference in response relationship under cloud coverage and clear sky conditions is calculated;

[0024] According to the position of the hyperspectral image pixel in the two-dimensional space relative to the midpoint straight line, it is divided into a cloud coverage area or a clear sky area.

[0025] In combination with the first aspect and the above implementation manners, based on the response mechanism of LAI disturbance to the characteristics, the optimal features for inversion are identified from the spectral feature dataset and the LiDAR feature dataset, including:

[0026] The first derivative is used to improve the two-dimensional correlation spectrum to analyze the response mechanism of reflectivity to LAI disturbance, and the optimal spectral feature is identified according to the autocorrelation peak intensity in the synchronous correlation spectrum;

[0027] Random forest regression and SHapley Additive exPlanations (SHAP) algorithm are combined to evaluate the importance of LiDAR features, and the optimal LiDAR features are selected according to the average SHAP value size.

[0028] In combination with the first aspect and the above implementation manners, based on the optimal layered inversion model, the spatial mapping of mangrove LAI is implemented, including:

[0029] An automatic peak intensity weighted regression algorithm is developed to explain the change of reflectivity with LAI disturbance, and a mangrove LAI layered inversion model suitable for cloud coverage area and clear sky area is constructed respectively;

[0030] The coefficient of determination, root mean square error and average relative error are used as accuracy indicators to evaluate the model performance; the optimal layered inversion model is selected for spatial mapping of mangrove LAI.

[0031] In the second aspect, the embodiments of the present application provide a mangrove LAI layered remote sensing inversion device, which comprises:

[0032] An acquisition module is configured to acquire unmanned aerial vehicle laser radar point cloud data, hyperspectral remote sensing image, mangrove leaf area index and gap rate data of a mangrove area;

[0033] The first generation module is used to preprocess lidar point cloud data and hyperspectral remote sensing images, and generate LiDAR feature dataset and spectral feature dataset respectively.

[0034] The second generation module is used to generate mangrove LAI samples based on LiDAR point cloud data and measured gap rate using a semi-supervised sample generation method.

[0035] The first identification module is used to stratify cloud cover and clear sky reflectance based on the response relationship between mangrove LAI samples and hyperspectral remote sensing image reflectance, so as to identify cloud-covered areas and clear sky areas.

[0036] The second identification module is used to identify the optimal features for inverting mangrove LAI from the spectral feature dataset and the LiDAR feature dataset based on the response mechanism of LAI perturbation to features;

[0037] The module is used to develop a highly interpretable mangrove LAI inversion algorithm based on optimal features, and to build hierarchical inversion models of mangrove LAI applicable to cloud-covered areas and clear-sky areas respectively. Then, spatial mapping of mangrove LAI is realized based on the optimal hierarchical inversion model.

[0038] Thirdly, embodiments of this application provide a computer storage medium storing multiple instructions adapted for loading by a processor and executing the steps of the above-described method.

[0039] Fourthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being adapted to be loaded by the processor and to execute the steps of the method described above.

[0040] The beneficial effects of the technical solutions provided in some embodiments of this application include at least the following: This invention utilizes the Beer-Lambert Law (PEBL) enhanced by prior knowledge to quickly obtain reliable mangrove LAI samples, achieves stratification of cloud cover and clear sky reflectance based on the response relationship between spectral reflectance and LAI, and then reveals the response mechanism of spectral reflectance to mangrove LAI perturbation based on the first derivative-enhanced two-dimensional correlation spectroscopy (FOD-2DCOS), thereby identifying the optimal spectral features and combining them with the optimal response features of LiDAR to achieve stratified inversion of mangrove LAI. Thus, it overcomes the technical difficulty of obtaining mangrove LAI samples and develops a novel mangrove LAI stratified inversion algorithm, integrating the response mechanism of spectral reflectance to LAI perturbation into the stratified inversion model, thereby achieving rapid and accurate inversion of mangrove LAI under complex weather conditions. Attached Figure Description

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings described in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0042] Figure 1 A system architecture diagram of a mangrove LAI hierarchical remote sensing inversion system under complex weather provided by the embodiments of the present application;

[0043] Figure 2 A flowchart of a mangrove LAI hierarchical remote sensing inversion method under complex weather provided by the embodiments of the present application;

[0044] Figure 3 A performance comparison diagram of a sample of mangrove LAI generated by an exemplary priori knowledge model and PEBL provided by the embodiments of the present application;

[0045] Figure 4 A precision comparison diagram of an exemplary hierarchical and non-hierarchical mangrove LAI model provided by the embodiments of the present application;

[0046] Figure 5 A precision comparison diagram of different hierarchical mangrove LAI models provided by the embodiments of the present application;

[0047] Figure 6 A mangrove LAI spatial distribution diagram drawn by an exemplary different hierarchical inversion model provided by the embodiments of the present application;

[0048] Figure 7 A structure block diagram of a mangrove LAI hierarchical remote sensing inversion device under complex weather provided by the embodiments of the present application;

[0049] Figure 8 A structure diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0050] In order to make the features and advantages of the present application more obvious and easy to understand, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0051] The following description refers to the accompanying drawings. Unless otherwise noted, reference to a term in

[0052] In the description of the present application, it should be understood that the terms "first", "second" and the like are used to describe and distinguish one element from another, but not necessarily used to describe or imply these elements' relative importance. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances. In addition, in the description of the present application, "a plurality of" means two or more, unless otherwise specified. The "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent: A exists alone, A and B exist together, and B exists alone. The character " / " generally represents that the associated objects before and after are in an "or" relationship.

[0053] As described above, the mangrove leaf area index (LAI) as a key parameter representing its canopy structure is closely related to the photosynthesis, transpiration and material and energy exchange of vegetation, and is the core index for monitoring the health status and growth dynamics of mangrove ecosystems. At present, large-scale mangrove LAI monitoring mainly relies on remote sensing technology. However, the spatial resolution of the existing mainstream medium and low spatial resolution LAI global products is usually hundreds of meters to several kilometers, which is difficult to accurately capture the details of the narrow-distributed and complex-structured mangrove canopy, and cannot meet the needs of fine management and protection of mangroves. Therefore, using unmanned aerial vehicle platform to carry hyperspectral and laser radar sensors for high-precision LAI inversion has become an important research direction.

[0054] However, there are still two key technical problems in realizing high-precision inversion of mangrove LAI. The first problem is the lack of training samples. Mangroves grow in the intertidal zone and are affected by tides. The sampling window is short and it is difficult to enter the beach environment, resulting in difficulties and high cost in obtaining ground measured LAI samples. Although airborne LiDAR technology provides the possibility for large-scale extraction of LAI, the traditional inversion method is limited in its laser pulse penetration ability when applied to dense mangrove canopy, and often ignores the actual changes of extinction coefficient, introducing significant errors and uncertainties, making it difficult to quickly and reliably generate a large number of high-quality LAI samples for model training.

[0055] The second difficulty lies in the inconsistency of spectral response relationship under complex weather conditions. The weather in mangrove forest regions is changeable, with frequent alternation of cloud cover and clear sky. Remote sensing images acquired in this environment contain both "dark pixels" affected by cloud shadows and "bright pixels" under sufficient sunlight, and the spectral reflectance of these two types of pixels has significant differences in response mechanism and sensitivity to LAI changes. If a uniform model is used for inversion, this inconsistency will directly lead to a significant decrease in LAI estimation accuracy, and traditional methods are difficult to overcome the disturbance caused by cloudy weather.

[0056] Therefore, the present application provides a mangrove LAI hierarchical remote sensing inversion method, device, computer storage medium and electronic equipment. By fusing LiDAR point cloud intensity and measured gap rate prior knowledge, the Beer-Lambert law is improved, and the PEBL model under the semi-supervised learning framework is constructed, so as to efficiently and reliably generate a large number of LAI training samples, thereby breaking through the problem of sample acquisition. Further, by analyzing the response relationship between the generated samples and the spectral reflectance, the reflectance hierarchical method is innovatively proposed, which effectively separates the cloud coverage and clear sky areas. On this basis, the FOD-2DCOS and SHAP and other explainability analysis techniques are used to deeply reveal the response mechanism of spectral and LiDAR features to LAI disturbance under different conditions, and the hierarchical inversion model is developed accordingly. Finally, the present application realizes the rapid, stable and high-precision inversion of mangrove LAI under complex weather conditions, and provides effective technical support for the accurate monitoring and protection of mangrove forests.

[0057] Please refer to Figure 1 , Figure 1 The present application provides a mangrove LAI hierarchical remote sensing inversion method, device, computer storage medium and electronic equipment. By fusing LiDAR point cloud intensity and measured gap rate prior knowledge, the Beer-Lambert law is improved, and the PEBL model under the semi-supervised learning framework is constructed, so as to efficiently and reliably generate a large number of LAI training samples, thereby breaking through the problem of sample acquisition. Further, by analyzing the response relationship between the generated samples and the spectral reflectance, the reflectance hierarchical method is innovatively proposed, which effectively separates the cloud coverage and clear sky areas. On this basis, the FOD-2DCOS and SHAP and other explainability analysis techniques are used to deeply reveal the response mechanism of spectral and LiDAR features to LAI disturbance under different conditions, and the hierarchical inversion model is developed accordingly. Finally, the present application realizes the rapid, stable and high-precision inversion of mangrove LAI under complex weather conditions, and provides effective technical support for the accurate monitoring and protection of mangrove forests.

[0058] As Figure 1 shown, the system architecture can include a terminal 101, a network 102 and a server 103. The network 102 is used to provide a communication link medium between the terminal 101 and the server 103. The network 102 can include various types of wired communication links or wireless communication links, for example: the wired communication links include optical fiber, twisted pair or coaxial cable, and the wireless communication links include Bluetooth communication link, Wireless-Fidelity (Wi-Fi) communication link or microwave communication link, etc.

[0059] The terminal 101 can interact with the server 103 through the network 102 to receive a message from the server 103 or send a message to the server 103, or the terminal 101 can interact with the server 103 through the network 102 to receive a message or data sent by other users to the server 103. The terminal 101 can be hardware or software. When the terminal 101 is hardware, it can be various electronic devices, including but not limited to smart watches, smart phones, tablet computers, laptop computers, desktop computers, and the like. When the terminal 101 is software, it can be installed in the above-mentioned electronic devices, and can be implemented as multiple software or software modules (for example, to provide distributed services), or as a single software or software module, which is not specifically limited here.

[0060] In the embodiments of the present application, the terminal 101 can obtain unmanned aerial vehicle LiDAR point cloud data, hyperspectral remote sensing images, mangrove leaf area index and gap ratio data of a mangrove area; the LiDAR point cloud data and the hyperspectral remote sensing images are preprocessed, and LiDAR feature data sets and spectral feature data sets are generated respectively; based on the LiDAR point cloud data and the measured gap ratio, a semi-supervised sample generation method is used to generate mangrove LAI samples; based on the response relationship between the mangrove LAI samples and the hyperspectral remote sensing image reflectivity, the cloud coverage and the clear sky reflectivity are layered to identify the cloud coverage area and the clear sky area; based on the response mechanism of LAI disturbance to the features, the optimal features for inverting the mangrove LAI are identified from the spectral feature data set and the LiDAR feature data set; based on the optimal features, a mangrove LAI inversion algorithm with high interpretability is developed, and mangrove LAI layered inversion models suitable for cloud coverage areas and clear sky areas are constructed respectively, and then spatial mapping of the mangrove LAI is realized based on the layered inversion models.

[0061] The server 103 can be a service server providing various services. It should be noted that the server 103 can be hardware or software. When the server 103 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server 103 is software, it can be implemented as multiple software or software modules (for example, to provide distributed services), or as a single software or software module, which is not specifically limited here.

[0062] Alternatively, the system architecture can also not include the server 103, in other words, the server 103 can be an optional device in the embodiments of the present specification, that is, the method provided in the embodiments of the present specification can be applied to a system structure including only the terminal 101, and the embodiments of the present application do not limit this.

[0063] It should be understood that, Figure 1The number of terminals, networks and servers in the system is only illustrative, and can be any number of terminals, networks and servers according to the needs of implementation.

[0064] Please refer to Figure 2 , Figure 2 A flowchart of a mangrove LAI hierarchical remote sensing inversion method under complex weather provided by an embodiment of the present application. The execution subject of the embodiment of the present application can be an electronic device for executing mangrove LAI hierarchical remote sensing inversion, or a processor in the electronic device for executing the mangrove LAI hierarchical remote sensing inversion method, or a mangrove LAI hierarchical remote sensing inversion service in the electronic device for executing the mangrove LAI hierarchical remote sensing inversion method. For the convenience of description, the specific execution process of the mangrove LAI hierarchical remote sensing inversion method will be introduced below by taking the execution subject as a processor in the electronic device.

[0065] As shown in Figure 2 , the mangrove LAI hierarchical remote sensing inversion method under complex weather can at least include:

[0066] S201, acquiring unmanned aerial vehicle laser radar point cloud data, hyperspectral remote sensing image, measured leaf area index and gap ratio data of the mangrove forest area.

[0067] Specifically, the unique ecological environment of mangrove forest brings great challenges to parameter inversion. Any single data source is difficult to independently cope with the uncertainty brought by complex canopy structure and variable environment. Unmanned aerial vehicle laser radar point cloud can penetrate the canopy gap and accurately capture the vertical three-dimensional structure information of trees, providing key geometric features for estimating canopy height, porosity and leaf area index. Hyperspectral remote sensing image can continuously record the reflection characteristics of ground objects in hundreds of narrow spectral bands, containing rich vegetation biochemical parameter information, which provides the possibility of LAI inversion from the spectral dimension. However, whether it is laser radar or hyperspectral data, the inversion model must rely on real ground measurement values for training and verification. Therefore, the synchronous acquisition of measured leaf area index and gap ratio data of mangrove forest becomes the "bridge" connecting remote sensing observation signals with real vegetation parameters, and is a decisive link to ensure the accuracy and reliability of the model.

[0068] In this embodiment, DJI Matrice M350 RTK and other unmanned aerial vehicle (UAV) platforms are used to carry LIAIRX3-H light laser radar system, AU20 laser radar scanning system and Cubert-X20P airborne hyperspectral imager respectively, and flight operations are carried out above the typical sample area of the mangrove forest protection area to ensure the acquisition of high-precision laser point cloud data and hyperspectral remote sensing image. The above devices are only examples, and the embodiment of the present application does not specially limit this.

[0069] The flight planning needs to ensure a certain degree of overlap and prepare for subsequent data registration and fusion. Therefore, for the UAV carrying the LiDAR sensor, the point cloud data is collected at a flight height of 100 m with an overlap rate of 50%; for the UAV carrying the hyperspectral sensor, the hyperspectral image is collected on the flight route planned in advance at a flight height of 200 m, and the heading and lateral overlap rates are 70% and 80%, respectively. Before the UAV carrying the hyperspectral sensor takes off, the reflectivity image of the calibrated calibration plate needs to be collected for subsequent correction of the reflectivity of each waveband of the hyperspectral sensor. At the same time, the ground measurement work is carried out synchronously, and the LAI-2200C plant canopy analyzer is used to obtain the canopy transmittance data in different directions within the sample plot of the UAV flight according to the standard measurement procedures, and then the measured leaf area index and gap fraction of the sample point are calculated. This synchronous sampling strategy maximizes the influence of the change of vegetation state and the change of light conditions on the consistency of the data caused by the time difference, and ensures the reliable correspondence between the remote sensing data and the ground true value.

[0070] S202, pre-process the LiDAR point cloud data and the hyperspectral remote sensing image, and generate a LiDAR feature data set and a spectral feature data set, respectively.

[0071] Specifically, the pre-processing of the UAV LiDAR raw data includes: generating high-precision true color point cloud data from the UAV LiDAR raw data, and performing resampling, point cloud denoising, ground point filtering and point cloud normalization processing, finally generating a digital elevation model (DEM), a digital surface model (DSM), a canopy height model (CHM) and derived point cloud features, including intensity features, height features, point cloud density features, coverage, gap fraction, normal vector. These features are used to construct a high-dimensional LiDAR point cloud feature data set.

[0072] Further, the UAV hyperspectral image is pre-processed: the Cubert Utils Touch 2.8.1 software is used for parameter adjustment and radiation correction. Then, the Agisoft PhotoScan software is used to mosaic the hyperspectral image. Finally, in order to remove the influence of noise on LAI inversion, only the spectral bands of 406-910 nm (spectral resolution of 4 nm) are used to construct a high-dimensional spectral feature data set.

[0073] S203, generating mangrove LAI samples based on the LiDAR point cloud data and the measured gap fraction using a semi-supervised sample generation method.

[0074] Specifically, traditional machine learning models heavily rely on a large amount of high-quality labeled data (i.e., measured LAI) for training. However, in the intertidal zone environment of mangrove forests, it is extremely difficult and costly to conduct large-scale ground measurements due to the difficulty of access to the beach and the short tidal window, and the number of samples obtained is far from enough to support a robust inversion model. Although airborne LiDAR technology provides the potential for large-scale LAI estimation, the traditional Beer-Lambert law method has inherent defects when applied to dense mangrove forests: it is difficult for laser pulses to completely penetrate the dense canopy, resulting in a significant deviation in the direct estimation of canopy bottom gap fraction. If the ground laser point ratio is simply used as a proxy for gap fraction, it will introduce a large error, making the directly estimated LAI value unreliable and unable to be directly used as a reliable sample.

[0075] In this embodiment, a statistical prior knowledge model is established using the gap fraction P(θ) value obtained by small-scale measurement and the LiDAR point cloud intensity information at the corresponding position to correct the influence of the difficulty of laser beams to penetrate the mangrove canopy, and then the prior knowledge model and the Beer-Lambert law are fused to generate mangrove LAI samples, i.e., prior knowledge enhanced Beer-Lambert law (PEBL). First, the prior knowledge model is constructed by the statistical relationship between the reflectance intensity value (0-255) of the LiDAR point cloud and the measured gap fraction to improve the accuracy of the Beer-Lambert law in extracting LAI. Since the range of P(θ) is 0-1, the intensity information is negatively correlated with the gap fraction, so an exponential function is more suitable for constructing the prior knowledge model of P(θ). Finally, the prior knowledge model is fitted by the least squares algorithm. The specific implementation formula is as follows:

[0076] f(x)=ae bx (1)

[0077] where f(x) is the prior knowledge model of P(θ); x is the intensity information of the LiDAR point cloud, and a and b are fitting coefficients, which are obtained based on the least squares method using the measured P(θ) and the intensity information of the LiDAR point cloud.

[0078] Further, the ratio of the ground points to the total point intensity or number at a certain height threshold is currently used as a proxy for gap fraction, and then LAI is obtained based on the Beer-Lambert law. However, this is not suitable for dense mangrove forests, because using these ratios as proxies for gap fraction to extract mangrove LAI will introduce large errors and uncertainties. Therefore, in this embodiment, the maximum intensity of the point cloud and the prior knowledge model constructed by P(θ) are fused with the Beer-Lambert law, i.e., using f(x)=ae bxInstead of P(θ) in formula 2, mangrove LAI samples are generated. The specific process is as follows:

[0079]

[0080] Where θ is the angle between the light incident direction and the vertical direction; G(θ) and P(θ) are the extinction coefficient and gap rate, respectively.

[0081] In order to generate mangrove LAI samples, it is necessary to first calculate the leaf inclination angle. The normal vector of each point cloud is calculated, and then the average inclination angle of all point clouds in the grid is obtained as the average leaf inclination angle. The specific formula is as follows:

[0082]

[0083] Where ALA is the average leaf inclination angle; N and N z are the unit normal vector and the normal vector in the z direction of a certain point cloud, respectively; n is the number of point clouds.

[0084] Secondly, the extinction coefficient is calculated. The shape parameter of the leaf inclination angle distribution is calculated by the average leaf inclination angle, and then G(θ) is calculated using the approximate ellipsoid model. The specific formula is as follows:

[0085]

[0086]

[0087] Where χ is the shape parameter of the leaf inclination angle distribution, representing the length ratio of the horizontal semi-axis and the vertical semi-axis; ALA is the average leaf inclination angle (unit: degree).

[0088] Finally, the mangrove LAI samples are generated. Substitute G(θ) (formula 5) and the prior knowledge model (formula 1) into the Beer-Lambert law to generate the mangrove LAI samples. The specific formula is as follows:

[0089]

[0090] Where f(x) is the prior knowledge model of P(θ), f(x) = 0.995 x e -0.022x ; x is the maximum intensity of each grid LiDAR point cloud.

[0091] S204, based on the response relationship between the mangrove LAI samples and the reflectivity of the hyperspectral remote sensing image, the cloud coverage and the clear sky reflectivity are layered to identify the cloud coverage area and the clear sky area.

[0092] Specifically, the mathematical relationship between spectral reflectance and LAI will be distorted due to the dramatic change of light conditions, which makes a unified inversion model unable to accurately apply to both dark pixels under cloud cover and bright pixels under clear sky. Therefore, to achieve high-precision inversion, the mixed signals need to be separated, i.e., to provide the model with two sets of rules of "how LAI affects reflectance under cloudy conditions" and "how LAI affects reflectance under sunny conditions" respectively.

[0093] In this embodiment, a reflectance stratification method is proposed to separate dark and bright reflectance in spectral images and identify cloud cover and clear sky areas. The specific steps are as follows:

[0094] First, construct a two-dimensional coordinate system with the mangrove LAI samples generated by PEBL and the corresponding hyperspectral reflectance, and extract the top points (P TD and P TB ) and bottom points (P BD and P BB ) of the two response relationships of spectral reflectance and LAI samples in the two-dimensional coordinate system. The specific formula is as follows:

[0095] P TD = (x TD , y TD ), P TB = (x TB , y TB ), P BD = (x BD , y BD ), P BB = (x BB , y BB ) (7)

[0096] Where x and y represent leaf area index and spectral reflectance respectively; P TD and P TB are the top point coordinates of dark and bright pixels in the two-dimensional coordinate system of the response relationship; P BD and P BB are the bottom point coordinates of the spectral reflectance and LAI response relationship in dark and bright pixels.

[0097] Secondly, based on the two-dimensional coordinate system, the midpoint coordinates (P T ) of the top points and the midpoint coordinates (P B ) of the bottom points in the two response relationships are calculated respectively, and the straight line equation of the two midpoints is solved to represent the difference between the response relationships under cloud cover and clear sky conditions. The specific formula is as follows:

[0098]

[0099] Where x Tand y B LAI sample value of the midpoint coordinates of the vertex and the bottom point respectively; y T and y B Spectral reflectance of the midpoint coordinates of the vertex and the bottom point respectively.

[0100] Finally, according to the position of the hyperspectral image pixel in the two-dimensional space relative to the midpoint straight line, it is divided into cloud coverage area or clear sky area. Specifically, the coordinates corresponding to the pixel point are brought into the midpoint straight line equation (formula 9) to realize the dark and bright reflectivity layering and identify the cloud coverage and clear sky area. The pixel point above the midpoint straight line is the bright pixel reflectivity and belongs to the mangrove area of clear sky. The pixel point below the midpoint straight line is the dark pixel reflectivity and belongs to the cloud-shielded mangrove area.

[0101]

[0102] wherein x pixel is the LAI extracted based on PEBL using LiDAR point cloud; y is the spectral reflectance varying with LAI.

[0103] S205, based on the response mechanism of the feature to the LAI disturbance, identifying the optimal feature for inverting the mangrove LAI from the spectral feature data set and the LiDAR feature data set.

[0104] Specifically, in order to identify those features that are most sensitive to the change of mangrove LAI, most stable in response, and have clear physical interpretation from a large number of spectral features and LiDAR features, so as to lay the foundation for building a high-precision and high-robustness inversion model.

[0105] In this embodiment, FOD-2DCOS, random forest regression and SHapley Additive exPlanations (SHAP) algorithm are used to clarify the response mechanism of the feature to the LAI disturbance and then identify the optimal feature. Among them, FOD-2DCOS is a two-dimensional correlation spectral technology (2DCOS) improved by first-order derivative (FOD) to enhance the response of vegetation dynamic spectrum to disturbed LAI to analyze the response mechanism of cloud coverage and clear sky spectral reflectance to disturbed LAI, including response intensity, response direction and response order, and then accurately identify the optimal spectral feature for inverting LAI based on the autocorrelation peak of synchronous FOD-2DCOS spectrum. Then, the above FOD-2DCOS technology is used to analyze its response mechanism to LAI disturbance, and the autocorrelation peak intensity in the synchronous correlation spectrum is used to identify the optimal spectral feature. The specific steps are as follows:

[0106] Firstly, the FOD spectrum of hyperspectral image is obtained, the change of LAI extracted by PEBL is regarded as the external disturbance of mangrove canopy system, and the direction of continuous FOD spectrum reflectance with the growth of LAI is sorted, and the corresponding spectrum is regarded as a dynamic spectrum. The dynamic spectrum can be calculated by the following formula:

[0107]

[0108] wherein, and are the dynamic first derivative spectrum and the reference spectrum of the first derivative with the disturbance variable t (LAI min ≤t≤LAI max ) change; is usually set as the average spectrum; S(b k ,t i ) is the first derivative spectrum; b k is the spectral variable (k = 1, 2, …, n).

[0109] Secondly, the intensity of FOD-2DCOS is calculated by using the dynamic spectrum, and the intensity is presented as a complex number, the real part and the imaginary part are synchronous FOD-2DCOS correlation spectrum intensity and asynchronous FOD-2DCOS correlation spectrum intensity respectively, see the following calculation formula:

[0110]

[0111]

[0112] wherein, X(b1, b2) is the intensity of FOD-2DCOS; φ(b1, b2) and φ(b1, b2) are synchronous FOD-2DCOS correlation spectrum intensity and asynchronous FOD-2DCOS correlation spectrum intensity respectively; N ij is the Hilbert-Noda transformation matrix

[0113] Further, the random forest regression and SHapley Additive exPlanations (SHAP) algorithm are combined to evaluate the feature importance, and the optimal LiDAR feature is selected according to the average SHAP value size. In this embodiment, the random forest regression algorithm and the SHapley Additive exPlanations algorithm are combined to clarify the response relationship between the LiDAR point cloud feature and the LAI, including the contribution and influence of the LiDAR point cloud feature to the LAI estimation. Finally, all the SHAP values are averaged, and the top 6 features in the average |SHAP| value sorting are selected as the optimal LiDAR features for estimating the LAI.

[0114] According to the reflectance stratification results, six mangrove LAI inversion model training schemes were designed based on spectral features and LiDAR features: (1) The sample data stratified by reflectance were input into FOD-2DCOS, and the spectral bands that were significantly responsive to LAI disturbance were selected from 406 to 910 nm to construct stratified inversion models for cloud cover and clear sky reflectance, respectively; (2) The sample data before stratification by reflectance were input into FOD-2DCOS, and the spectral bands that were significantly responsive to LAI disturbance were selected from 406 to 910 nm to construct non-stratified inversion models; (3) The sample data stratified by reflectance were based on the autocorrelation peaks of FOD-2DCOS to select the optimal spectral features under cloud cover and clear sky for constructing stratified inversion models; (4) The sample data before stratification by reflectance were based on the autocorrelation peaks of FOD-2DCOS to select the optimal spectral features for constructing non-stratified inversion models; (5) On the basis of the spectral features selected in scheme 3, the optimal LiDAR features under cloud cover and clear sky based on SHAP were added, respectively; (6) On the basis of the optimal spectral features selected in scheme 4, the optimal LiDAR features based on SHAP were added.

[0115] The main purpose of the above six schemes is to verify the effectiveness of the optimal features identified by FOD-2DCOS for mangrove LAI inversion, the feasibility of the LAI inversion in this embodiment, and the effectiveness of LiDAR features for improving the accuracy of mangrove LAI inversion.

[0116] S206, based on the optimal features, develop a mangrove LAI inversion algorithm with high interpretability, and construct mangrove LAI stratified inversion models suitable for cloud cover and clear sky areas, respectively, and then realize the spatial mapping of mangrove LAI based on the optimal stratified inversion model.

[0117] Specifically, two novel automatic peak weighted regression algorithms that can effectively explain the changes of reflectance with LAI disturbance were developed based on the automatic peak intensity of FOD-2DCOS, and were embedded into random forest and partial least squares regression algorithms to construct mangrove LAI stratified inversion models.

[0118] Based on the optimal feature identification module based on the LAI disturbance response mechanism, the sensitivity of reflectance to LAI under cloud cover and clear sky was evaluated, and two novel mangrove LAI stratified inversion algorithms were developed based on the optimal spectral features that were sensitive to LAI disturbance and had significant response, namely, automatic peak intensity weighted regression algorithm based on reflectance (AIWRA) and automatic peak intensity weighted regression algorithm based on model (AIWMA). This algorithm uses the intensity value of the automatic peak in FOD-2DCOS to gradually adjust the weight of each optimal band reflectance or model, so as to more effectively explain the changes of reflectance with LAI disturbance. See the following formula:

[0119]

[0120] where LRM ij (λ j ) is the linear regression model (y = ax + b) constructed by spectral reflectance at wavelength λ j , i = 1 and i = 2 represent dark and bright pixels, respectively; R rs (λ j ) is the spectral reflectance at wavelength λ j , m represents the number of optimal spectral features; w ij is the weight factor based on the automatic peak intensity adjustment of synchronous FOD-2DCOS correlation spectrum; φ(λ j , λ j ) is the automatic peak intensity of the synchronous FOD-2DCOS correlation spectrum.

[0121] Further, the embodiment further embeds the random forest and partial least squares regression algorithm in the module, integrates the optimal spectrum and LiDAR features to construct the hierarchical mangrove LAI inversion model. Then, the coefficient of determination, root mean square error and average relative error are used as the accuracy index to evaluate the model performance; the optimal hierarchical inversion model is selected to map the spatial distribution of mangrove LAI. Specifically, the coefficient of determination (R 2 ), root mean square error (RMSE) and mean relative error (MRE) are used to evaluate the performance of the hierarchical and non-hierarchical mangrove LAI inversion model, respectively, and the optimal mangrove LAI hierarchical inversion model is selected to draw the spatial distribution of mangrove LAI. Please refer to Figures 3-5 , Figure 3 is a kind of example priori knowledge model provided in the embodiment of the application, PEBL generates the performance comparison diagram of mangrove LAI sample, Figure 4 is a kind of example hierarchical and non-hierarchical mangrove LAI model accuracy comparison diagram provided in the embodiment of the application, Figure 5 is a kind of example hierarchical mangrove LAI model accuracy comparison diagram provided in the embodiment of the application, finally, according to the content in the figure, the model can be selected to carry out inversion prediction and carry out the spatial mapping of mangrove LAI, it needs to be explained that it can be the combination prediction of multiple models, for example, random forest regression and AIWMA hierarchical inversion model complete the spatial distribution inversion of mangrove LAI. The specific calculation method of accuracy evaluation index is shown in the following formula:

[0122]

[0123] where y i , and The measured LAI, the inverted LAI and the average value of the measured LAI, respectively; n is the sample number of the measured LAI.

[0124] Please refer to Figure 6 , Figure 6 is a schematic diagram of the LAI spatial distribution of mangrove provided by an embodiment of the present application. As shown in Figure 6 , the spatial distribution details of the LAI of the mangrove in the study area are obtained by the method of the present application, and the results clearly show the spatial heterogeneity of mangroves with different growth potentials (corresponding to different LAI values), verifying the effectiveness of the method in actual mapping application. Among them Figure 6 (a) is the LAI spatial distribution map of the mangrove inverted by the optimal model combination, and the growth potential of different species is based on the results.

[0125] The embodiment of the present application provides a mangrove LAI hierarchical remote sensing inversion method under complex weather. The prior knowledge enhanced Beer-Lambert law (PEBL) proposed in the present application is used to quickly obtain reliable mangrove LAI samples, the response relationship between spectral reflectance and LAI is used to realize cloud coverage and clear sky reflectance layering, and then the response mechanism of spectral reflectance to mangrove LAI disturbance is revealed based on FOD-2DCOS, and then the optimal spectral feature is identified to develop two weighted regression algorithms that can effectively explain the change of reflectance with LAI disturbance, and the optimal response feature of LiDAR is combined to realize the hierarchical inversion of mangrove LAI. Thus, the technical difficulty of obtaining mangrove LAI samples is broken through, and a novel mangrove LAI hierarchical inversion algorithm is developed, the response mechanism of spectral reflectance to LAI disturbance is integrated into the hierarchical inversion model, so that the mangrove LAI inversion under complex weather is quickly and accurately realized.

[0126] Please refer to Figure 7 , Figure 7 is a structural block diagram of a mangrove LAI hierarchical remote sensing inversion device provided by an embodiment of the present application. As shown in Figure 7 , the mangrove LAI hierarchical remote sensing inversion device comprises an acquisition module 710, a first generation module 720, a second generation module 730, a first identification module 740, a second identification module 750 and a construction module 760. Wherein:

[0127] The acquisition module 710 is configured to acquire unmanned aerial vehicle laser radar point cloud data, hyperspectral remote sensing images, mangrove leaf area index and gap rate data of a mangrove area;

[0128] The first generation module 720 is configured to pre-process the laser radar point cloud data and the hyperspectral remote sensing images, and generate a LiDAR feature data set and a spectral feature data set, respectively;

[0129] The second generation module 730 is configured to generate mangrove LAI samples based on LiDAR point cloud data and measured gap fraction by using a semi-supervised sample generation method.

[0130] The first identification module 740 is configured to perform hierarchical division on cloud coverage and clear sky reflectivity based on a response relationship between the mangrove LAI samples and hyperspectral remote sensing image reflectivity, so as to identify cloud coverage areas and clear sky areas.

[0131] The second identification module 750 is configured to identify optimal features for mangrove LAI inversion from a spectral feature dataset and a LiDAR feature dataset based on a response mechanism of LAI disturbance on the features.

[0132] The construction module 760 is configured to develop a mangrove LAI inversion algorithm with high interpretability based on the optimal features, and construct a mangrove LAI hierarchical inversion model suitable for cloud coverage areas and clear sky areas respectively, and then realize spatial mapping of the mangrove LAI based on the optimal hierarchical inversion model.

[0133] In some possible embodiments, the unmanned aerial vehicle LiDAR point cloud data and the hyperspectral remote sensing image are obtained by corresponding sensors carried on the unmanned aerial vehicle; and the measured leaf area index and gap fraction of the mangrove are obtained in a sample area by a ground measuring instrument while the unmanned aerial vehicle is flying.

[0134] In some possible embodiments, the first generation module 720 includes:

[0135] The preprocessing unit is configured to perform resampling, denoising, ground point filtering and normalization processing on the LiDAR point cloud data, and perform radiation correction, tiling and denoising processing on the hyperspectral remote sensing image.

[0136] The first construction unit is configured to extract LiDAR features including height, intensity and density from the LiDAR point cloud data, and construct a spectral feature dataset based on the hyperspectral remote sensing image.

[0137] In some possible embodiments, the second generation module 730 includes:

[0138] The second construction unit is configured to construct a prior knowledge model between LiDAR point cloud intensity and measured gap fraction.

[0139] The generation unit is configured to fuse the prior knowledge model and the Beer-Lambert law to form a prior knowledge enhanced Beer-Lambert law, and generate the mangrove LAI samples.

[0140] In some possible embodiments, the first identification module 740 includes:

[0141] A third construction unit is configured to construct a two-dimensional spatial coordinate system for the mangrove LAI sample and the corresponding hyperspectral reflectance;

[0142] A calculation unit is configured to calculate a mid-point straight line equation representing the difference in response relationship between cloud cover and clear sky conditions based on the two-dimensional spatial coordinate system;

[0143] A division unit is configured to divide the hyperspectral image pixels into cloud cover areas or clear sky areas according to their positions relative to the mid-point straight line in the two-dimensional spatial coordinate system.

[0144] In some possible embodiments, the second identification module 750 includes:

[0145] A first screening unit is configured to use a two-dimensional correlation spectrum based on a first-order derivative to improve the response mechanism of the LAI disturbance, and to identify optimal spectral features according to the autocorrelation peak intensity in the synchronous correlation spectrum;

[0146] A second screening unit is configured to use a random forest regression and a SHapley Additive exPlanations (SHAP) algorithm to evaluate feature importance, and to screen optimal LiDAR features according to the average SHAP value size.

[0147] In some possible embodiments, the construction module 760 includes:

[0148] A selection unit is configured to use a coefficient of determination, a root mean square error, and an average relative error as accuracy indicators to evaluate model performance, and to select an optimal hierarchical inversion model for spatial mapping of the mangrove LAI.

[0149] It should be noted that the mangrove LAI hierarchical remote sensing inversion device provided in the above embodiments is used to execute the mangrove LAI hierarchical remote sensing inversion method, and only the division of the above functional modules is used as an example for illustration. In actual applications, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the mangrove LAI hierarchical remote sensing inversion device and the mangrove LAI hierarchical remote sensing inversion method provided in the above embodiments belong to the same concept, and the implementation process is described in detail in the method embodiments. Therefore, it is not repeated here.

[0150] The serial numbers of the above embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0151] Please refer to Figure 8 , Figure 8 is a structural schematic diagram of an electronic device provided by the embodiments of the present application. As shown in Figure 8As shown, the electronic device 800 can include at least one processor 801, at least one network interface 804, a user interface 803, a memory 805, at least one communication bus 802.

[0152] The communication bus 802 is configured to realize the connection communication between the components.

[0153] The user interface 803 can include a display, a camera, and optionally includes a standard wired interface, a wireless interface.

[0154] The network interface 804 can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface).

[0155] The processor 801 can include one or more processing cores. The processor 801 connects various parts in the entire electronic device 800 through various interfaces and lines, executes various functions of the electronic device 800 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 805, and calling data stored in the memory 805. Optionally, the processor 801 can be implemented in at least one of a hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 801 can be integrated with a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU is mainly used to process operating systems, user interfaces, and application programs; the GPU is used to render and draw the content to be displayed on the display; and the modem is used to process wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 801, but can be realized by a separate chip.

[0156] The memory 805 can include a random access memory (RAM) and can also include a read-only memory (ROM). Optionally, the memory 805 includes a non-transitory computer-readable storage medium. The memory 805 can be used to store instructions, programs, codes, code sets, or instruction sets. The memory 805 can include a program storage area and a data storage area, where the program storage area can store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area can store data involved in the above-mentioned various method embodiments, etc. The memory 805 can also be at least one storage device located away from the aforementioned processor 801. As shown in Figure 8 the memory 805 as a computer storage medium can include an operating system, a network communication module, a user interface module, and a mangrove LAI hierarchical remote sensing inversion application program.

[0157] In the electronic device 800 shown in Figure 8 the user interface 803 is mainly used to provide an interface for user input and obtain user input data; and the processor 801 can be used to call the mangrove LAI hierarchical remote sensing inversion application program stored in the memory 805 and specifically perform the following operations:

[0158] obtain unmanned aerial vehicle laser radar point cloud data, hyperspectral remote sensing images, mangrove leaf area index, and gap ratio data of a mangrove area;

[0159] preprocess the laser radar point cloud data and the hyperspectral remote sensing images, and generate a LiDAR feature data set and a spectral feature data set, respectively;

[0160] based on the LiDAR point cloud data and the measured gap ratio, generate mangrove LAI samples using a semi-supervised sample generation method;

[0161] based on the response relationship between the mangrove LAI samples and the hyperspectral remote sensing image reflectivity, hierarchically classify cloud coverage and clear sky reflectivity to identify cloud coverage areas and clear sky areas;

[0162] based on the response mechanism of LAI disturbance to features, identify optimal features for mangrove LAI inversion from the spectral feature data set and the LiDAR feature data set;

[0163] Based on the optimal features, a mangrove LAI inversion algorithm with high interpretability is used to construct a mangrove LAI hierarchical inversion model suitable for cloud-covered areas and clear-sky areas, and the spatial mapping of mangrove LAI is realized based on the optimal hierarchical inversion model.

[0164] In some possible embodiments, the unmanned aerial vehicle LiDAR point cloud data and the hyperspectral remote sensing image are obtained by corresponding sensors carried on the unmanned aerial vehicle; the mangrove leaf area index and gap fraction are obtained by ground measuring instruments in the sample area while the unmanned aerial vehicle is flying.

[0165] In some possible embodiments, when the processor 801 performs preprocessing on the LiDAR point cloud data and the hyperspectral remote sensing image, and generates a LiDAR feature data set and a spectral feature data set respectively, it is specifically used to perform:

[0166] Resampling, denoising, ground point filtering and normalization processing are performed on the LiDAR point cloud data, and preprocessing of the hyperspectral remote sensing image includes radiation correction, tiling and denoising processing; LiDAR features including height, intensity and density are extracted from the LiDAR point cloud data; and a spectral feature data set is constructed based on the hyperspectral remote sensing image.

[0167] In some possible embodiments, when the processor 801 generates mangrove LAI samples by using a semi-supervised sample generation method based on the LiDAR point cloud data and the measured gap fraction, it is specifically used to perform:

[0168] A priori knowledge model between the LiDAR point cloud intensity and the measured gap fraction is constructed; the priori knowledge model is fused with the Beer-Lambert law to form a priori knowledge enhanced Beer-Lambert law, and the mangrove LAI samples are generated.

[0169] In some possible embodiments, when the processor 801 performs hierarchical division of cloud-covered and clear-sky reflectance based on the response relationship between the mangrove LAI samples and the hyperspectral reflectance to identify cloud-covered areas and clear-sky areas, it is specifically used to perform:

[0170] A two-dimensional spatial coordinate system is constructed based on the mangrove LAI samples and the corresponding hyperspectral reflectance; based on the two-dimensional spatial coordinate system, a midpoint straight line equation representing the difference in response relationship under cloud-covered and clear-sky conditions is calculated; and the hyperspectral image pixel is divided into a cloud-covered area or a clear-sky area according to its position relative to the midpoint straight line in the two-dimensional space.

[0171] In some possible embodiments, when the processor 801 performs hierarchical division of cloud-covered and clear-sky reflectance based on the response relationship between the mangrove LAI samples and the hyperspectral reflectance to identify cloud-covered areas and clear-sky areas, it is specifically used to perform:

[0172] The first-order derivative improved two-dimensional correlation spectroscopy analysis technique is used to analyze the response mechanism to the LAI disturbance, and the optimal spectral features are screened according to the autocorrelation peak intensity in the synchronous correlation spectrum; the feature importance is evaluated by combining the random forest regression and the SHAP algorithm, and the optimal LiDAR features are screened according to the average SHAP value size.

[0173] In some possible embodiments, the processor 801 executes the mangrove LAI inversion algorithm with high interpretability based on the optimal features, respectively constructs the mangrove LAI inversion models suitable for the cloud coverage area and the clear sky area, and then realizes the spatial mapping of the mangrove LAI based on the optimal layered inversion model, and is specifically used for executing:

[0174] An automatic peak intensity weighted regression algorithm capable of explaining the change of reflectivity with the LAI disturbance is developed, and the LAI layered inversion models suitable for the cloud coverage and the clear sky area are respectively constructed;

[0175] The coefficient of determination, the root mean square error and the average relative error are used as the accuracy indicators to evaluate the model performance; and the optimal layered inversion model is selected to realize the spatial mapping of the mangrove LAI.

[0176] The embodiment of the application further provides a computer readable storage medium, which stores instructions, when the instructions are run on a computer or a processor, the computer or the processor executes one or more steps in the above-mentioned Figure 2 The above-mentioned complex weather mangrove LAI layered remote sensing inversion device can be stored in the computer readable storage medium if the various component modules are realized in the form of software function units and sold or used as independent products.

[0177] In the above embodiments, all or part of the methods can be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the methods can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted by the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes one or more available media sets. The available media can be magnetic media (for example, floppy disk, hard disk, magnetic tape), optical media (for example, digital versatile disc (DVD)), or semiconductor media (for example, solid state disk (SSD)) and the like.

[0178] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by a computer program instructing related hardware, which can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. The storage medium includes read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and various media that can store program codes. In the case of no conflict, the technical features in the embodiments and the embodiments can be combined arbitrarily.

[0179] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. It will be apparent to those skilled in the art that various modifications can be made to the embodiments, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for remote sensing inversion of mangrove LAI stratification under complex weather, characterized in that, The method comprises: acquiring unmanned aerial vehicle laser radar point cloud data, hyperspectral remote sensing images, measured leaf area index and gap fraction data of mangrove forest regions; preprocessing the laser radar point cloud data and hyperspectral remote sensing images, and generating a LiDAR feature data set and a spectral feature data set, respectively; based on the LiDAR point cloud data and the measured gap fraction, generating mangrove forest LAI samples by using a semi-supervised sample generation method; based on the response relationship between the mangrove forest LAI samples and the hyperspectral remote sensing image reflectivity, stratifying the cloud coverage and clear sky reflectivity to identify cloud coverage areas and clear sky areas; based on the response mechanism of LAI disturbance to features, identifying optimal features for mangrove forest LAI inversion from the spectral feature data set and the LiDAR feature data set; based on the optimal features, developing a mangrove forest LAI inversion algorithm with high interpretability, and constructing a mangrove forest LAI stratified inversion model suitable for cloud coverage areas and clear sky areas, respectively, and then realizing spatial mapping of mangrove forest LAI based on the optimal stratified inversion model.

2. The method of claim 1, wherein, The unmanned aerial vehicle laser radar point cloud data and hyperspectral remote sensing images are acquired by corresponding sensors carried on unmanned aerial vehicles; the mangrove forest leaf area index and gap fraction are obtained by ground measuring instruments in the sample area while the unmanned aerial vehicle is flying.

3. The method of claim 1, wherein, The preprocessing of the laser radar point cloud data and hyperspectral remote sensing images, and the generation of a LiDAR feature data set and a spectral feature data set, respectively, comprises: resampling, denoising, ground point filtering and normalization processing of the laser radar point cloud data, and preprocessing of the hyperspectral remote sensing images including radiometric correction, mosaicking and denoising processing; extracting LiDAR features including height, intensity and density from the laser radar point cloud data; constructing a spectral feature data set based on the hyperspectral remote sensing images.

4. The method of claim 1, wherein, The generation of mangrove forest LAI samples based on the LiDAR point cloud data and the measured gap fraction by using a semi-supervised sample generation method comprises: constructing a prior knowledge model between the LiDAR point cloud intensity and the measured gap fraction; fusing the prior knowledge model and the Beer-Lambert law to form a prior knowledge enhanced Beer-Lambert law, and generating mangrove forest LAI samples.

5. The method of claim 1, wherein, The stratification of cloud coverage and clear sky reflectivity based on the response relationship between the mangrove forest LAI samples and the hyperspectral remote sensing image reflectivity to identify cloud coverage areas and clear sky areas comprises: constructing a two-dimensional spatial coordinate system of the mangrove forest LAI samples and the corresponding hyperspectral reflectivity; based on the two-dimensional spatial coordinate system, calculating a midpoint straight line equation representing the difference in response relationship under cloud coverage and clear sky conditions; according to the position of the hyperspectral image pixel in the two-dimensional space relative to the midpoint straight line, dividing it into a cloud coverage area or a clear sky area.

6. The method of claim 1, wherein, The identification of optimal features for mangrove forest LAI inversion from the spectral feature data set and the LiDAR feature data set based on the response mechanism of LAI disturbance to features comprises: The first-order derivative is used to improve the two-dimensional correlation spectroscopy to analyze the response mechanism of reflectance to LAI disturbance, and the autocorrelation peak intensity in the synchronous correlation spectrum is used to identify the optimal spectral feature; The random forest regression and SHapley Additive exPlanations (SHAP) algorithm are combined to evaluate the importance of LiDAR features, and the optimal LiDAR feature is selected according to the average SHAP value.

7. The method of claim 1, wherein, The spatial mapping of mangrove LAI is realized based on the optimal hierarchical inversion model, including: An automatic peak intensity weighted regression algorithm that can explain the change of reflectance to LAI disturbance is developed, and the hierarchical inversion models of mangrove LAI suitable for cloud-covered areas and clear-sky areas are constructed, respectively. The coefficient of determination, root mean square error and average relative error are used as accuracy indicators to evaluate the performance of the model; the optimal hierarchical inversion model is selected to draw the spatial distribution of mangrove LAI.

8. A mangrove LAI hierarchical remote sensing inversion device, characterized in that, The device comprises: An acquisition module is configured to acquire unmanned aerial vehicle laser radar point cloud data, hyperspectral remote sensing images, mangrove leaf area index and gap fraction data of a mangrove area; A first generation module is configured to preprocess the laser radar point cloud data and hyperspectral remote sensing images, and generate a LiDAR feature data set and a spectral feature data set, respectively; A second generation module is configured to generate mangrove LAI samples by using a semi-supervised sample generation method based on the LiDAR point cloud data and the measured gap fraction; A first identification module is configured to layer cloud-covered and clear-sky reflectance based on the response relationship between the mangrove LAI samples and the hyperspectral remote sensing image reflectance, to identify cloud-covered areas and clear-sky areas; A second identification module is configured to identify optimal features for mangrove LAI inversion from the spectral feature data set and the LiDAR feature data set based on the response mechanism of LAI disturbance to the features; A construction module is configured to develop a highly interpretable mangrove LAI inversion algorithm based on the optimal features, and construct hierarchical inversion models of mangrove LAI suitable for cloud-covered areas and clear-sky areas, respectively, and then realize the spatial mapping of mangrove LAI based on the optimal hierarchical inversion model.

9. A computer storage medium, characterized in that The computer storage medium stores a plurality of instructions, which are suitable for being loaded and executed by a processor to perform the steps of the method of any one of claims 1-7.

10. An electronic device, comprising: The computer program is stored on a memory and can be run on a processor, and the processor performs the steps of the method of any one of claims 1-7 when executing the program.

Images