A method for remotely sensing and monitoring the surface ecological diversity of satellites for verification

Abstract

The present invention discloses a method for remotely sensing and monitoring the verification of surface ecological diversity by satellite, belonging to the technical field of image processing, including data collection and preprocessing, ecological parameter calculation, interpretation and classification, dynamic change analysis and result output, which solves the technical problems of inconsistent data processing, insufficient dynamic change analysis and low classification accuracy in surface ecological diversity monitoring. The present invention ensures the consistency and high quality of data, can comprehensively reflect the dynamic characteristics of the ecosystem, quantifies the change trend of the ecosystem, provides an accurate reference basis for ecological protection, and improves the accuracy of complex ecosystem classification.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method for satellite remote sensing monitoring and verification of surface ecological diversity. Background Art

[0002] Biodiversity is the basis of ecosystem productivity, the cornerstone of the stable operation of the earth's ecosystem, and the foundation for maintaining the earth's ecological balance, ensuring human well-being and natural sustainable development. Biodiversity encompasses the richness and variability of all life forms on earth. Diverse organisms depend on and restrict each other, jointly maintaining the stability and functions of the ecosystem.

[0003] Although traditional ground survey methods have a certain degree of accuracy at the micro scale, their low efficiency and limited coverage make it difficult to meet the requirements of large-scale and multi-temporal dynamic monitoring. In recent years, ecological monitoring technologies based on remote sensing images have been widely applied. However, the existing technologies still have the following deficiencies:

[0004] Insufficient data processing: Traditional remote sensing image analysis methods rely on a single data source and are difficult to integrate the spatio-temporal information of multi-source data (such as meteorological and land use data), which limits the accuracy of ecosystem dynamic analysis.

[0005] Single calculation of ecological parameters: At present, the existing technologies only focus on the normalized difference vegetation index (NDVI) or simple ecological parameters, and have weak capabilities in jointly calculating and analyzing key ecological indicators (such as net primary productivity, aboveground biomass, vegetation coverage, etc.), making it difficult to comprehensively reflect the status of ecological diversity.

[0006] Limited dynamic change analysis: Current dynamic analysis is mostly limited to simple trend assessment, lacking quantitative analysis of the spatial autocorrelation and habitat fragmentation of the ecosystem, and unable to accurately capture the complex characteristics of ecosystem changes.

[0007] Insufficient classification accuracy: Traditional interpretation and classification methods (such as pixel-based classification or expert interpretation rules) perform poorly in diverse and complex ecosystems, with low classification accuracy and difficulty in meeting the needs of formulating ecological protection policies. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for satellite remote sensing monitoring and verification of surface ecological diversity, which solves the technical problems of inconsistent data processing, insufficient dynamic change analysis, and low classification accuracy in surface ecological diversity monitoring.

[0009] To achieve the above purpose, the present invention adopts the following technical solutions:

[0010] A method for satellite remote sensing monitoring and verification of surface ecological diversity includes the following steps:

[0011] Step 1: Establish a data collection module. The data collection module collects remote sensing images and auxiliary data from the Internet, preprocesses the remote sensing images and auxiliary data respectively, and obtains the original data set;

[0012] The preprocessing of remote sensing images includes radiometric correction, atmospheric correction, geometric correction, and cropping of the study area; the preprocessing of auxiliary data includes unifying to raster data and standardizing meteorological data;

[0013] Step 2: The data modeling module performs mathematical modeling on ecological data and calculates ecological parameters, including calculation of normalized difference vegetation index, net primary productivity, vegetation coverage, and aboveground biomass;

[0014] Step 3: The interpretation and classification module constructs an interpretation signature library and classifies the ecological systems shown by the ecological data;

[0015] Step 4: The dynamic change analysis module conducts dynamic analysis on ecological data, analyzes the spatial autocorrelation of ecosystem changes, quantifies the ecosystem change process, and the dynamic analysis includes trend analysis calculation and habitat fragmentation calculation;

[0016] Step 5: The result output module outputs the final results, including ecological system classification maps, dynamic change reports, and protection strategy reports.

[0017] Preferably, when performing Step 1, the remote sensing images are Landsat and MODIS image data obtained from a public platform, covering multi-temporal data within a preset year;

[0018] The auxiliary data are land use data and meteorological data obtained from a third-party platform;

[0019] The preprocessing of remote sensing images includes:

[0020] Radiometric correction processing, specifically using LandsatLEDAPS or MODIS correction tools for radiometric correction to standardize the spectral reflectance of different image sources;

[0021] Atmospheric correction processing, specifically using the FLAASH or 6S model to remove the atmospheric effect and make the image truly reflect the surface characteristics;

[0022] Geometric correction processing, specifically making all remote sensing images consistent with the ground through geometric division;

[0023] Cropping of the study area processing, specifically cropping the remote sensing images according to the preset study scope;

[0024] The preprocessing of land use data and meteorological data includes:

[0025] Unify the format of land use data into raster data, with the spatial resolution aligned with Landsat LEDAPS;

[0026] Standardize the meteorological data and interpolate it monthly to the annual time scale.

[0027] Preferably, when performing step 2, the mathematical models for calculating the normalized difference vegetation index, net primary productivity, vegetation coverage, and aboveground biomass are as follows:

[0028] Normalized difference vegetation index NDVI calculation model:

[0029] ;

[0030] Where, is the reflectance in the near-infrared band, is the reflectance in the red light band;

[0031] The normalized difference vegetation index NDVI calculation model finally outputs a multi-temporal NDVI time series data raster map within the preset year;

[0032] Net primary productivity NPP calculation model, namely the CASA model:

[0033] ;

[0034] Where, represents the net primary productivity of pixel x in month t; represents the photosynthetically active radiation absorbed by pixel x in month t; represents the actual light use efficiency of pixel x in month t; represents the vegetation growing season distribution curve factor, with a value in the range of [0,1];

[0035] Photosynthetically active radiation absorbed The calculation formula of is as follows:

[0036] ;

[0037] Where, represents the total solar radiation at pixel x in month t; 0.5 is a constant, representing the proportion of the effective solar radiation that vegetation can utilize in the total solar radiation;

[0038] represents the absorption ratio of incident photosynthetically active radiation PAR by the vegetation layer:

[0039] ;

[0040] Where, represents the NDVI value at pixel x in month t, and respectively represent the maximum and minimum NDVI values corresponding to the i-th vegetation type;

[0041] and take values of 0.01 and 0.95 respectively;

[0042] Based on sample statistics, dynamically adjust and :

[0043] ;

[0044] ;

[0045] where is the average value of NDVI at pixel x, obtained by statistically analyzing the sample NDVI data; is the standard deviation of NDVI at pixel x, used to measure the data dispersion of NDVI;

[0046] and are empirical parameters, both taking values of 1.5;

[0047] The calculation formula for the actual light energy utilization rate is as follows:

[0048] ;

[0049] where and respectively represent the influence of plants on the light energy conversion rate under different temperature conditions; represents the influence of plants on the light energy conversion rate under different moisture conditions; represents the maximum light energy utilization rate under ideal conditions;

[0050] ;

[0051] where represents the average temperature of the month when NDVI reaches the highest in the area where pixel x is located, that is, the optimum temperature, representing the optimum temperature at point x in month t;

[0052] ;

[0053] where represents the actual temperature at point x in month t;

[0054] ;

[0055] where represents the actual evapotranspiration occurring at point x in month t, Represents the potential evapotranspiration in the area at point x in month t;

[0056] The specific model for calculating vegetation coverage is as follows:

[0057] ;

[0058] Among them, represents the vegetation coverage, represents the maximum NDVI value of pure vegetation pixels, represents the minimum NDVI value of pure non-vegetation pixels;

[0059] The calculation model of aboveground biomass is as follows:

[0060] ;

[0061] Among them, B represents biomass, NDVI represents the vegetation index, represents the latitude, and α, β, and γ represent regression coefficients;

[0062] Vegetation growing season distribution curve factor The calculation formula is as follows:

[0063] ;

[0064] Among them, represents the NDVI value at pixel x in month t, is used to reflect the dynamic distribution of the vegetation growing season and depends on the change of NDVI over time.

[0065] Preferably, when performing step 3, it specifically includes the following steps:

[0066] Step 3-1: Collect characteristic samples of different ecosystem types and construct an interpretation sign library, which specifically includes collecting direct signs and indirect signs. The direct signs include shape, size, shadow, tone, color, texture, pattern, position, and layout, and the indirect signs include water system, landform, and vegetation;

[0067] Construct raster data of ecosystem changes and calculate the change value of each pixel;

[0068] Step 3-2: Construct a spatial weight matrix based on Euclidean distance or adjacency relationship , and use the Moran's I index model to evaluate the spatial autocorrelation of ecosystem changes. The specific formula is as follows:

[0069] ;

[0070] Among them, It represents Moran's I index, which reflects spatial autocorrelation. Its value range is [-1, 1]. A positive value indicates positive correlation, and a negative value indicates negative correlation; represents the number of sample points; 、 respectively represent the ecological change values at the i-th and j-th positions; represents the average value of all samples; represents the weight matrix, which represents the spatial weight between position i and position j (specifically, the reciprocal of the distance);

[0071] represents the normalization factor of the weight matrix:

[0072] ;

[0073] Step 3-3: Use deep learning methods for ecosystem classification. Specifically, use the convolutional neural network CNN model for ecosystem classification. The input data of the convolutional neural network CNN model is remote sensing images and raster data. The specific model formula is as follows:

[0074] ;

[0075] Among them, represents the k-th feature map in the l-th layer; represents the convolutional kernel weight; represents the bias term; represents the activation function ReLU;

[0076] In the pooling layer of the convolutional neural network CNN model, reduce the dimension of the feature map and retain key features. In the fully connected layer, perform classification probability prediction and output the ecological type classification result;

[0077] Step 3-4: Use the confusion matrix to evaluate the classification accuracy, calculate the overall accuracy OA and the Kappa coefficient, and the goal is that the overall accuracy OA reaches 85%.

[0078] Preferably, when performing Step 4, the specific models for trend analysis calculation and habitat fragmentation calculation are as follows:

[0079] The model for trend analysis calculation is the Sen+MK model, and the specific formula is as follows:

[0080] Sen slope formula:

[0081] ;

[0082] Among them, represents the slope of the time series change;

[0083] Using the Mann-Kendall (MK) significance test, calculate the significance p-value to determine whether the trend is significant;

[0084] The model formula for calculating habitat fragmentation is as follows:

[0085] ;

[0086] Where, is the fragmentation degree of the i-th ecosystem, is the total area, is the number of patches.

[0087] A method for satellite remote sensing monitoring and verification of surface ecological diversity according to the present invention solves the technical problems of inconsistent data processing, insufficient dynamic change analysis, and low classification accuracy in surface ecological diversity monitoring. The present invention integrates Landsat and MODIS remote sensing images, meteorological data, and land use data, and through preprocessing processes such as radiometric correction and atmospheric correction, ensures the consistency and high quality of the data. A multi-dimensional calculation model of the normalized difference vegetation index (NDVI), net primary productivity (NPP), vegetation coverage, and aboveground biomass is introduced, which can comprehensively reflect the dynamic characteristics of the ecosystem. Through trend analysis (Sen+MK model) and calculation of habitat fragmentation, the change trend of the ecosystem is quantified. The spatial autocorrelation of ecological changes is evaluated through the Moran's I index model, providing an accurate reference basis for ecological protection. The convolutional neural network (CNN) is used to classify the ecosystem. By optimizing the convolutional layer and classification model, the accuracy of complex ecosystem classification is improved, and the classification effect is evaluated using a confusion matrix to ensure that the overall accuracy (OA) reaches more than 85%. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 is a schematic diagram of the system architecture of the present invention;

[0089] Figure 2 is the main flowchart of the present invention;

[0090] Figure 3 is the flowchart of step 2 of the present invention;

[0091] Figure 4 is the flowchart of step 4 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0092] By Figures 1-4 shown, a method for satellite remote sensing monitoring and verification of surface ecological diversity includes the following steps:

[0093] Step 1: Establish a data collection module. The data collection module collects remote sensing images and auxiliary data from the Internet, and preprocesses the remote sensing images and auxiliary data respectively to obtain an original data set;

[0094] In this embodiment, in practical applications, the data collection module is deployed at the data input layer, mainly including a remote sensing image acquisition module, an auxiliary data acquisition module, and a preprocessing module. The remote sensing image acquisition module is used to obtain public remote sensing images such as Landsat and MODIS, covering multi-temporal image data; the auxiliary data acquisition module is used to obtain land use data (such as rasterized land use classification data), and at the same time obtain meteorological data (monthly meteorological observation data); the preprocessing module is used to preprocess remote sensing images and auxiliary data.

[0095] The preprocessing of remote sensing images includes radiometric correction, atmospheric correction, geometric correction, and cropping of the study area; the preprocessing of auxiliary data includes unifying to raster data and standardizing meteorological data;

[0096] When performing step 1, the remote sensing images are Landsat and MODIS image data obtained from a public platform, covering multi-temporal data within a preset year;

[0097] The auxiliary data are land use data and meteorological data obtained from a third-party platform;

[0098] In this embodiment, the image data are obtained from the official data release platforms of Landsat and MODIS satellites. The sources of the remote sensing images are as follows:

[0099] Sentinel remote sensing data: The Sentinel remote sensing data used comes from the European Space Agency website. Sentinel-2 is a high-resolution multi-spectral imaging satellite, carrying a multi-spectral imager (MSI) for land monitoring, which can provide images of vegetation, soil and water coverage, inland waterways, and coastal areas, etc. It is divided into two satellites, Sentinel-2A and Sentinel-2B. Sentinel-2B is in the same group as the Sentinel-2A satellite launched in June 2015. The altitude of the Sentinel-2 satellite is 786 km, covering 13 spectral bands, with a swath width of 290 kilometers. The ground resolutions are 10m, 20m, and 60m respectively. The revisit period of one satellite is 10 days, and the two are complementary with a revisit period of 5 days. It has different spatial resolutions from visible light and near-infrared to short-wave infrared. Among optical data, Sentinel-2 data is the only data containing three bands in the red edge range, which is very effective for monitoring vegetation health information.

[0100] Sources of MODIS satellite remote sensing data:

[0101] The MODIS remote sensing satellite data used is sourced from the MODIS download website of NASA in the United States. The MODIS NDVI (Normalized Difference Vegetation Index) and the NDVI products of the NOAA Advanced Very High Resolution Radiometer (AVHRR) can be used in combination to show the temporal series changes of the vegetation index. The MODIS NDVI products are calculated from the bidirectional atmospheric correction surface reflectance that has been processed for water, clouds, heavy aerosols, and cloud shadow masking. The global MYD13Q1 data is a 250-meter L3 data product synthesized every 16 days, with a sinusoidal projection.

[0102] Source of LANDSAT satellite remote sensing data:

[0103] The LANDSAT satellite remote sensing data mainly includes the data of two satellites, Landsat-8 and Landsat-9. The data is sourced from the NASA data download website in the United States.

[0104] The high-resolution remote sensing dataset is mainly sourced from Google imagery, and its main use is to assist in the delineation of ecosystem samples and the verification of ecosystem classification data.

[0105] The preprocessing of remote sensing images includes:

[0106] Radiometric correction processing, specifically using the Landsat LEDAPS or MODIS correction tools for radiometric correction to standardize the spectral reflectance of different image sources;

[0107] Atmospheric correction processing, specifically using the FLAASH or 6S model to remove the atmospheric effect so that the image can truly reflect the surface characteristics;

[0108] Geometric correction processing, specifically making all remote sensing images consistent with the ground through geometric partitioning;

[0109] Cropping the study area processing, specifically cropping the remote sensing image according to the preset study scope;

[0110] The preprocessing of land use data and meteorological data includes:

[0111] Unify the format of land use data into raster data, with the spatial resolution aligned with Landsat LEDAPS;

[0112] Standardize the meteorological data and interpolate it monthly to the annual time scale.

[0113] Step 2: The data modeling module performs mathematical modeling on ecological data to calculate ecological parameters, including the calculation of normalized difference vegetation index, net primary productivity, vegetation coverage, and aboveground biomass;

[0114] In this embodiment, the data modeling module is deployed in the data processing layer and is mainly used for calculating ecological parameters, specifically including the calculation of the Normalized Difference Vegetation Index (NDVI), the calculation of the Net Primary Productivity (NPP) based on the CASA model, the calculation of vegetation coverage, and the calculation of above-ground biomass.

[0115] When performing step 2, the mathematical models for the calculation of the Normalized Difference Vegetation Index, the Net Primary Productivity, the vegetation coverage, and the above-ground biomass are as follows:

[0116] Normalized Difference Vegetation Index NDVI calculation model:

[0117] ;

[0118] Among them, is the reflectance in the near-infrared band, is the reflectance in the red light band;

[0119] The Normalized Difference Vegetation Index NDVI calculation model finally outputs a raster map of multi-temporal NDVI time series data within a preset year;

[0120] In this embodiment, NDVI characterizes the vegetation growth status through the difference in reflectance between the red (Red) and near-infrared (NIR) bands. Vegetation absorbs red light for photosynthesis and strongly reflects near-infrared light. The NDVI value ranges from [-1, 1]. Generally:

[0121] NDVI > 0.2: Indicates that there is vegetation coverage, and the higher the value, the healthier the vegetation;

[0122] NDVI ≤ 0.2: Usually represents soil, rock, or water body;

[0123] The Normalized Difference Vegetation Index NDVI provides the vegetation growth status of pixels in remote sensing images and provides input for calculating the Fraction of Photosynthetically Active Radiation (FPAR) and the Vegetation Coverage (FC).

[0124] Net Primary Productivity NPP calculation model, i.e., the CASA model:

[0125] ;

[0126] Among them, represents the net primary productivity of pixel x in month t; represents the photosynthetically active radiation absorbed by pixel x in month t; represents the actual light use efficiency of pixel x in month t; represents the vegetation growing season distribution curve factor, with a value in the range of [0, 1];

[0127] In this embodiment, is the net primary productivity of pixel x during period t, which is an important indicator for measuring the amount of carbon fixed by vegetation through photosynthesis. NPP is the core output of the entire model and is used for the study of the carbon cycle in ecosystems.

[0128] Absorbed photosynthetically active radiation The calculation formula is as follows:

[0129] ;

[0130] Among them, represents the total solar radiation at pixel x in month t; 0.5 is a constant, representing the proportion of the effective solar radiation that vegetation can utilize in the total solar radiation; represents the absorption ratio of the incident photosynthetically active radiation PAR by the vegetation layer:

[0131] ;

[0132] Among them, represents the NDVI value at pixel x in month t, and represent the maximum and minimum NDVI values of the corresponding i-th vegetation type respectively, which determine the normalization range of FPAR; and take the values of 0.01 and 0.95 respectively;

[0133] In this embodiment, FPAR is a key part of APAR. They quantify the efficiency of plants in absorbing light energy and are closely related to the vegetation coverage and health status.

[0134] Based on sample statistics, dynamically adjust and :

[0135] ;

[0136] ;

[0137] Among them, is the average value of NDVI at pixel x, obtained by statistically analyzing the NDVI data of the samples; is the standard deviation of NDVI at pixel x, used to measure the data dispersion of NDVI;

[0138] and are empirical parameters, both taking the value of 1.5;

[0139] Actual light energy utilization efficiency The calculation formula is as follows:

[0140] ;

[0141] Among them, and respectively represent the influence of plants on the light energy conversion rate under different temperature conditions; represents the influence of plants on the light energy conversion rate under different water conditions; represents the maximum light energy utilization rate under ideal conditions;

[0142] In this embodiment, The energy conversion efficiency for APAR reflects the impact of environmental constraints on photosynthesis.

[0143] ;

[0144] Among them, represents the average temperature of the month when NDVI reaches the highest in the area where pixel x is located, that is, the optimal temperature, representing the optimal temperature of point x in month t;

[0145] ;

[0146] Among them, represents the actual temperature of point x in month t;

[0147] ;

[0148] Among them, represents the actual evapotranspiration that occurs at point x in month t, represents the regional potential evapotranspiration of point x in month t;

[0149] The specific model for calculating vegetation coverage is as follows:

[0150] ;

[0151] Among them, represents the vegetation coverage, represents the maximum NDVI value of pure vegetation pixels, represents the minimum NDVI value of pure non-vegetation pixels;

[0152] In this embodiment, the vegetation coverage reflects the vegetation coverage level of pixels in the study area, provides support for ecosystem evaluation and classification, and is highly correlated with the change of NDVI.

[0153] The aboveground biomass calculation model is as follows:

[0154] ;

[0155] Among them, B represents biomass, NDVI represents the vegetation index, represents latitude, and α, β, and γ represent regression coefficients;

[0156] Vegetation growing season distribution curve factor The calculation formula is as follows:

[0157] ;

[0158] Among them, represents the NDVI value at pixel x in month t, which is used to reflect the dynamic distribution of the vegetation growing season and depends on the change of NDVI over time.

[0159] Step 3: The interpretation and classification module constructs an interpretation flag library to classify the ecosystems shown by the ecological data;

[0160] When performing Step 3, it specifically includes the following steps:

[0161] Step 3-1: Collect characteristic samples of different ecosystem types to construct an interpretation flag library, specifically including collecting direct flags and indirect flags. The direct flags include shape, size, shadow, tone, color, texture, pattern, position, and layout, and the indirect flags include water systems, landforms, and vegetation;

[0162] Construct raster data of ecosystem changes and calculate the change value of each pixel;

[0163] Step 3-2: Construct a spatial weight matrix based on Euclidean distance or adjacency relationship , and use the Moran’s I index model to evaluate the spatial autocorrelation of ecosystem changes. The specific formula is as follows:

[0164] ;

[0165] Among them, represents the Moran’s I index, which reflects spatial autocorrelation, and its value range is [-1, 1]. A positive value indicates positive correlation, and a negative value indicates negative correlation; represents the number of sample points; , respectively represent the ecological change values at the i-th and j-th positions; represents the average value of all samples; represents the weight matrix, which represents the spatial weight between position i and position j (specifically the reciprocal of the distance);

[0166] represents the normalization factor of the weight matrix:

[0167] ;

[0168] Step 3-3: Use deep learning methods for ecosystem classification. Specifically, a convolutional neural network (CNN) model is adopted for ecosystem classification. The input data of the CNN model are remote sensing images and raster data. The specific model formula is as follows:

[0169] ;

[0170] Among them, represents the k-th feature map in the l-th layer; represents the convolutional kernel weight; represents the bias term; represents the activation function ReLU;

[0171] In the pooling layer of the CNN model, the dimension of the feature map is reduced to retain key features. In the fully connected layer, classification probability prediction is performed, and the classification result of the ecological type is output;

[0172] The training data for the CNN model are labeled ecological type sample data;

[0173] The loss function is cross-entropy loss:

[0174] ;

[0175] Among them, is the true value that the sample i belongs to the category c, represents the probability value predicted by the model;

[0176] Step 3-4: Use the confusion matrix to evaluate the classification accuracy, calculate the overall accuracy OA and the Kappa coefficient, and the goal is that the overall accuracy OA reaches 85%.

[0177] In this embodiment, the confusion matrix, as a prior art, has a mature application foundation and is widely used in the accuracy evaluation of classification problems, such as the fields of image recognition, ecological monitoring, and remote sensing interpretation. In this embodiment, it is necessary to accurately evaluate the classification result of the ecological type output by the CNN model. The confusion matrix can provide multi-dimensional evaluation indicators such as precision, recall, and overall accuracy (OA), which are suitable for such multi-classification tasks. Through the evaluation indicators (such as the Kappa coefficient and the overall accuracy) generated by the confusion matrix, the performance of the classification model can be intuitively quantified, and the classification error can be traced and optimized.

[0178] When specifically used, the specific operation of using the confusion matrix to evaluate the classification accuracy is as follows:

[0179] The input of the confusion matrix is: the classification result predicted by the CNN model + the true ecological system annotation sample;

[0180] The confusion matrix calculates the following key metrics to evaluate the performance of the classification model:

[0181] Overall Accuracy (OA): The ratio of the number of correctly classified samples to the total number of samples;

[0182] Kappa coefficient: Evaluates the random consistency of the classification results for a more rigorous performance assessment;

[0183] Objective requirement: OA ≥ 85% to ensure the high reliability of the ecological classification results;

[0184] The Kappa coefficient is significantly higher than 0.8, indicating strong consistency in the classification results.

[0185] Step 4: The dynamic change analysis module conducts dynamic analysis on ecological data, analyzes the spatial autocorrelation of ecosystem changes, quantifies the ecosystem change process, and the dynamic analysis includes trend analysis calculation and habitat fragmentation calculation;

[0186] In this embodiment, both the interpretation and classification module and the dynamic change analysis module are deployed in the analysis and modeling layer, mainly used for constructing the interpretation mark library, building a convolutional neural network (CNN) classification model, quantifying the change trend of ecological parameters based on the Sen + MK model, analyzing spatial autocorrelation based on the Moran's I index model, and evaluating the degree of ecosystem fragmentation.

[0187] When performing Step 4, the specific models for trend analysis calculation and habitat fragmentation calculation are as follows:

[0188] The model for trend analysis calculation is the Sen + MK model, and the specific formula is as follows:

[0189] Sen slope formula:

[0190] ;

[0191] where, represents the slope of the time series change;

[0192] Using the MK significance test, calculate the significance p-value to determine whether the trend is significant;

[0193] The model formula for habitat fragmentation calculation is as follows:

[0194] ;

[0195] where, is the fragmentation degree of the i-th type of ecosystem, is the total area, is the number of patches.

[0196] Step 5: The result output module outputs the final results, including the ecosystem classification map, the dynamic change report, and the protection strategy report.

[0197] In this embodiment, the result output module is deployed in the result output layer and is responsible for outputting the analysis results, the dynamic change report, and the protection strategy suggestions.

[0198] In this embodiment, the content shown in the ecosystem classification map is as follows:

[0199] The remotely sensed images, land use data, and meteorological data that have undergone standardized processing, and these data mainly provide the basic input data for classification;

[0200] The generated ecological parameters (such as NDVI, NPP, vegetation coverage, and aboveground biomass), and these data mainly serve as feature input data to support the classification model;

[0201] The classification results output after classifying the remotely sensed images and ecological parameters through a convolutional neural network (CNN).

[0202] The content shown in the dynamic change report is as follows:

[0203] Provide multi-temporal NDVI data, NPP, vegetation coverage and other temporal ecological parameters, and these data are also used for dynamic analysis;

[0204] Quantify the dynamic change process of the ecosystem, such as:

[0205] The results of trend analysis (Sen + MK model), used to evaluate the change trend and significance of the ecosystem;

[0206] The results of spatial autocorrelation analysis (Moran's I index model), used to analyze the spatial distribution law of ecological changes;

[0207] The results of habitat fragmentation calculation, used to quantify the fragmentation degree of the ecosystem.

[0208] The content of the protection strategy report includes:

[0209] Provide ecological parameters (such as NPP, vegetation coverage, aboveground biomass), reflecting the ecological health status;

[0210] Provide the ecosystem classification results to clarify the distribution of ecological types;

[0211] Show indicators such as trend analysis and habitat fragmentation, so as to show the change direction and potential threats of the ecosystem.

[0212] A method for satellite remote sensing monitoring and verification of surface ecological diversity according to the present invention solves the technical problems of inconsistent data processing, insufficient dynamic change analysis, and low classification accuracy in surface ecological diversity monitoring. The present invention integrates Landsat and MODIS remote sensing images, meteorological data, and land use data, and through preprocessing processes such as radiometric correction and atmospheric correction, ensures the consistency and high quality of the data. It introduces multi-dimensional calculation models of the normalized difference vegetation index (NDVI), net primary productivity (NPP), vegetation coverage, and aboveground biomass, which can comprehensively reflect the dynamic characteristics of the ecosystem. Through trend analysis (Sen+MK model) and calculation of habitat fragmentation, the change trend of the ecosystem is quantified. Through the Moran's I index model, the spatial autocorrelation of ecological changes is evaluated, providing an accurate reference basis for ecological protection. A convolutional neural network (CNN) is used to classify the ecosystem. By optimizing the convolutional layer and classification model, the accuracy of complex ecosystem classification is improved, and the classification effect is evaluated using a confusion matrix to ensure that the overall accuracy (OA) reaches more than 85%.

Claims

1. A method for remotely sensing and monitoring the surface ecological diversity by satellite, characterized in that: It includes the following steps: Step 1: Establish a data collection module. The data collection module collects remote sensing images and auxiliary data from the Internet, and preprocesses the remote sensing images and auxiliary data respectively to obtain an original dataset; The preprocessing of remote sensing images includes radiometric correction, atmospheric correction, geometric correction, and cropping of the study area; the preprocessing of auxiliary data includes unifying to raster data and standardizing meteorological data; The preprocessing of remote sensing images includes: Geometric correction processing, specifically, through geometric division, making all remote sensing images consistent with the ground; The preprocessing of land use data and meteorological data includes: Unifying the format of land use data to raster data, with the spatial resolution aligned with Landsat LEDAPS; Standardizing meteorological data and interpolating it monthly to an annual time scale; Step 2: The data modeling module performs mathematical modeling on ecological data and calculates ecological parameters, including calculation of the normalized difference vegetation index, calculation of net primary productivity, calculation of vegetation coverage, and calculation of aboveground biomass; Net primary productivity NPP calculation model, that is, the CASA model: ； Among them, represents the net primary productivity of pixel x in month t; represents the photosynthetically active radiation absorbed by pixel x in month t; represents the actual light use efficiency of pixel x in month t; represents the vegetation growing season distribution curve factor, with a value in the range of [0, 1]; Vegetation growing season distribution curve factor The calculation formula is as follows: ； Among them, represents the NDVI value at pixel x in month t, which is used to reflect the dynamic distribution of the vegetation growing season and depends on the change of NDVI over time; Step 3: The interpretation and classification module constructs an interpretation sign library and classifies the ecological systems shown by the ecological data; When performing Step 3, it specifically includes the following steps: Step 3-1: Collect characteristic samples of different ecosystem types and construct an interpretation sign library, specifically including collecting direct signs and indirect signs. Direct signs include shape, size, shadow, tone, color, texture, pattern, position, layout, and indirect signs include water system, landform, and vegetation; Construct raster data of ecosystem changes and calculate the change value of each pixel; Step 3-2: Construct a spatial weight matrix based on Euclidean distance or adjacency relationship , and use the Moran's I index model to evaluate the spatial autocorrelation of ecosystem changes. The specific formula is as follows: ； Among them, represents the Moran's I index, which reflects spatial autocorrelation and ranges from [-1, 1]. A positive value indicates positive correlation, and a negative value indicates negative correlation. represents the number of sample points; , respectively represent the ecological change values at the i-th and j-th positions; represents the average value of all samples; represents the weight matrix, indicating the spatial weight between position i and position j. Denotes the normalization factor of the weight matrix: ； Step 3-3: Use deep learning methods for ecosystem classification. Specifically, use the convolutional neural network CNN model for ecosystem classification. The input data of the convolutional neural network CNN model are remote sensing images and raster data. The specific model formula is as follows: ； Among them, represents the k-th feature map in the l-th layer; represents the convolutional kernel weights; represents the bias term; represents the activation function ReLU; Reduce the dimension of the feature map in the pooling layer of the convolutional neural network CNN model, retain key features, and perform classification probability prediction in the fully connected layer to output the ecological type classification result; Step 3-4: Use a confusion matrix to evaluate the classification accuracy and calculate the overall accuracy OA and Kappa coefficient; Step 4: The dynamic change analysis module performs dynamic analysis on ecological data, analyzes the spatial autocorrelation of ecosystem changes, quantifies the ecosystem change process, and the dynamic analysis includes trend analysis calculation and habitat fragmentation calculation; Step 5: The result output module outputs the final results, including the ecosystem classification map, dynamic change report, and protection strategy report.

2. The method for remotely sensing and monitoring the surface ecological diversity by satellite as claimed in claim 1, wherein: When performing Step 1, the remote sensing images are Landsat and MODIS image data obtained from a public platform, covering multi-temporal data within a preset year; The auxiliary data are land use data and meteorological data obtained from a third-party platform; The preprocessing of remote sensing images includes: Radiometric correction processing, specifically, using the Landsat LEDAPS or MODIS correction tool for radiometric correction to standardize the spectral reflectance of different image sources; Atmospheric correction processing, specifically, using the FLAASH or 6S model to remove the atmospheric effect and make the image truly reflect the surface characteristics; Cropping the research area, specifically cropping the remote sensing image according to the preset research scope.

3. The method for remotely sensing and monitoring the surface ecological diversity of a satellite as claimed in claim 1, wherein: When performing step 2, the mathematical models for calculating the normalized difference vegetation index, net primary productivity, vegetation coverage, and aboveground biomass are as follows: Normalized difference vegetation index (NDVI) calculation model: ； wherein, is the reflectance in the near-infrared band, is the reflectance in the red light band; The normalized difference vegetation index (NDVI) calculation model finally outputs a raster map of multi-temporal NDVI time series data within the preset year; Absorbed photosynthetically active radiation The calculation formula is as follows: ； Among them, represents the total solar radiation at pixel x in month t; 0.5 is a constant representing the proportion of the available solar radiation that can be utilized by vegetation in the total solar radiation; represents the absorption ratio of the incident photosynthetically active radiation PAR by the vegetation layer: ； Among them, represents the NDVI value at pixel x in month t, and respectively represent the maximum and minimum NDVI values corresponding to the i-th vegetation type; and take the values of 0.01 and 0.95 respectively; Dynamically adjust based on sample statistics and : ; ； Among them, is the average value of NDVI at pixel x, obtained by statistically sampling NDVI data; is the standard deviation of NDVI at pixel x, used to measure the data dispersion of NDVI; and are empirical parameters, and their values are both 1.5; Actual light energy utilization rate The calculation formula is as follows: ； Among them, and respectively represent the effects of plants on the light energy conversion rate under different temperature conditions; represents the effect of plants on the light energy conversion rate under different water conditions; represents the maximum light energy utilization rate under ideal conditions; ； Among them, represents the average temperature in the month when NDVI reaches the highest in the area where pixel x is located, that is, the optimal temperature, representing the optimal temperature at point x in month t; ； Among them, represents the actual temperature at point x in month t; ； Among them, represents the actual evapotranspiration occurring at point x in month t, represents the regional potential evapotranspiration at point x in month t; The specific model for calculating vegetation coverage is as follows: ； Among them, represents the vegetation coverage, represents the maximum NDVI value of pure vegetation pixels, represents the minimum NDVI value of pure non-vegetation pixels; The aboveground biomass calculation model is as follows: ； Among them, B represents biomass, NDVI represents vegetation index, represents latitude, and α, β and γ represent regression coefficients.

4. A method for remotely sensing and monitoring the verification of surface ecological diversity by satellite as claimed in claim 1, characterized in that: When performing step 4, the specific models for trend analysis calculation and habitat fragmentation calculation are as follows: The model for trend analysis calculation is the Sen+MK model, and the specific formula is as follows: Sen slope formula: ； Among them, represents the slope of the time series change; Using the MK significance test, calculate the significance p-value to determine whether the trend is significant; The model formula for habitat fragmentation calculation is as follows: ； Among them, is the fragmentation degree of the i-th type of ecosystem, is the total area, is the number of patches.

Images