Remote sensing recognition method and system applied to ecological system investigation

Abstract

The invention relates to the technical field of remote sensing recognition, in particular to a remote sensing recognition method and system applied to ecological system investigation. The method comprises the following steps: acquiring multi-temporal remote sensing image data of a target area; extracting a land cover type of the multi-temporal remote sensing image data, and performing dominant human activity area identification on the multi-temporal remote sensing image data according to the land cover type to generate dominant human activity area data; acquiring night light data of the target area; performing space-time registration on the night light data of the target area and the multi-temporal remote sensing image data to generate fused night light remote sensing data; performing boundary region extraction on the multi-temporal remote sensing image data through the land cover type to obtain edge region data; and calculating a vegetation index and a noctilucence index of the marginal region data based on the fused noctilucence remote sensing data. According to the method, through multi-temporal and multi-source data fusion and multi-index time sequence analysis, the accuracy and comprehensiveness of ecological system investigation remote sensing recognition are improved.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing identification technology, and in particular to a remote sensing identification method and system applied to ecosystem investigation. Background Art

[0002] Thanks to the rapid development of satellite remote sensing technology, the application of multispectral and hyperspectral imaging techniques has greatly enhanced the ability to identify ecosystems. Multispectral remote sensing, by capturing spectral information across different wavelengths, enables the classification and monitoring of ecological elements such as vegetation types, water bodies, and soil. Hyperspectral remote sensing further enhances the ability to discern species and ecological structures in detail, promoting refined analysis of ecosystem diversity and health. With the advancement of computer technology and artificial intelligence, remote sensing image analysis methods based on machine learning and deep learning have emerged. These methods improve the efficiency and accuracy of ecosystem identification by automatically extracting image features and are capable of effectively processing large-scale, high-dimensional remote sensing data. However, traditional methods currently focus primarily on overt human activities, ignoring the complex ecological impacts of marginal areas. Furthermore, existing technologies often focus on identifying overt threats, ignoring potential threats within ecosystems. This results in low accuracy and comprehensiveness in remote sensing identification of ecosystems. Summary of the Invention

[0003] Based on this, it is necessary to provide a remote sensing identification method and system for ecosystem survey to solve at least one of the above technical problems.

[0004] To achieve the above-mentioned purpose, a remote sensing identification method for ecosystem survey is provided, the method comprising the following steps:

[0005] Step S1: Acquire multi-temporal remote sensing image data of the target area; extract the surface cover type of the multi-temporal remote sensing image data, and identify the explicit human activity area of ​​the multi-temporal remote sensing image data according to the surface cover type to generate explicit human activity area data;

[0006] Step S2: Acquire nighttime light data of the target area; perform spatiotemporal registration of the nighttime light data of the target area with the multi-temporal remote sensing image data to generate fused nighttime light remote sensing data; extract the boundary area of ​​the multi-temporal remote sensing image data according to the surface cover type to obtain edge area data; calculate the vegetation index and nighttime light index of the edge area data based on the fused nighttime light remote sensing data;

[0007] Step S3: Performing biological behavior correlation analysis on edge area data using vegetation index and night light index to generate edge effect identification data; analyzing the temporal changes of low-frequency disturbances in explicit human activity area data to generate low-frequency disturbance identification data; performing ecosystem health assessment prediction based on the low-frequency disturbance identification data and edge effect identification data to generate ecosystem health assessment prediction data for the target area;

[0008] Step S4: Based on the ecosystem health assessment prediction data of the target area, the regional latent threat output is performed on the ecosystem of the target area, thereby obtaining the latent threat level and distribution map.

[0009] This invention effectively improves the collaborative processing capabilities of multi-source heterogeneous data in terms of spatial resolution and temporal consistency by performing spatiotemporal registration and fusion of multi-temporal remote sensing imagery and nighttime light data. It extracts explicit human activity areas based on land cover type and analyzes edge effects by combining vegetation indices and nighttime light indices in boundary areas, enhancing the ability to identify the spatial extent of human activity and areas with blurred boundaries. It analyzes biobehavioral associations using vegetation and nighttime light variation in edge areas, and combines low-frequency disturbance time series data from explicit areas to achieve a deep coupling analysis between dynamic ecological responses and human activity interference. Using edge effect and low-frequency disturbance identification data as core input features, an ecosystem health prediction model is established, which exhibits strong data-driven characteristics and model scalability. Using ecosystem health prediction data, a latent threat level classification mechanism is constructed and spatial distribution maps are output, effectively improving the accuracy and visualization capabilities of ecological threat identification. Therefore, through multi-temporal and multi-source data fusion and multi-index time series analysis, this invention improves the accuracy and comprehensiveness of remote sensing identification in ecosystem surveys.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: performing multi-period sampling control of remote sensing images on the target area to obtain multi-temporal remote sensing image data;

[0012] Step S12: performing radiation correction and geometric correction on the multi-temporal remote sensing image data to generate corrected multi-temporal remote sensing image data;

[0013] Step S13: extracting the surface spectral features of the corrected multi-temporal remote sensing image data, and performing temporal change analysis and classification processing on the surface spectral features to generate surface cover type data;

[0014] Step S14: interpreting the surface cover type data for human activity indicators to generate preliminary human activity mask data; performing spatial consistency screening and boundary correction on the preliminary human activity mask data to generate explicit human activity area data.

[0015] The present invention can systematically capture remote sensing information of different time nodes in the target area through a multi-period sampling control mechanism, providing a high-quality spatiotemporal data foundation for subsequent change detection and dynamic analysis. It effectively eliminates radiation errors caused by factors such as sensors, sun angles, atmospheric interference, and geometric distortions caused by changes in viewing angles and terrain, ensuring that remote sensing data has high analytical availability and spatial registration accuracy. By extracting the spectral change characteristics of multi-phase remote sensing images and combining them with temporal dynamic patterns for classification analysis, the accuracy of distinguishing complex surface types (such as conversion of farmland, woodland and construction land) can be improved. By using the surface cover classification results to extract human activity indicators, supplemented by spatial consistency screening and boundary correction, the spatial integrity and boundary accuracy of human activity masks can be significantly improved, reducing the misjudgment rate.

[0016] Preferably, step S13 includes the following steps:

[0017] Step S131: extracting multi-band reflectance values ​​of surface pixels at different time phases from the corrected multi-temporal remote sensing image data to obtain a surface spectral feature sequence with a time tag;

[0018] Step S132: Calculate the temporal variation trend of the reflectance of key bands in the multi-temporal remote sensing image data based on the surface spectral feature sequence and determine whether its variation pattern meets any of the following conditions and mark it accordingly to generate temporal variation judgment mark data: the NDVI index has seasonal periodic fluctuations and the amplitude within the year is greater than 0.3; the position of the intersection of the blue light band and the red edge band reflectance varies by more than ±15nm in three consecutive temporal phases; the number of inflection points of the spectral curve differs by more than 2 in different growth periods;

[0019] Step S133: Use the time series change judgment mark data to identify the coverage type of the surface spectral characteristics. If any of the following conditions is met, it can be determined as the specified coverage type and the surface coverage type data is generated: if the maximum NDVI value exceeds 0.6 and the NDVI fluctuation range within the year is greater than 0.35, it is classified as "crops"; if the short-wave infrared reflectivity changes by less than 5% throughout the year and the NDVI is stable below 0.2 for a long time, it is classified as "building surface"; if the normalized difference index of red light and near-infrared shows a double peak within the year and the peak times correspond to spring and autumn respectively, it is classified as "deciduous woodland".

[0020] The present invention generates a spectral feature sequence with time labels by extracting multi-band reflectance at different phases, providing a high-dimensional and continuous remote sensing information foundation for subsequent temporal change analysis, and significantly improving the ability to identify dynamic characteristics such as seasonality and growing period of land features. Innovatively combining multi-scale features such as NDVI seasonal fluctuation amplitude, displacement of the intersection of blue light and red edge, and change of spectral curve inflection point to judge temporal change patterns, it breaks through the traditional classification method based on a single vegetation index or static spectral value and enhances the ability to discriminate changes in complex ecological cover types. By matching the criteria data of surface temporal characteristics with rules and setting fine classification standards in combination with multiple biophysical parameters (such as NDVI, red light / near infrared index, short-wave infrared stability), the recognition accuracy of typical surface types such as "crops", "building surfaces", and "deciduous woodlands" can be effectively improved, and the classification system's generalization ability for cross-seasonal and cross-regional data can be enhanced. The overall process can automatically extract key change features and perform discrimination and classification during large-scale remote sensing data processing, reducing the cost of manual intervention and adapting to the remote sensing classification needs under different regional ecological backgrounds.

[0021] Preferably, step S2 includes the following steps:

[0022] Step S21: Acquire nighttime light data of the target area;

[0023] Step S22: performing radiation normalization and noise suppression on the raw radiation data of nighttime lights to generate net radiation data of nighttime lights; performing spatiotemporal reference point alignment on the net radiation data of nighttime lights and the multi-temporal remote sensing image correction data to generate a spatiotemporal registration parameter set;

[0024] Step S23: performing grid-level registration and resampling fusion on the target area nighttime light data and the multi-temporal remote sensing image data according to the spatiotemporal registration parameter set to generate fused nighttime light remote sensing data; performing edge detection and morphological region growing analysis on the boundary change areas of the remote sensing image using the surface cover type data to generate a boundary area layer;

[0025] Step S24: performing spatial extraction and segmentation on the boundary area layer to generate edge area data; performing vegetation index calculation on the edge area data based on the fused night light remote sensing data to generate edge area vegetation index data;

[0026] Step S25: extracting the night light index of the edge area data based on the fused night light remote sensing data to generate the night light index data of the edge area.

[0027] The present invention constructs fused night light remote sensing data under a unified spatiotemporal reference system by normalizing radiation, spatiotemporal registration and fusion processing of night light data and multi-temporal remote sensing images, effectively integrating the intensity of human activities and natural surface information, and improving the comprehensive perception of dynamic changes in edge areas. By constructing a raster-level registration model based on precise spatiotemporal reference points and resampling and fusing image data of different resolutions, it can effectively alleviate the problems of spatial dislocation and scale mismatch in the process of fusion of heterogeneous remote sensing data, and improve the consistency and credibility of data integration. By combining edge detection and morphological region growth processing with surface cover types, the boundary change area of ​​the object can be located more accurately, which is particularly suitable for edge transition zone analysis scenarios such as urban expansion and land use conversion. The boundary area is further refined into edge areas using spatial extraction and segmentation methods, providing accurate spatial boundary support for subsequent indicator extraction and ecological analysis, and improving the analysis model's ability to identify ecological transition zones and human intervention boundaries. The vegetation index and night light index of the edge area are extracted simultaneously based on the uniformly registered data, which can reflect the interactive relationship between the ecological status and human interference, and provide highly sensitive, multi-dimensional indicator support for subsequent biological behavior analysis and ecological risk assessment.

[0028] Preferably, in step S23, performing edge detection and morphological region growing analysis on the boundary change area of ​​the remote sensing image using the surface cover type data includes:

[0029] The edge of the remote sensing image boundary change area is extracted using the surface cover type data. The high threshold is set to 100 pixel gray value and the low threshold is set to 50 pixel gray value. A 3×3 Gaussian kernel is used for smoothing to obtain contour line data with clear boundaries and convert it into vector polygon format.

[0030] The contour data is used as the seed area, and the region growing algorithm is used to expand to the area with similar features. The spectral similarity threshold is set to ±10% reflectivity, and the area threshold is greater than or equal to 1000m. 2 , the growth radius is within 90 meters, and the growth results are re-labeled and rasterized into the boundary area;

[0031] The boundary area and the remote sensing image boundary change area are superimposed to generate a vector layer; the following attribute fields are assigned to each boundary area unit: change type, change area, boundary length, and change period, thereby outputting the boundary area layer.

[0032] The present invention introduces the surface cover type as a priori information in edge extraction, combines the set double threshold (high threshold 100, low threshold 50) with 3×3 Gaussian kernel smoothing, effectively suppresses noise and strengthens the boundary contour, thereby obtaining contour line data with clear boundaries and complete structure, significantly improving the accuracy of remote sensing image boundary change detection. Using contour line data as the seed area, spectral similarity (±10% reflectance) and area (≥1000m 2 ) and spatial growth radius (≤90 meters) and other multi-dimensional restriction conditions are used to perform regional growth analysis, effectively preventing over-expansion or missed detection, and improving the spatial modeling capabilities of complex boundary areas. In the process of regional identification, the extraction results are converted from contour lines to vector polygons and then rasterized, which not only improves the flexibility and accuracy of data expression, but also facilitates subsequent spatial operations and ecological indicator extraction, and improves the engineering adaptability of the remote sensing processing process. By superimposing images on the boundary area and assigning attribute fields such as change type, change area, boundary length and change period, a structured and quantifiable boundary area layer is constructed, providing strong data support and visualization foundation for subsequent ecological monitoring, edge effect analysis and regional dynamic modeling. With the help of the boundary superposition and attribute labeling mechanism of multi-temporal images, the fine tracking and classification identification of the changes in the boundary area over time are achieved, which improves the controllability and accuracy of the monitoring of changes in the ecological edge area.

[0033] Preferably, in step S3, performing biological behavior correlation analysis on edge area data using vegetation index and night light index includes:

[0034] The edge area data is divided into spatial grids, and vegetation index changes are extracted by combining time series remote sensing images to generate edge vegetation dynamic index data;

[0035] Overlay nighttime light remote sensing images on edge area data, extract nighttime light intensity and stability indicators, and generate edge nighttime light index data;

[0036] Conduct bivariate spatiotemporal synergistic analysis of vegetation index and night light index to generate vegetation-night light coupling characteristic data;

[0037] Clustering of spatiotemporal patterns of vegetation-nightlight coupling characteristic data to identify the coupling effects of human activity intervention and ecological disturbances, and generating biological behavior response layer data;

[0038] Conduct edge sensitivity analysis and biodiversity impact assessment on biological behavior response layer data to generate edge effect identification data.

[0039] By dividing the spatial grid of edge areas and dynamically extracting multi-temporal vegetation indices, the present invention can meticulously capture the temporal variation trends of vegetation coverage, providing high-precision quantitative data support for the spatiotemporal dynamics of ecosystems. Night light data is used to extract night light intensity and its stability, and combined with vegetation indices to construct composite indicators, effectively reflecting the intensity of human activities and their spatial distribution characteristics, and improving the ability to identify explicit and implicit interference. Through bivariate collaborative analysis of vegetation index and night light index, the mutual influence and coupling mechanism between vegetation dynamics and human activities are revealed, which helps to identify the human behavioral drivers behind ecological disturbances and deepen the understanding of ecosystem behavior. Spatiotemporal pattern clustering technology is used to classify and identify coupling features, effectively separating different intervention patterns and ecological responses, enhancing the classification accuracy and stability of biological behavioral responses, and facilitating the formulation of targeted ecological management and protection strategies. By conducting edge sensitivity and biodiversity impact assessments on biological behavioral response layers, vulnerable areas and potential risks of ecosystems can be identified, assisting in ecosystem health assessments and scientific determination of implicit threats, and achieving early warning and precise intervention of ecological risks.

[0040] Preferably, analyzing the low-frequency disturbance time series changes of the explicit human activity area data in step S3 includes:

[0041] Perform long-term resampling and smoothing on the data of explicit human activity areas to extract stable change trends and generate low-frequency activity change baseline data;

[0042] Extract remote sensing index sequences from multi-temporal remote sensing image data, perform wavelet decomposition, extract low-frequency disturbance principal components, and generate disturbance frequency band feature data;

[0043] Perform time window sliding analysis on disturbance frequency band feature data using low-frequency activity change baseline data to identify low-frequency trend deviations and breakpoints, and generate low-frequency change anomaly detection data;

[0044] The low-frequency change anomaly detection data and the explicit human activity area data are spatially constrained to be consistent, short-period high-frequency false interference is filtered out, and a steady-state disturbance response layer is generated;

[0045] The disturbance pattern classification and intensity calibration are performed on the steady-state disturbance response layer to generate low-frequency disturbance identification data.

[0046] The present invention can effectively filter out short-period fluctuations and noise through long-term time series resampling and smoothing, extract a stable and continuous baseline of low-frequency activity changes, and provide reliable basic data for subsequent disturbance analysis. The wavelet decomposition method can finely separate the low-frequency disturbance components in remote sensing indicators, reveal the potential abnormal changes in the ecosystem on a time scale, and improve the sensitivity and resolution of disturbance detection. Through sliding window analysis of disturbance frequency band characteristic data and baseline data, trend deviations and breakpoints can be identified in real time, and fine dynamic monitoring of low-frequency abnormal changes can be achieved, thereby improving the real-time and accuracy of time series anomaly detection. By utilizing the spatial constraints of explicit human activity areas, short-period high-frequency noise interference can be effectively eliminated, the spatial coherence and stability of the disturbance response can be ensured, and the reliability of low-frequency disturbance identification can be improved. Pattern classification and intensity calibration of the steady-state disturbance response layer can achieve a quantitative description of low-frequency disturbances, supporting scientific assessment and management decisions on the health status of the ecosystem.

[0047] Preferably, performing ecosystem health assessment prediction based on the low-frequency disturbance identification data and the edge effect identification data in step S3 includes:

[0048] The low-frequency disturbance identification data and edge effect identification data were spatially semantically fused to construct an ecological disturbance comprehensive index layer. The ecological sensitivity weighted overlay analysis was performed on the ecological disturbance comprehensive index layer to generate multi-factor ecological vulnerability assessment data.

[0049] The multi-factor ecological vulnerability assessment data were divided into data sets to generate model training sets and model testing sets;

[0050] The model training set is trained using the long short-term memory neural network algorithm to generate an ecosystem health assessment pre-model; the ecosystem health assessment pre-model is optimized and iterated based on the model test set to generate an ecosystem health assessment model;

[0051] The multi-factor ecological vulnerability assessment data is input into the ecosystem health assessment model for health assessment prediction, generating ecosystem health assessment prediction data for the target area.

[0052] The present invention constructs an ecological interference comprehensive index layer through spatial semantic fusion of low-frequency disturbance identification data and edge effect identification data, thereby improving the comprehensive perception of multi-dimensional interference factors in the ecosystem. The comprehensive index layer is weighted and superimposed using ecological sensitivity weights to scientifically reflect the impact of different ecological factors on regional vulnerability and enhance the accuracy and detail of ecological vulnerability assessment. By dividing the model training set and test set, and combining the long short-term memory neural network (LSTM) algorithm for model training and optimization, dynamic learning and accurate prediction of the health status of the ecosystem are achieved. By continuously optimizing the pre-model through the model test set, the model's adaptability to unknown data is enhanced, ensuring the stability and reliability of the ecosystem health assessment results. By inputting multi-factor ecological vulnerability data into the trained ecosystem health assessment model, the ecological health status of the target area can be accurately predicted, assisting in ecological environmental protection and management decisions.

[0053] Preferably, step S4 includes the following steps:

[0054] Step S41: Perform factor stratification extraction on the ecosystem health assessment prediction data to obtain data on potential latent interference factors in the region;

[0055] Step S42: performing a regional comprehensive analysis of the ecosystem in the target area based on the data of potential latent interference factors in the area to extract data on sensitive areas of latent threats, wherein the regional comprehensive analysis includes analysis of land use, vegetation continuity, and water body connectivity;

[0056] Step S43: Comparing the fluctuations of sensitive area data in time series, identifying the trend of implicit interference impact, generating a threat interference trend layer, and overlaying and analyzing the threat interference trend layer with the ecosystem health assessment prediction data to divide the implicit threat level intervals of different intensities;

[0057] Step S44: geographically plotting the latent threat level intervals and defining the distribution range, and generating a latent threat level and distribution map of the target area.

[0058] The present invention uses factor hierarchical extraction technology to effectively capture potential hidden interference factors in the region and improve the depth and precision of identifying complex threat factors in the ecosystem. Combined with land use, vegetation continuity and water connectivity analysis, the scientific extraction of sensitive areas of hidden threats to the ecosystem is achieved, and the spatial correlation and ecological rationality of hidden threat identification are enhanced. By analyzing the time series fluctuation changes of sensitive area data, the impact trend of hidden interference is dynamically identified, and the understanding and grasp of the spatiotemporal evolution of hidden threats to the ecosystem are enhanced. The threat interference trend layer is superimposed on the ecosystem health assessment prediction data, and a comprehensive analysis is conducted in combination with multi-source information to achieve quantitative division of hidden threat levels and improve the accuracy and practicality of threat assessment. Through geographic spatial mapping and scoping, intuitive hidden threat levels and distribution maps are generated to facilitate risk monitoring, early warning and ecological protection decision support for ecological management departments.

[0059] In this specification, a remote sensing identification system for ecosystem survey is provided, which is used to execute the above-mentioned remote sensing identification method for ecosystem survey. The remote sensing identification system for ecosystem survey includes:

[0060] The multi-temporal remote sensing recognition module is used to obtain multi-temporal remote sensing image data of the target area; extract the surface cover type of the multi-temporal remote sensing image data, and identify the explicit human activity area based on the surface cover type to generate explicit human activity area data;

[0061] The night light remote sensing recognition module is used to obtain night light data of the target area; perform spatiotemporal registration of the target area night light data and multi-temporal remote sensing image data to generate fused night light remote sensing data; extract the boundary area of ​​the multi-temporal remote sensing image data based on the surface cover type to obtain edge area data; and calculate the vegetation index and night light index of the edge area data based on the fused night light remote sensing data;

[0062] The correlation analysis module is used to conduct biological behavior correlation analysis on edge area data through vegetation index and night light index to generate edge effect identification data; analyze the temporal changes of low-frequency disturbances in data of obvious human activity areas to generate low-frequency disturbance identification data; and perform ecosystem health assessment and prediction based on low-frequency disturbance identification data and edge effect identification data to generate ecosystem health assessment prediction data for the target area;

[0063] The ecosystem assessment module is used to output regional hidden threats to the ecosystem of the target area based on the ecosystem health assessment prediction data of the target area, thereby obtaining the hidden threat level and distribution map.

[0064] The beneficial effects of the present invention are that the multi-temporal remote sensing identification module can achieve dynamic acquisition and comprehensive analysis of multi-temporal remote sensing images of the target area, improving the temporal continuity and spatial accuracy of surface cover type identification; it can identify areas of explicit human activity in combination with surface cover types, enhancing the spatial positioning and monitoring capabilities of human activity impacts, and providing a reliable data basis for subsequent ecological analysis. The night light remote sensing identification module achieves deep fusion of multi-source remote sensing data through precise spatiotemporal registration of night light data with multi-temporal remote sensing images, improving the accuracy of extracting vegetation and night light features in edge areas; it uses surface cover types to effectively extract image boundary change areas, accurately obtain ecological edge area data, and enhance the scientific nature of edge effect identification. The correlation analysis module comprehensively reveals the interaction between human activities and vegetation dynamics in the ecosystem through biological behavior correlation analysis of vegetation index and night light index, and enhances the spatiotemporal understanding of ecological disturbances; combines the time series change analysis of low-frequency disturbances in the data of explicit human activity areas to achieve accurate identification of long-term disturbance trends in the ecosystem; conducts ecosystem health assessment and prediction based on multi-dimensional disturbance data, improves the scientific nature and foresight of ecological status assessment, and provides decision support for ecological protection and restoration. The ecosystem assessment module relies on ecosystem health assessment prediction data, and the system outputs regional hidden threats to achieve in-depth identification and hierarchical management of potential ecological risks; through the generation of hidden threat levels and distribution maps, it provides intuitive and scientific spatial risk distribution information, which facilitates ecological environment management and precise intervention. Therefore, the present invention improves the accuracy and comprehensiveness of remote sensing identification of ecosystem surveys through multi-phase and multi-source data fusion and multi-indicator time series analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flowchart of the steps of a remote sensing identification method applied to ecosystem surveys;

[0066] Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG.

[0067] Figure 3 for Figure 1 Detailed implementation steps of step S4 in FIG.

[0068] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0069] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0070] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0071] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0072] To achieve this, please refer to Figures 1 to 3 , a remote sensing identification method for ecosystem investigation, the method comprising the following steps:

[0073] Step S1: Acquire multi-temporal remote sensing image data of the target area; extract the surface cover type of the multi-temporal remote sensing image data, and identify the explicit human activity area of ​​the multi-temporal remote sensing image data according to the surface cover type to generate explicit human activity area data;

[0074] Step S2: Acquire nighttime light data of the target area; perform spatiotemporal registration of the nighttime light data of the target area with the multi-temporal remote sensing image data to generate fused nighttime light remote sensing data; extract the boundary area of ​​the multi-temporal remote sensing image data according to the surface cover type to obtain edge area data; calculate the vegetation index and nighttime light index of the edge area data based on the fused nighttime light remote sensing data;

[0075] Step S3: Performing biological behavior correlation analysis on edge area data using vegetation index and night light index to generate edge effect identification data; analyzing the temporal changes of low-frequency disturbances in explicit human activity area data to generate low-frequency disturbance identification data; performing ecosystem health assessment prediction based on the low-frequency disturbance identification data and edge effect identification data to generate ecosystem health assessment prediction data for the target area;

[0076] Step S4: Based on the ecosystem health assessment prediction data of the target area, the regional latent threat output is performed on the ecosystem of the target area, thereby obtaining the latent threat level and distribution map.

[0077] This invention effectively improves the collaborative processing capabilities of multi-source heterogeneous data in terms of spatial resolution and temporal consistency by performing spatiotemporal registration and fusion of multi-temporal remote sensing imagery and nighttime light data. It extracts explicit human activity areas based on land cover type and analyzes edge effects by combining vegetation indices and nighttime light indices in boundary areas, enhancing the ability to identify the spatial extent of human activity and areas with blurred boundaries. It analyzes biobehavioral associations using vegetation and nighttime light variation in edge areas, and combines low-frequency disturbance time series data from explicit areas to achieve a deep coupling analysis between dynamic ecological responses and human activity interference. Using edge effect and low-frequency disturbance identification data as core input features, an ecosystem health prediction model is established, which exhibits strong data-driven characteristics and model scalability. Using ecosystem health prediction data, a latent threat level classification mechanism is constructed and spatial distribution maps are output, effectively improving the accuracy and visualization capabilities of ecological threat identification. Therefore, through multi-temporal and multi-source data fusion and multi-index time series analysis, this invention improves the accuracy and comprehensiveness of remote sensing identification in ecosystem surveys.

[0078] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a flow chart showing the steps of a remote sensing identification method for ecosystem survey according to the present invention. In this example, the remote sensing identification method for ecosystem survey includes the following steps:

[0079] Step S1: Acquire multi-temporal remote sensing image data of the target area; extract the surface cover type of the multi-temporal remote sensing image data, and identify the explicit human activity area of ​​the multi-temporal remote sensing image data according to the surface cover type to generate explicit human activity area data;

[0080] Step S2: Acquire nighttime light data of the target area; perform spatiotemporal registration of the nighttime light data of the target area with the multi-temporal remote sensing image data to generate fused nighttime light remote sensing data; extract the boundary area of ​​the multi-temporal remote sensing image data according to the surface cover type to obtain edge area data; calculate the vegetation index and nighttime light index of the edge area data based on the fused nighttime light remote sensing data;

[0081] Step S3: Performing biological behavior correlation analysis on edge area data using vegetation index and night light index to generate edge effect identification data; analyzing the temporal changes of low-frequency disturbances in explicit human activity area data to generate low-frequency disturbance identification data; performing ecosystem health assessment prediction based on the low-frequency disturbance identification data and edge effect identification data to generate ecosystem health assessment prediction data for the target area;

[0082] Step S4: Based on the ecosystem health assessment prediction data of the target area, the regional latent threat output is performed on the ecosystem of the target area, thereby obtaining the latent threat level and distribution map.

[0083] In this embodiment of the present invention, NPP-VIIRS nighttime light remote sensing data with the same time span as step S1 is obtained, with a temporal accuracy of the data on a monthly scale and a spatial resolution of 500 meters. The VIIRS data is resampled using the nearest neighbor interpolation method and unified to a 10-meter resolution to ensure spatial consistency with the Sentinel-2 data. Spatial alignment is performed using the gdalwarp tool of the GDAL library in Python. The control points are manually corrected by selecting six stable and invariant points of the ground object, with the correction error not exceeding 0.3 pixels. Temporal alignment is performed by taking the average value of the VIIRSDN for three months in each quarter to make it consistent with the time points of the quarterly remote sensing imagery. Edge area extraction is achieved by constructing a classification boundary of surface cover. The Canny edge detection algorithm is used with threshold parameters set to a low threshold of 0.05 and a high threshold of 0.15 to extract the boundaries between buildings and non-building areas. Subsequently, a buffer analysis operation is applied with a buffer radius set to ±150 meters to form an edge buffer zone with a width of 300 meters. The National Density Observatory (NDVI) was calculated using the formula: NDVI = (B8 - B4) / (B8 + B4), where B8 represents the near-infrared band and B4 represents the red band. NDVI results were normalized to a range of 0 to 1 and extracted using a masked edge region. The night light index was calculated using the normalized VIIRS DN value: NLI = DN / 63, where the maximum DN value is 63. All processed results were exported in GeoTIFF format with a WGS 84 spatial reference. Edge region data was provided by the buffer output from step S2, with spatial units of 10 meters. A two-dimensional matrix of NDVI and NLI was constructed for all pixels within the region. A 5×5 sliding window (50 m × 50 m actual area) was constructed for local statistical processing, and the mean NDVI and NLI values ​​were calculated within each window. Pearson correlation analysis was used to determine the correlation coefficient, r, between NDVI and NLI. Regions with |r| > 0.6 were identified as areas with significant behavioral correlations. All sliding window results were reconstructed into a new GeoTIFF-formatted edge effect identification map, with each pixel value representing the r-value at that location. For the data on areas with overt human activity, the NDVI time series at their central points were extracted and decomposed using the Daubechies wavelet (db4) with orders 1 to 4. Low-frequency components with periods greater than one year were extracted, and the standard deviation σ of this component was calculated. Areas meeting this standard, with a perturbation threshold of σ > 0.12, were marked as low-frequency perturbation areas. The output perturbation map was then converted to a 10-meter-resolution GeoTIFF format. Finally, the edge effect identification map and the low-frequency perturbation map were pixel-wise overlaid, and the following ecological health index assessment model was used: H = 1-(0.5·σNDVI + 0.3·NLI + 0.2·rabs), where σNDVI is the standard deviation of the low-frequency perturbation, NLI is the night light index, and rabs is the absolute correlation coefficient between NDVI and NLI. The index was normalized to the range [0, 1], and a 10-meter-resolution health index layer was output.Using the ecosystem health assessment map (health index layer) generated in S3 as the primary data, the target area was classified into implicit threat levels. Based on the health index value (H), the following classification criteria were used: H ≥ 0.8 for Level I (no threat), 0.6 ≤ H < 0.8 for Level II (mild threat), 0.4 ≤ H < 0.6 for Level III (moderate threat), and H < 0.4 for Level IV (severe threat). During the classification process, the RasterCalculator tool was used to assign values ​​from 1 to 4 within the aforementioned interval. All results were exported in GeoTIFF format with legends. To generate a regional spatial distribution map, the "Reclassify" tool in ArcGIS was used to color the layer, setting the color coding: Level I green, Level II yellow, Level III orange, and Level IV red. The "Fishnet" grid tool was used to construct 100 m × 100 m analysis cells. The area percentage of each threat level within each grid cell was calculated to generate a raster heat map with a resolution of 100 meters. The final output is a hidden threat level distribution map (GeoTIFF format), a grid statistical heat map (Shapefile format), and a threat level spatial proportion table (CSV format).

[0084] Preferably, step S1 includes the following steps:

[0085] Step S11: performing multi-period sampling control of remote sensing images on the target area to obtain multi-temporal remote sensing image data;

[0086] Step S12: performing radiation correction and geometric correction on the multi-temporal remote sensing image data to generate corrected multi-temporal remote sensing image data;

[0087] Step S13: extracting the surface spectral features of the corrected multi-temporal remote sensing image data, and performing temporal change analysis and classification processing on the surface spectral features to generate surface cover type data;

[0088] Step S14: interpreting the surface cover type data for human activity indicators to generate preliminary human activity mask data; performing spatial consistency screening and boundary correction on the preliminary human activity mask data to generate explicit human activity area data.

[0089] In an embodiment of the present invention, by performing multi-period sampling control of remote sensing images in the target area, the sampling frequency and sampling period need to be determined according to the temporal variation characteristics of the ecosystem. Taking the East Asian monsoon region as an example, remote sensing image data for the three typical seasons of spring, summer and autumn should be obtained each year to cover the growth, prosperity and decline periods of the ecosystem. The remote sensing image data source uses Sentinel-2MSI remote sensing data with a resolution of not less than 10 meters, covering the entire target area, and ensuring a cloud coverage filtering rate of not less than 95%. Multi-period sampling adopts a monthly download method, and the image date interval (for example, the 10th of each month ± 3 days), spatial overlap rate greater than 95%, and cloud cover less than 5% are set as screening conditions, and the download operation is performed on a data platform such as the ESA Copernicus Open Access Hub. The downloaded original remote sensing images are sorted according to the naming rules and a multi-temporal data directory structure is established. They are sorted by timestamp to form a continuous time series remote sensing data stack for subsequent processing and calling. Radiometric correction is based on the sensor radiometric response (DN) of the original remote sensing imagery, using the SEN2COR tool for atmospheric correction and surface reflectance conversion. Parameters such as solar altitude, sensor observation angle, surface atmospheric model (e.g., mid-latitude summer atmospheric model), and aerosol type (continental) are set, and radiometric conversion is performed uniformly. Geometric correction is performed using high-precision geographic control point (GCP) data and DEM data. GCP data is matched to the coordinates of control points marked on a national 1:50,000 topographic map and corrected using a least-squares fitting algorithm. Terrain distortion is compensated for using the SRTM 30-meter resolution DEM. All corrected images are uniformly output in the UTM projection format using the WGS84 coordinate system (specific partition numbers are determined by the image center longitude). The spatial resolution is uniformly resampled to 10 meters, ensuring that all image coverage areas have an overlap error of less than one pixel within the boundary lines. Surface spectral features are extracted using a combination of principal component analysis (PCA) and vegetation index calculation. Using remote sensing imagery from each temporal phase as input, we calculate the Normalized Difference Vegetation Index (NDVI), Normalized Difference Building Index (NDBI), and NDWI (Water Body Index). The specific calculation formulas are: NDVI = (NIR-RED) / (NIR+RED), NDBI = (SWIR-NIR) / (SWIR+NIR), and NDWI = (GREEN-NIR) / (GREEN+NIR). NIR represents the near-infrared band, RED represents the red band, SWIR represents the shortwave infrared band, and GREEN represents the green band. The band positions are based on the Sentinel-2MSI sensor specifications: Band 8 (NIR), Band 4 (RED), Band 11 (SWIR), and Band 3 (GREEN).Subsequently, NDVI data were extracted pixel by pixel in time series. A maximum value composite (MVC) strategy was used to generate a seasonal NDVI peak layer. K-means clustering (k = 5) was then performed to classify the land cover types into five categories: grassland, woodland, water bodies, bare land, and built-up areas. During classification, NDVI thresholds were set (e.g., NDVI > 0.5 for woodland, NDVI < 0.2 for bare land). Pixels experiencing two or more major changes in a year were excluded (deemed unstable). The land cover data were interpreted as indicators of human activity, primarily using a mask screening algorithm based on the triple criteria of a building index (NDBI) greater than 0.2, NDVI less than 0.3, and NDWI less than 0 to construct a preliminary human activity mask. These thresholds were determined based on analysis of field samples. For the sample set, the NDBI threshold was set at the 95th percentile for pixels in the urban core area, while the NDVI and NDWI were set based on seasonal minimum values. After initial mask data extraction, it is screened for spatial consistency. Connected domains larger than 0.5 square kilometers are retained, while areas smaller than this threshold are considered isolated noise points and removed. Boundary correction uses a boundary buffer fusion operation. Specifically, boundary pixels are expanded by 5 pixels to form a boundary buffer. A majority voting rule is applied to determine their identity and adjust excessively fragmented boundary areas. The final output is explicit human activity area data, maintaining a spatial resolution of 10 meters and using the same coordinate system as the corrected image. It is output in GeoTIFF format for direct loading and processing in subsequent analysis modules.

[0090] Preferably, step S13 includes the following steps:

[0091] Step S131: extracting multi-band reflectance values ​​of surface pixels at different time phases from the corrected multi-temporal remote sensing image data to obtain a surface spectral feature sequence with a time tag;

[0092] Step S132: Calculate the temporal variation trend of the reflectance of key bands in the multi-temporal remote sensing image data based on the surface spectral feature sequence and determine whether its variation pattern meets any of the following conditions and mark it accordingly to generate temporal variation judgment mark data: the NDVI index has seasonal periodic fluctuations and the amplitude within the year is greater than 0.3; the position of the intersection of the blue light band and the red edge band reflectance varies by more than ±15nm in three consecutive temporal phases; the number of inflection points of the spectral curve differs by more than 2 in different growth periods;

[0093] Step S133: Use the time series change judgment mark data to identify the coverage type of the surface spectral characteristics. If any of the following conditions is met, it can be determined as the specified coverage type and the surface coverage type data is generated: if the maximum NDVI value exceeds 0.6 and the NDVI fluctuation range within the year is greater than 0.35, it is classified as "crops"; if the short-wave infrared reflectivity changes by less than 5% throughout the year and the NDVI is stable below 0.2 for a long time, it is classified as "building surface"; if the normalized difference index of red light and near-infrared shows a double peak within the year and the peak times correspond to spring and autumn respectively, it is classified as "deciduous woodland".

[0094] In this embodiment of the present invention, band reflectance values ​​are extracted pixel by pixel from the multi-temporal remote sensing image data after radiometric and geometric correction in step S12. The 10 primary bands of Sentinel-2 MSI data are used, including blue light (Band 2, center wavelength 490nm), green light (Band 3, 560nm), red light (Band 4, 665nm), red edge (Band 5, 705nm; Band 6, 740nm; Band 7, 783nm), near infrared (Band 8, 842nm), and shortwave infrared (Band 11, 1610nm; Band 12, 2190nm). Using geographic coordinates as indexes, the reflectance values ​​of the above-mentioned bands are extracted for each pixel in all temporal images, and a surface spectral feature sequence with the following structure is constructed: {pixel ID:(t1,[R_b2,R_b3,...,R_b12]),(t2,[R_b2,R_b3,...,R_b12]),...,(tn,[R_b2,R_b3,...,R_b12])}, where ti represents the i-th timestamp and R_bj is the apparent reflectance of band j. All temporal data are archived in chronological order, and the time tags are recorded in the Julian Day encoding format of the observation date, forming a spectral sequence with standard time attributes, ensuring that each pixel has complete spectral variation characteristics within the year. Based on the spectral feature sequence extracted in step S131, the NDVI value sequence and the blue light and red edge band reflectance sequence of each pixel are calculated in sequence, and a temporal variation analysis is performed. The NDVI value was calculated using the formula: NDVI = (Band 8 - Band 4) / (Band 8 + Band 4). For each pixel, the maximum and minimum NDVI values ​​within the year were counted, and the amplitude (maximum minus minimum) was calculated to determine whether it was greater than 0.3. Fast Fourier transform (FFT) analysis was also used to detect seasonal periodic fluctuations. If the frequency domain principal component period was within the range of 300 to 370 days, the seasonal fluctuation was marked as significant. For the red-edge bands (Bands 5 to 7) and the blue band (Band 2), the spectral curves were fitted using linear interpolation, and the wavelength position of the intersection was calculated. The intersection point was defined as the wavelength where the reflectance curves of the two bands intersected. The curves were reconstructed by interpolating at 1 nm intervals within the range of 705 nm ± 30 nm. The wavelength corresponding to the reflectance intersection point was determined. Comparisons were performed at three consecutive time points. If the intersection point position varied by more than ±15 nm, a significant change in the red-edge blue reflectance crossover was marked. The second-order derivative method is used to count the number of inflection points in the spectral curve. The second-order difference sequence of the spectral reflectance sequence of each phase is calculated, and the number of extreme points is counted as the inflection point number indicator. If the difference between the maximum and minimum number of inflection points within a year exceeds 2, it is marked as a significant structural change.Finally, the three judgment conditions are combined to form a three-digit logical vector (e.g., [1, 0, 1]), which is the "temporal change judgment mark data" for the pixel. The "temporal change judgment mark data" obtained in step S132 is combined with the specific values ​​of the NDVI and SWIR bands in step S131 for classification. The land cover type is determined according to the following rules: For the NDVI sequence, if the maximum NDVI value within the year is greater than 0.6, and the difference between the maximum and minimum NDVI values ​​(intra-year amplitude) exceeds 0.35, and the NDVI seasonal fluctuation item in the corresponding temporal change mark is 1 (i.e., there is a seasonal cycle), then the pixel is classified as "crop." A year-round time series analysis of shortwave infrared reflectance (Bands 11 and 12) was performed, calculating the difference between the maximum and minimum values. If the annual variation was less than 5% (i.e., the difference between the maximum and minimum values ​​was within 0.05), the NDVI maximum was less than 0.2, the NDVI showed no significant fluctuation throughout the year, and the "inflection point difference" in the change flag was 0 (no significant structural change), the pixel was classified as "building surface." A normalized difference series derived from the NDVI was calculated for the red band (Band 4) and the near-infrared band (Band 8), testing whether it exhibited two peaks within the year. Extreme value analysis was used to identify the peak locations. If the two main peaks occurred between March and November, the peak difference was greater than 0.2, and both the "NDVI seasonal fluctuation" and "inflection point change" in the change flags were 1, the pixel was classified as "deciduous woodland." Finally, each pixel was evaluated once according to the above criteria, and a land cover type label was output if any of the criteria were met. The classification results are output in GeoTIFF format, and different types are encoded by value ranges in single-band rasters (for example, 1 for crops, 2 for building surfaces, 3 for deciduous woodlands, etc.) for subsequent interpretation.

[0095] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0096] Step S21: Acquire nighttime light data of the target area;

[0097] Step S22: performing radiation normalization and noise suppression on the raw radiation data of nighttime lights to generate net radiation data of nighttime lights; performing spatiotemporal reference point alignment on the net radiation data of nighttime lights and the multi-temporal remote sensing image correction data to generate a spatiotemporal registration parameter set;

[0098] Step S23: performing grid-level registration and resampling fusion on the target area nighttime light data and the multi-temporal remote sensing image data according to the spatiotemporal registration parameter set to generate fused nighttime light remote sensing data; performing edge detection and morphological region growing analysis on the boundary change areas of the remote sensing image using the surface cover type data to generate a boundary area layer;

[0099] Step S24: performing spatial extraction and segmentation on the boundary area layer to generate edge area data; performing vegetation index calculation on the edge area data based on the fused night light remote sensing data to generate edge area vegetation index data;

[0100] Step S25: extracting the night light index of the edge area data based on the fused night light remote sensing data to generate the night light index data of the edge area.

[0101] In this embodiment of the present invention, the VIRSDNB (Visible Infrared Imaging Radiometer Suite Day / Night Band) nighttime light remote sensing image data is selected to obtain a nighttime light image sequence that matches the time range of the multi-temporal remote sensing image. The monthly average nighttime light image can be downloaded from the Earth Observation Group platform provided by NOAA. The spatial resolution is about 500m, the data format is GeoTIFF, and the value range unit is nW / cm 2 / sr. Using spatial clipping operations (e.g., using gdalwarp), the nighttime light imagery was cropped to the same spatial extent as the remote sensing imagery, and the data was converted to a floating-point single-band layer for subsequent radiometric correction and fusion. VIIRS nighttime light data was normalized to a minimum-maximum range between 0 and 1. The formula is: L_norm = (L_raw - L_min) / (L_max - L_min); where L_raw is the raw nighttime light intensity, L_min is the minimum value within the region (after removing background values), and L_max is the 95th percentile (excluding anomalous highlights). Gaussian noise was suppressed using a median filter (3x3 window) or wavelet denoising (e.g., Haar wavelet), preserving the structure of continuous illuminated areas and removing discrete light points. Basic geographic reference points (e.g., urban road intersections, water boundaries, and nighttime highlights) that correspond to the nighttime light data were extracted from the multi-temporal remote sensing imagery. Use geographic location similarity algorithms (such as SIFT or ORB feature point matching) to extract co-located points and calculate the affine registration matrix (or TPS transformation). The output result is a "space-time registration parameter set" containing affine parameters, coordinate offsets, and time indexes. For night light data and multi-temporal remote sensing image data, gdalwarp is used to perform a unified coordinate system transformation based on the space-time registration parameter set. Use the bilinear interpolation resampling method to adjust the night light data to a spatial resolution consistent with the remote sensing image (such as 10m or 30m). The night light layer is used as a new band and superimposed on the remote sensing multi-band image to generate fused night light remote sensing data (NFRSD). Use the generated surface cover type data to extract boundary combination areas such as "crop-building surface" or "crop-deciduous forest land". Use the Sobel operator or Canny operator for edge detection to extract cover type mutation edges. A morphological region growing algorithm (e.g., grayscale region expansion + closing operation) was applied to generate a spatially extensible "junction area layer." A vectorization operation (e.g., gdal_polygonize) was used to extract the planar edge regions of the junction area layer, generating vector edge data in Shapefile format. Attribute information (e.g., boundary category, degree of spectral variation) was combined to perform regional division and filter out small error areas. The red band (Band 4) and near-infrared band (Band 8) were extracted from the fused night light remote sensing data. The standard NDVI formula was used: NDVI = (NIR-Red) / (NIR+Red); in each edge area, the average NDVI value of the pixels was calculated to generate "edge area vegetation index data" (one vegetation index value per area). The night light band (a newly added band) was extracted from the fused night light remote sensing data.The Night Light Index (NLI) is defined as the average value of the night light band within a region, or calculated using the normalized night light rate of change: NLI = mean(L_night) / max(L_night_in_region). For each edge region, the Night Light Index is calculated and combined with land cover type information for temporal and spatial comparative analysis. The resulting vector boundary file is accompanied by a Night Light Index attribute, supporting applications such as urban edge development intensity assessment and land use zoning identification.

[0102] Preferably, in step S23, performing edge detection and morphological region growing analysis on the boundary change area of ​​the remote sensing image using the surface cover type data includes:

[0103] The edge of the remote sensing image boundary change area is extracted using the surface cover type data. The high threshold is set to 100 pixel gray value and the low threshold is set to 50 pixel gray value. A 3×3 Gaussian kernel is used for smoothing to obtain contour line data with clear boundaries and convert it into vector polygon format.

[0104] The contour data is used as the seed area, and the region growing algorithm is used to expand to the area with similar features. The spectral similarity threshold is set to ±10% reflectivity, and the area threshold is greater than or equal to 1000m. 2 , the growth radius is within 90 meters, and the growth results are re-labeled and rasterized into the boundary area;

[0105] The boundary area and the remote sensing image boundary change area are superimposed to generate a vector layer; the following attribute fields are assigned to each boundary area unit: change type, change area, boundary length, and change period, thereby outputting the boundary area layer.

[0106] In this embodiment of the present invention, raster-formatted land cover data (e.g., each pixel value represents a land class code: farmland = 1, building = 2, woodland = 3, etc.) is input. The current and historical land cover data are overlaid, and areas of change are extracted as boundary change regions. A 3×3 Gaussian kernel is applied to the overlaid layer to eliminate small-scale noise and spurious edge changes. Examples of Gaussian kernels are [[1,2,1], [2,4,2], [1,2,1]] / 16. Contours are extracted using the dual-threshold Canny edge detection algorithm, with a high threshold of 100 grayscale and a low threshold of 50 grayscale. The output is a pixel-level contour image, reflecting the sudden change in land class boundaries. A contour tracing algorithm (such as findContours in OpenCV) is used to convert the edge lines into vector polygons. Simplification (Douglas–Peucker algorithm with a threshold of 2 pixels) is performed to reduce the number of vertices and improve subsequent analysis efficiency. This results in "contour vector data," the initial boundary seed region. The contour vectors were converted into raster seed layers and combined with the multi-band data of the fused remote sensing image. The growth conditions were set as follows: spectral similarity threshold: ±10% reflectance range (compared with NDVI or blue / red / near infrared bands to ensure the consistency of the ground features); area threshold: ≥1000m 2 , eliminate small patch interference; growth radius: ≤90 meters, that is, the expansion from the seed point does not exceed 90 meters. Reassign the growth results to independent boundary area codes to maintain regional closure. Use connected domain markers (4 or 8 neighborhoods) to ensure that each growth area is an independent surface unit. Rasterize the output to generate a "boundary area raster layer". Perform a Boolean intersection operation on the boundary area layer and the original remote sensing image boundary change area layer, and only retain the overlapping areas of the two. Use GIS overlay analysis methods (such as Intersect or gdal_rasterize+gdal_polygonize in ArcGIS) to generate the final vector layer of the boundary area. For each boundary area unit, calculate and assign the following attribute fields: Change type (change_type): Assign values ​​according to the change land type combination, such as "farmland→building"; Change area (area_m2): Calculated through vector surface data, the actual area of ​​each polygon (unit: m 2 ); boundary_length: Calculates the total length of the polygon boundary (unit: meters); change_period: Combines the imagery's temporal index and assigns a value such as "2021Q2–2023Q1." Output is a standard vector layer (such as Shapefile or GeoJSON format), supporting overlay analysis, chart statistics, and 3D presentation.

[0107] Preferably, in step S3, performing biological behavior correlation analysis on edge area data using vegetation index and night light index includes:

[0108] The edge area data is divided into spatial grids, and vegetation index changes are extracted by combining time series remote sensing images to generate edge vegetation dynamic index data;

[0109] Overlay nighttime light remote sensing images on edge area data, extract nighttime light intensity and stability indicators, and generate edge nighttime light index data;

[0110] Conduct bivariate spatiotemporal synergistic analysis of vegetation index and night light index to generate vegetation-night light coupling characteristic data;

[0111] Clustering of spatiotemporal patterns of vegetation-nightlight coupling characteristic data to identify the coupling effects of human activity intervention and ecological disturbances, and generating biological behavior response layer data;

[0112] Conduct edge sensitivity analysis and biodiversity impact assessment on biological behavior response layer data to generate edge effect identification data.

[0113] In this embodiment of the present invention, a spatial grid system for the edge region is formed by dividing edge region data into regular grid cells at a certain spatial resolution (e.g., 500 m x 500 m or 1 km x 1 km). Multi-temporal remote sensing images (e.g., MODIS, Landsat, Sentinel-2, etc.) are collected to cover the edge region, ensuring the continuity and integrity of the time series. Vegetation indices (e.g., NDVI, EVI, etc.) are calculated for the remote sensing images at each time point to obtain a vegetation index value for each spatial grid cell. Combined with the time series, the temporal variation characteristics of the vegetation index for each grid cell (e.g., seasonal trends, abnormal fluctuations) are extracted to generate a dataset of edge vegetation dynamic indices. The spatial grid of the edge region is spatially overlaid with nighttime light remote sensing data (e.g., DMSP-OLS or VIIRS data) for the corresponding time period. The nighttime light radiation intensity of each grid cell is calculated to obtain a nighttime light index representing the intensity of human activities. Stability indicators of the nighttime light index (e.g., time series standard deviation, coefficient of fluctuation, etc.) are calculated to distinguish between stable and fluctuating human activities. This results in a dataset of edge nighttime light indices containing both nighttime light intensity and stability indicators. The vegetation index and nightlight index were normalized to eliminate dimensional differences, ensuring the synchronization of the two index data in time and space. Bivariate spatiotemporal correlation analysis (e.g., spatiotemporal cross-correlation function) was used to assess the coordinated changes in vegetation and nightlight at different temporal and spatial scales. A coupling coordination model was used to calculate the coupling strength and coordination between the vegetation index and nightlight index, quantifying the relationship between the two. This generated a spatiotemporal dataset describing the coupling characteristics of vegetation and nightlight, reflecting the interactive relationship between ecological and human activities. Based on the vegetation-nightlight coupling characteristic data, typical spatiotemporal pattern characteristics (e.g., periodicity, trends, and mutation points) were extracted. Pattern recognition was performed using spatiotemporal clustering algorithms (e.g., spatiotemporal K-means, DBSCAN, and spatiotemporal spectral clustering). Areas of strong human activity interference (high nightlight, low vegetation coupling patterns) and ecological disturbance (abnormal vegetation decline accompanied by nightlight changes) were identified. A zonal biobehavioral response layer was generated to identify the spatial distribution of different coupling effects. Spatial statistical analysis of the biobehavioral response layer was performed to identify grid cells in edge areas that were sensitive to human activity or ecological changes. A spatial weight matrix is ​​used to analyze the spatial spread and impact of edge effects. Combined with existing biodiversity monitoring data or model predictions, the impact of coupled effect areas on species diversity and ecosystem function is assessed. A multi-metric comprehensive evaluation method (such as species richness, species evenness, and ecological function index) is used to rank impacts. This produces a comprehensive dataset reflecting the impacts and sensitivity of edge areas to biodiversity, which can be used for subsequent ecological protection or intervention decisions.

[0114] Of particular importance is the clustering of spatiotemporal patterns of vegetation-nightlight coupling characteristic data to identify the coupling effects of human activity intervention and ecological disturbances.

[0115] Extract the spatiotemporal characteristics of vegetation-nightlight coupling feature data, and use density peak clustering algorithm to perform spatiotemporal pattern recognition to generate cluster label data;

[0116] Map the spatiotemporal distribution of cluster label data and construct a spatiotemporal pattern clustering layer;

[0117] Statistical correlation analysis methods were used to calculate the coupling effect index between vegetation characteristics and night light intensity, generating a coupling effect coefficient matrix. A comprehensive evaluation of the coupling effect was conducted by combining the spatiotemporal pattern clustering layer with the coupling effect coefficient matrix, generating distribution data on the intensity of biological behavioral responses.

[0118] The biological behavior response intensity distribution data is mapped to the spatial coordinate system to generate biological behavior response layer data.

[0119] In an embodiment of the present invention, edge region data is spatially gridded and combined with long-term remote sensing imagery to extract vegetation indices (such as NDVI) that change over time, generating edge vegetation dynamic index data. Simultaneously, nighttime light remote sensing images are superimposed with edge region data to extract indicators such as nighttime light intensity and nighttime light stability, generating edge nighttime light index data. On this basis, a bivariate spatiotemporal collaborative analysis is performed on the vegetation index and the nighttime light index to construct a coupled multidimensional feature set and generate vegetation-nighttime light coupling feature data. Subsequently, the main spatiotemporal features of the vegetation-nighttime light coupling feature data are extracted, including dynamic features such as mean, trend, volatility, periodicity, and mutation points in the time series, as well as indicators such as spatial distribution concentration, scalability, spatial gradient, and local spatial autocorrelation. Based on this, the Density Peaks Clustering (DPC) algorithm is used to cluster and identify high-dimensional spatiotemporal coupling features. By calculating the local density and relative distance between samples, significantly different coupling patterns are identified, generating cluster label data. The cluster labels are mapped to geographic space, and a spatiotemporal pattern clustering layer is constructed to reflect the distribution pattern and evolution trend of different coupling patterns in the region. Subsequently, statistical correlation analysis methods (such as Pearson correlation coefficient, mutual information, canonical correlation analysis, etc.) are used to calculate the synergistic relationship between vegetation changes and night light changes in each cluster category to generate a coupling effect coefficient matrix. Furthermore, the coefficient matrix is ​​combined with the clustering layer to conduct a comprehensive evaluation of the coupling effect to form the distribution data of the intensity of biological behavior response. This data reflects the synergistic mechanism and spatiotemporal strength differences between changes in the intensity of human activities and responses to ecological disturbances. Finally, the biological behavior response intensity distribution data is mapped to a unified spatial coordinate system to generate high-resolution biological behavior response layer data, providing spatial support for subsequent edge sensitivity analysis, biodiversity risk assessment, and ecosystem health status monitoring.

[0120] Of particular importance is the combination of the spatiotemporal pattern clustering layer and the coupling effect coefficient matrix to conduct a comprehensive assessment of the coupling effect, which also includes:

[0121] For each cluster unit in the spatiotemporal pattern clustering layer, the corresponding coupling effect coefficient matrix subset is extracted to generate local coupling index data. Based on the local coupling index data, a weighted fusion algorithm is used to integrate the multi-factor coupling strength to generate a comprehensive coupling strength score.

[0122] Perform spatial interpolation on the comprehensive coupling strength score to fill in the areas not covered by the spatiotemporal pattern clustering layer and generate a continuous coupling strength distribution map;

[0123] Threshold segmentation technology is used to identify areas with significant coupling strength and determine hotspots of highly responsive biological behaviors;

[0124] Conduct spatiotemporal change distribution analysis on high-response biological behavior hotspots to generate biological behavior response intensity distribution data.

[0125] In an embodiment of the present invention, edge area data is spatially gridded and combined with long-term remote sensing imagery to extract vegetation indices (such as NDVI) that change over time, generating edge vegetation dynamic index data. Simultaneously, nighttime light remote sensing images are superimposed with edge area data to extract indicators such as nighttime light intensity and nighttime light stability, generating edge nighttime light index data. On this basis, a bivariate spatiotemporal collaborative analysis is performed on the vegetation index and the nighttime light index to construct a coupled multidimensional feature set and generate vegetation-nighttime light coupling feature data. Subsequently, the main spatiotemporal features of the coupled feature data are extracted, including dynamic features such as mean, trend, volatility, periodicity, and mutation points in the time series, as well as indicators such as spatial distribution concentration, spatial gradient, and local spatial autocorrelation. Based on this, the Density Peaks Clustering (DPC) algorithm is used to cluster and identify high-dimensional spatiotemporal coupling features. By analyzing the local density and relative distance between samples, significantly different coupling patterns are identified, cluster label data is generated, and this data is mapped to geographic space to construct a spatiotemporal pattern clustering layer. Next, statistical correlation analysis methods (such as the Pearson correlation coefficient, canonical correlation coefficient, and mutual information) were used to calculate the synergistic relationship between vegetation change and night light change, forming a coupling effect coefficient matrix. Based on this, a comprehensive assessment of the coupling effect was conducted by combining the spatiotemporal pattern clustering layer with the coupling effect coefficient matrix. Specifically, for each cluster unit in the spatiotemporal pattern clustering layer, a subset of the coupling effect coefficient matrix corresponding to the cluster unit was extracted to generate local coupling index data. Based on these indices, a weighted fusion algorithm was used to fuse multiple coupling factors (such as the correlation between night light intensity stability and vegetation change rate) to generate a comprehensive coupling intensity score. To improve spatial continuity, the comprehensive coupling intensity score data was spatially interpolated to fill in areas not covered by the spatiotemporal pattern clustering layer, generating a continuous coupling intensity distribution map. Subsequently, a threshold segmentation technique was used to identify areas with significant comprehensive coupling intensity, extracting areas with high coupling intensity responses and identifying them as high-response biobehavioral hotspots. Furthermore, the spatiotemporal distribution and evolution trends of these high-response biobehavioral hotspots were analyzed to identify their expansion, migration, or decline patterns, ultimately generating a biobehavioral response intensity distribution map.

[0126] Preferably, analyzing the low-frequency disturbance time series changes of the explicit human activity area data in step S3 includes:

[0127] Perform long-term resampling and smoothing on the data of explicit human activity areas to extract stable change trends and generate low-frequency activity change baseline data;

[0128] Extract remote sensing index sequences from multi-temporal remote sensing image data, perform wavelet decomposition, extract low-frequency disturbance principal components, and generate disturbance frequency band feature data;

[0129] Perform time window sliding analysis on disturbance frequency band feature data using low-frequency activity change baseline data to identify low-frequency trend deviations and breakpoints, and generate low-frequency change anomaly detection data;

[0130] The low-frequency change anomaly detection data and the explicit human activity area data are spatially constrained to be consistent, short-period high-frequency false interference is filtered out, and a steady-state disturbance response layer is generated;

[0131] The disturbance pattern classification and intensity calibration are performed on the steady-state disturbance response layer to generate low-frequency disturbance identification data.

[0132] In this embodiment of the present invention, multi-temporal remote sensing images and related activity indicator data of areas with overt human activity are collected to ensure a long time span (e.g., several years) with uniform or nearly uniform time intervals. The original time series data is resampled to a uniform time interval (e.g., monthly, quarterly, or annual), and missing values ​​are filled. Interpolation methods (e.g., linear interpolation, spline interpolation) are used to ensure time series continuity. Sliding average, LOESS smoothing, or local weighted regression methods are used to remove high-frequency fluctuations, extract long-term stable trends, and form baseline data for low-frequency activity changes. Key remote sensing indicators such as NDVI, NDBI (building index), and night light intensity are calculated from multi-temporal images to form indicator sequences at multiple time points. An appropriate wavelet basis (e.g., Daubechies wavelet db4) is selected, and multi-scale wavelet decomposition is performed on each indicator sequence to separate low-frequency components (approximation coefficients) from high-frequency components (detail coefficients). The principal components reflecting low-frequency disturbances are extracted to generate disturbance frequency band feature data. An appropriate time window size (e.g., 12 months or 24 months) is set and the data is moved across the disturbance frequency band feature data sequence. Calculate statistical indicators (mean, variance, trend change rate, etc.) within each window. Use breakpoint detection algorithms (such as CUSUM and Pelt algorithms) to identify trend deviations and breakpoints, generating low-frequency change anomaly detection data containing low-frequency trend deviations and breakpoints. Using the low-frequency activity change baseline data generated in step 1 as a reference, perform a sliding window comparative analysis on the disturbance frequency band feature data. Confirm whether the detected trend deviations and breakpoints are significantly different from the baseline trend to avoid misjudgments. Combine the baseline and disturbance feature analysis results to generate more accurate low-frequency change anomaly detection data. Use the degree of overlap between the spatial boundaries of visible human activity areas and the spatial locations of the anomaly detection data to eliminate inconsistent anomalies. Use spatial clustering (such as DBSCAN) to identify the spatial coherence of the anomaly areas. Combined with time series high-frequency signal analysis, remove noise interference with a period less than a set threshold and generate a steady-state disturbance response layer constrained by spatial consistency. Based on the spatial morphology and temporal characteristics of the disturbance response, classify and identify disturbance patterns (such as slowly increasing, sudden, and periodic). This can be accomplished by using machine learning classification algorithms (such as random forests and support vector machines) combined with expert knowledge. The disturbance intensity can be graded (weak, medium, and strong) according to the disturbance amplitude, duration, and spatial range, forming low-frequency disturbance identification data that reflects the low-frequency disturbance characteristics in areas with obvious human activity, which can be used for subsequent ecological impact analysis and management decisions.

[0133] Preferably, performing ecosystem health assessment prediction based on the low-frequency disturbance identification data and the edge effect identification data in step S3 includes:

[0134] The low-frequency disturbance identification data and edge effect identification data were spatially semantically fused to construct an ecological disturbance comprehensive index layer. The ecological sensitivity weighted overlay analysis was performed on the ecological disturbance comprehensive index layer to generate multi-factor ecological vulnerability assessment data.

[0135] The multi-factor ecological vulnerability assessment data were divided into data sets to generate model training sets and model testing sets;

[0136] The model training set is trained using the long short-term memory neural network algorithm to generate an ecosystem health assessment pre-model; the ecosystem health assessment pre-model is optimized and iterated based on the model test set to generate an ecosystem health assessment model;

[0137] The multi-factor ecological vulnerability assessment data is input into the ecosystem health assessment model for health assessment prediction, generating ecosystem health assessment prediction data for the target area.

[0138] In an embodiment of the present invention, by reading low-frequency disturbance identification data (reflecting explicit human activity disturbance) and edge effect identification data (reflecting ecosystem edge sensitivity and diversity impact), the spatial coordinates of the two data sets are unified and the grids are aligned to ensure that the data between corresponding spatial units can be directly compared. A spatial semantic fusion method (such as weight-based multi-index fusion or fuzzy logic-based fusion method) is used to integrate the spatial information of the two data sources. A comprehensive index calculation model is designed, for example: Ieco = w1 × Dlowfreq + w2 × Eedge; where Ieco is the ecological disturbance comprehensive index, Dlowfreq is the low-frequency disturbance identification index, Eedge is the edge effect identification index, and w1 and w2 are weight coefficients (determined by expert scoring or data-driven). Generate an ecological disturbance comprehensive index layer covering the target area. Collect relevant ecological sensitivity factors, such as soil type, terrain slope, precipitation, vegetation coverage, etc. Assign weights to each sensitivity factor, combine with the ecological disturbance comprehensive index for superposition calculation, and form a multi-factor comprehensive evaluation. For example, the weighted linear superposition (WLA) model is used: Where V eco is the ecological vulnerability assessment value, F i is the value of the i-th ecological sensitivity factor, w i is the corresponding weight, w eco is the comprehensive index weight of ecological disturbance, I eco=Ecological disturbance comprehensive index. Generate multi-factor ecological vulnerability assessment data reflecting the ecological vulnerability of the target area. Compile the ecological vulnerability assessment data and corresponding ecosystem health assessment reference data (such as historical monitoring indicators and biodiversity indicators). Using random sampling, spatial stratified sampling, or time series partitioning, divide the dataset into a training set (e.g., 70%-80%) and a test set (e.g., 20%-30%). Standardize and normalize the data to ensure numerical stability during model training. Construct an LSTM network architecture, input the multi-factor ecological vulnerability assessment data, and output ecosystem health assessment prediction results. Set an appropriate number of network layers, units, and activation functions to prevent overfitting. Iteratively train the model using the training set, and evaluate prediction errors using a loss function (e.g., mean squared error (MSE)). Update network weights using an optimization algorithm (e.g., Adam). Evaluate model performance using the test set, and combine early stopping, cross-validation, and other techniques for model adjustment and hyperparameter optimization. Iteratively update model parameters to ultimately obtain an ecosystem health assessment model with stable performance and strong generalization capabilities. Input the multi-factor ecological vulnerability assessment data into the trained LSTM model. The model outputs ecosystem health assessment prediction data for the target area, quantifying the health of the ecosystem. The prediction results can include health level classifications (such as healthy, sub-healthy, sub-poor, and degraded) or continuous numerical indicators.

[0139] As an example of the present invention, refer to Figure 3 As shown, in this example, step S4 includes:

[0140] Step S41: Perform factor stratification extraction on the ecosystem health assessment prediction data to obtain data on potential latent interference factors in the region;

[0141] Step S42: performing a regional comprehensive analysis of the ecosystem in the target area based on the data of potential latent interference factors in the area to extract data on sensitive areas of latent threats, wherein the regional comprehensive analysis includes analysis of land use, vegetation continuity, and water body connectivity;

[0142] Step S43: Comparing the fluctuations of sensitive area data in time series, identifying the trend of implicit interference impact, generating a threat interference trend layer, and overlaying and analyzing the threat interference trend layer with the ecosystem health assessment prediction data to divide the implicit threat level intervals of different intensities;

[0143] Step S44: geographically plotting the latent threat level intervals and defining the distribution range, and generating a latent threat level and distribution map of the target area.

[0144] In an embodiment of the present invention, by inputting ecosystem health assessment prediction data, quantitative indicators of the current health status of the ecosystem in the region are included. Multi-factor hierarchical analysis technology is used to decompose the comprehensive prediction indicators into different ecological factor levels, such as potential hidden interference factors such as soil quality, groundwater conditions, and ecological connectivity. Principal component analysis (PCA), factor analysis (FA) or hierarchical clustering methods are used to identify hidden key influencing factors. A data set of potential hidden interference factors in the region is generated, and each factor is represented in the form of a spatial grid or vector unit to reflect its spatial distribution characteristics. Potential hidden interference factor data and basic spatial data such as regional land use, vegetation cover, and water body distribution are input. Types such as cultivated land, construction land, and forest land are identified through land use classification data, and the land use change rate and ecological interference potential are calculated. Remote sensing vegetation index data (such as NDVI) is used in combination with spatial connectivity indicators (such as Moran's index and connectivity index) to assess the degree of vegetation continuity and the risk of disruption. Based on water body remote sensing images and water system vector data, water body spatial connectivity indicators (such as water body connectivity index) are calculated to evaluate the supporting function of water resources for the ecology. A spatial layer of hidden threat-sensitive areas is generated through multi-factor spatial overlay and weighted scoring. Data on hidden threat-sensitive areas within the target region is extracted to identify potential ecosystem risk areas. Multi-temporal time series fluctuation analysis is conducted on the hidden threat-sensitive area data, using statistical fluctuation indicators (such as standard deviation and coefficient of variation) and time series analysis methods (such as sliding window trend detection and change point detection). Temporal trends and abnormal fluctuation points of hidden interference factors are identified, revealing the dynamic evolution of potential threats. Based on the trend analysis results, a spatially distributed threat interference trend layer is generated, showing the strength and direction of the hidden interference trend. The threat interference trend layer is overlaid with the ecosystem health assessment prediction data, and the spatial weights and multi-factor scoring system are combined to classify hidden threat levels into three levels (e.g., low, medium, and high). This hidden threat level range spatial layer is generated to provide a basis for subsequent spatial mapping. Using a GIS platform, vector processing is performed on the hidden threat level range layer to draw threat level boundaries. Spatial analysis tools such as buffer analysis and hotspot analysis are applied to help clarify the distribution of threat areas. The spatial extension of hidden threats is rationally defined based on ecosystem boundaries and natural geographical conditions. Generate a map product that includes the classification of hidden threat levels and their spatial distribution, marking the level and distribution range of hidden threats in the target area, and providing spatial decision-making support for ecological management and risk prevention and control.

[0145] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0146] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing identification method for ecosystem investigation, characterized in that: The following steps are involved: Step S1: Acquire multi-temporal remote sensing image data of the target area; extract the surface cover type of the multi-temporal remote sensing image data, and identify the explicit human activity area of ​​the multi-temporal remote sensing image data according to the surface cover type to generate explicit human activity area data; Step S2: Acquire nighttime light data of the target area; Perform spatiotemporal registration of the target area's nighttime light data and multi-temporal remote sensing image data to generate fused nighttime light remote sensing data; extract the boundary area of ​​the multi-temporal remote sensing image data based on the surface cover type to obtain edge area data; and calculate the vegetation index and nighttime light index of the edge area data based on the fused nighttime light remote sensing data. Step S3: Performing biological behavior correlation analysis on edge area data using vegetation index and night light index to generate edge effect identification data; analyzing the temporal changes of low-frequency disturbances in explicit human activity area data to generate low-frequency disturbance identification data; Conduct ecosystem health assessment and prediction based on low-frequency disturbance identification data and edge effect identification data to generate ecosystem health assessment prediction data for the target area; Step S4: Based on the ecosystem health assessment prediction data of the target area, the regional latent threat output is performed on the ecosystem of the target area, thereby obtaining the latent threat level and distribution map.

2. The remote sensing identification method for ecosystem investigation according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: performing multi-period sampling control of remote sensing images on the target area to obtain multi-temporal remote sensing image data; Step S12: performing radiation correction and geometric correction on the multi-temporal remote sensing image data to generate corrected multi-temporal remote sensing image data; Step S13: extracting the surface spectral features of the corrected multi-temporal remote sensing image data, and performing temporal change analysis and classification processing on the surface spectral features to generate surface cover type data; Step S14: interpreting the surface cover type data for human activity indicators to generate preliminary human activity mask data; performing spatial consistency screening and boundary correction on the preliminary human activity mask data to generate explicit human activity area data.

3. The remote sensing identification method for ecosystem investigation according to claim 2, characterized in that: Step S13 includes the following steps: Step S131: extracting multi-band reflectance values ​​of surface pixels at different time phases from the corrected multi-temporal remote sensing image data to obtain a surface spectral feature sequence with a time tag; Step S132: Calculate the temporal variation trend of the reflectance of key bands in the multi-temporal remote sensing image data based on the surface spectral feature sequence and determine whether its variation pattern meets any of the following conditions and mark it accordingly to generate temporal variation judgment mark data: the NDVI index has seasonal periodic fluctuations and the amplitude within the year is greater than 0.3; the position of the intersection of the blue light band and the red edge band reflectance varies by more than ±15nm in three consecutive temporal phases; the number of inflection points of the spectral curve differs by more than 2 in different growth periods; Step S133: Use the time series change judgment mark data to identify the cover type of the surface spectral characteristics. If any of the following conditions are met, it can be determined as a specified cover type and the surface cover type data is generated: if the maximum NDVI value exceeds 0.6 and the NDVI fluctuation range within the year is greater than 0.35, it is classified as "crop"; if the shortwave infrared reflectivity changes by less than 5% throughout the year and the NDVI is stable below 0.2 for a long time, it is classified as "building surface"; if the normalized difference index of red light and near infrared shows a double peak within the year, and the peak times correspond to spring and autumn respectively, it is classified as "deciduous woodland".

4. The remote sensing identification method for ecosystem investigation according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: Acquire nighttime light data of the target area; Step S22: performing radiation normalization and noise suppression on the raw radiation data of nighttime lights to generate net radiation data of nighttime lights; performing spatiotemporal reference point alignment on the net radiation data of nighttime lights and the multi-temporal remote sensing image correction data to generate a spatiotemporal registration parameter set; Step S23: performing grid-level registration and resampling fusion on the target area nighttime light data and the multi-temporal remote sensing image data according to the spatiotemporal registration parameter set to generate fused nighttime light remote sensing data; performing edge detection and morphological region growing analysis on the boundary change areas of the remote sensing image using the surface cover type data to generate a boundary area layer; Step S24: performing spatial extraction and segmentation on the boundary area layer to generate edge area data; performing vegetation index calculation on the edge area data based on the fused night light remote sensing data to generate edge area vegetation index data; Step S25: extracting the night light index of the edge area data based on the fused night light remote sensing data to generate the night light index data of the edge area.

5. The remote sensing identification method for ecosystem investigation according to claim 4, characterized in that: In step S23, edge detection and morphological region growing analysis are performed on the remote sensing image boundary change area using the surface cover type data, including: The edge of the remote sensing image boundary change area is extracted using the surface cover type data. The high threshold is set to 100 pixel gray value and the low threshold is set to 50 pixel gray value. A 3×3 Gaussian kernel is used for smoothing to obtain contour line data with clear boundaries and convert it into vector polygon format. The contour data is used as the seed area, and the region growing algorithm is used to expand to the area with similar features. The spectral similarity threshold is set to ±10% reflectivity, and the area threshold is greater than or equal to 1000m. 2 , the growth radius is within 90 meters, and the growth results are re-labeled and rasterized into the boundary area; The boundary area and the remote sensing image boundary change area are superimposed to generate a vector layer; the following attribute fields are assigned to each boundary area unit: change type, change area, boundary length, and change period, thereby outputting the boundary area layer.

6. The remote sensing identification method for ecosystem investigation according to claim 1, characterized in that: In step S3, the biological behavior correlation analysis of the edge area data using the vegetation index and the night light index includes: The edge area data is divided into spatial grids, and vegetation index changes are extracted by combining time series remote sensing images to generate edge vegetation dynamic index data; Overlay nighttime light remote sensing images on edge area data, extract nighttime light intensity and stability indicators, and generate edge nighttime light index data; Conduct bivariate spatiotemporal synergistic analysis of vegetation index and night light index to generate vegetation-night light coupling characteristic data; Clustering of spatiotemporal patterns of vegetation-nightlight coupling characteristic data to identify the coupling effects of human activity intervention and ecological disturbances, and generating biological behavior response layer data; Conduct edge sensitivity analysis and biodiversity impact assessment on biological behavior response layer data to generate edge effect identification data.

7. The remote sensing identification method for ecosystem investigation according to claim 1, characterized in that: The analysis of low-frequency disturbance time series changes in explicit human activity area data in step S3 includes: Perform long-term resampling and smoothing on the data of explicit human activity areas to extract stable change trends and generate low-frequency activity change baseline data; Extract remote sensing index sequences from multi-temporal remote sensing image data, perform wavelet decomposition, extract low-frequency disturbance principal components, and generate disturbance frequency band feature data; Perform time window sliding analysis on disturbance frequency band feature data using low-frequency activity change baseline data to identify low-frequency trend deviations and breakpoints, and generate low-frequency change anomaly detection data; The low-frequency change anomaly detection data and the explicit human activity area data are spatially constrained to be consistent, short-period high-frequency false interference is filtered out, and a steady-state disturbance response layer is generated; The disturbance pattern classification and intensity calibration are performed on the steady-state disturbance response layer to generate low-frequency disturbance identification data.

8. The remote sensing identification method for ecosystem investigation according to claim 1, characterized in that: In step S3, the ecosystem health assessment prediction based on the low-frequency disturbance identification data and the edge effect identification data includes: The low-frequency disturbance identification data and edge effect identification data were spatially semantically fused to construct an ecological disturbance comprehensive index layer. The ecological sensitivity weighted overlay analysis was performed on the ecological disturbance comprehensive index layer to generate multi-factor ecological vulnerability assessment data. The multi-factor ecological vulnerability assessment data were divided into data sets to generate model training sets and model testing sets; The model training set is trained using the long short-term memory neural network algorithm to generate an ecosystem health assessment pre-model; the ecosystem health assessment pre-model is optimized and iterated based on the model test set to generate an ecosystem health assessment model; The multi-factor ecological vulnerability assessment data is input into the ecosystem health assessment model for health assessment prediction, generating ecosystem health assessment prediction data for the target area.

9. The remote sensing identification method for ecosystem investigation according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing factor stratification extraction on the ecosystem health assessment prediction data to obtain data on potential latent interference factors in the region; Step S42: performing a regional comprehensive analysis of the ecosystem in the target area based on the data of potential latent interference factors in the area to extract data on sensitive areas of latent threats, wherein the regional comprehensive analysis includes analysis of land use, vegetation continuity, and water body connectivity; Step S43: Comparing the fluctuations of sensitive area data in time series, identifying the trend of implicit interference impact, generating a threat interference trend layer, and overlaying and analyzing the threat interference trend layer with the ecosystem health assessment prediction data to divide the implicit threat level intervals of different intensities; Step S44: geographically plotting the latent threat level intervals and defining the distribution range, and generating a latent threat level and distribution map of the target area.

10. A remote sensing identification system for ecosystem survey, characterized in that: The method for remote sensing identification applied to ecosystem survey according to claim 1 is used to execute the method, and the remote sensing identification system applied to ecosystem survey comprises: The multi-temporal remote sensing recognition module is used to obtain multi-temporal remote sensing image data of the target area; extract the surface cover type of the multi-temporal remote sensing image data, and identify the explicit human activity area based on the surface cover type to generate explicit human activity area data; The night light remote sensing recognition module is used to obtain night light data of the target area; perform spatiotemporal registration of the target area night light data and multi-temporal remote sensing image data to generate fused night light remote sensing data; extract the boundary area of ​​the multi-temporal remote sensing image data based on the surface cover type to obtain edge area data; and calculate the vegetation index and night light index of the edge area data based on the fused night light remote sensing data; The correlation analysis module is used to conduct biological behavior correlation analysis on edge area data through vegetation index and night light index to generate edge effect identification data; analyze the temporal changes of low-frequency disturbances in data of obvious human activity areas to generate low-frequency disturbance identification data; and perform ecosystem health assessment and prediction based on low-frequency disturbance identification data and edge effect identification data to generate ecosystem health assessment prediction data for the target area; The ecosystem assessment module is used to output regional hidden threats to the ecosystem of the target area based on the ecosystem health assessment prediction data of the target area, thereby obtaining the hidden threat level and distribution map.

Images