Ecologically sensitive area intelligent identification and evaluation system and method based on big data analysis

Abstract

The application discloses an ecological sensitive area intelligent identification and evaluation system and method based on big data analysis, and relates to the technical field of ecological environment monitoring and data processing; multi-source heterogeneous data such as remote sensing, meteorology, soil and human activities are acquired to construct a time-space ecological data cube; network and typical correlation analysis are used to extract high-dimensional features such as surface texture, time sequence evolution and soil-plant correlation; an improved isolation forest algorithm is used to automatically identify potential sensitive areas; a two-way coupling formula of ecological vulnerability and human activity pressure is constructed to calculate coupling coordination; future ecological trends are predicted to generate a dynamic sensitivity grading atlas; a two-dimensional decision matrix is constructed in combination with ecosystem service value to output differentiated protection suggestions; the application solves the problems of insufficient data fusion, lagging evaluation and missing quantitative human-earth relationship of traditional methods, and realizes accurate identification, trend early warning and scientific control of ecological sensitive areas.

Description

Technical Field

[0001] This invention belongs to the field of ecological environment monitoring and data processing technology, and specifically relates to an intelligent identification and assessment system and method for ecologically sensitive areas based on big data analysis. Background Technology

[0002] With the acceleration of urbanization and the intensification of human activities, the ecological environment is facing increasingly severe pressure. Ecologically sensitive areas, as key regions that are highly susceptible to human disturbance, prone to ecological degradation, and difficult to restore, require accurate identification and scientific assessment as a crucial foundation for land-use planning, ecological red line delineation, and environmental protection policy formulation. Traditional ecological assessments largely rely on manual field surveys, which are inefficient and have limited coverage, failing to meet the needs of large-scale, dynamic monitoring.

[0003] Currently, ecological sensitivity assessment based on remote sensing (RS) and geographic information systems (GIS) has become mainstream. Existing technologies typically employ a "factor-weighted overlay method," which selects single factors such as elevation, slope, and vegetation cover, determines weights using expert scoring (AHP), and then performs spatial overlay analysis. For example, in soil erosion sensitivity assessment, layers such as rainfall erosivity, soil erodibility, and topographic factors are often weighted and summed; if the total score exceeds a threshold, the area is identified as sensitive. In practical applications, this is often used for compiling static planning base maps, such as generating the base map for a city's "14th Five-Year Plan" for ecological and environmental protection. Furthermore, some studies have begun to introduce machine learning algorithms (such as random forests) to assist in determining weights, attempting to reduce the influence of human subjectivity.

[0004] This reveals the following problems with existing technologies: 1. Insufficient data fusion depth, often compressing time series data into a single mean, ignoring the dynamic evolution characteristics of ecological processes, and failing to effectively integrate soil micro-physicochemical data; 2. Lack of correlation mechanism mining, simple weighted superposition cannot reveal the complex nonlinear coupling relationship between natural background and human activities, leading to confusion between "high-sensitivity low-pressure" and "high-sensitivity high-pressure" regions; 3. Lagging assessment, only reflecting the state at the time of assessment, lacking the ability to predict future trends, and difficult to guide preventive protection. Summary of the Invention

[0005] (a) Technical problems to be solved In response to the problems in related technologies, this invention provides an intelligent identification and assessment system and method for ecologically sensitive areas based on big data analysis, so as to overcome the aforementioned technical problems existing in the existing related technologies.

[0006] (II) Technical Solution To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: S1. Acquire remote sensing image data, meteorological and hydrological monitoring data, ground soil sampling data and human activity trajectory data of the target area respectively, and perform spatiotemporal benchmark unification, noise reduction and interpolation reconstruction preprocessing to construct a spatiotemporal ecological data cube. S2. Based on the spatiotemporal ecological data cube, extract the spatial texture features of land cover and the temporal evolution features of environmental parameters, and mine the nonlinear correlation between features to generate a high-dimensional ecological background feature matrix. S3. Input the high-dimensional ecological background feature matrix into the improved isolated forest algorithm model, calculate the average path length and anomaly score of the data points, and generate a potential sensitive area mask file; S4. Extract the corresponding data of the area covered by the potential sensitive area mask file from the spatiotemporal ecological data cube, and calculate the coupling coordination degree between the ecological vulnerability index and the human activity pressure index to obtain the ecological coupling feature vector. S5. Using a long short-term memory network, the ecological coupling feature vector within the historical time window is temporally predicted to obtain the prediction result; based on the prediction result, dynamic classification is performed to generate a dynamic ecological sensitivity classification map. S6. Calculate the ecosystem service value distribution map of the target area; based on the dynamic ecological sensitivity grading map and the ecosystem service value distribution map, construct a two-dimensional decision matrix and output priority suggestions; This invention constructs a spatiotemporal ecological data cube by acquiring heterogeneous data from multiple sources, including remote sensing, meteorology, soil, and human activities. It mines features such as surface texture, temporal evolution, and soil-plant relationships. An improved isolated forest algorithm automatically identifies potential sensitive areas. By calculating coupling coordination degree, a genetically optimized LSTM network predicts future ecological trends, generating a dynamic sensitivity grading map. A two-dimensional decision matrix is ​​constructed based on ecosystem service value, outputting differentiated protection recommendations. This invention solves the problems of insufficient data fusion, lagging assessment, and lack of quantitative analysis of human-land relationships in traditional methods, achieving accurate identification, trend warning, and scientific management of ecologically sensitive areas.

[0007] Preferably, step S1 includes the following steps: S11. Use satellite remote sensing platform to acquire multispectral images, and calculate and generate vegetation index time series based on red light band and near-infrared band. Obtain rainfall and evaporation data from ground meteorological stations. Obtain soil organic matter content and soil erosion modulus data through IoT sensors and manual sampling. Obtain nighttime light data and road network density data to obtain the initial multi-source heterogeneous dataset. S12. The Savitzky-Golay filtering algorithm is used to smoothly reconstruct the vegetation index time series in the initial multi-source heterogeneous dataset and remove cloud and fog noise; the Kriging interpolation algorithm is used to map the discrete soil sampling data into continuous raster data to obtain a soil attribute raster map. S13. Establish a unified projection coordinate system and time reference axis, and resample and spatially register the multispectral images, reconstructed vegetation index, meteorological data, soil attribute raster map and nighttime light data to form a spatiotemporal ecological data cube containing time dimension, spatial dimension and attribute dimension. This invention addresses the challenges of remote sensing cloud and fog obstruction and sparse ground sampling points by acquiring data and utilizing SG filtering and Kriging interpolation techniques. Its advantages lie in breaking the limitations of traditional ecological assessment data sources that rely solely on remote sensing or ground surveys. It constructs a standardized data cube containing multi-dimensional information on time, space, and attributes, providing a high-quality, comprehensive data foundation for subsequent exploration of deeper ecological patterns.

[0008] Preferably, step S2 includes the following steps: S21. Input the image layer in the spatiotemporal ecological data cube into the residual network to extract spatial texture features; input the meteorological and soil layers in the spatiotemporal ecological data cube into the convolutional network to extract temporal evolution features; S22. Extract soil attribute data from the spatiotemporal ecological data cube as the first variable group, and extract vegetation index data as the second variable group. Calculate the canonical correlation coefficients of the first and second variable groups and their corresponding canonical variables to obtain the bio-geochemical association characteristics; S23. Spatial texture features, temporal evolution features and bio-geochemical correlation features are spliced ​​and normalized to construct a high-dimensional ecological background feature matrix; This invention achieves multi-perspective mining of ecological background characteristics by combining deep learning networks with statistical analysis. It not only extracts the macroscopic visual features of the land surface and the microscopic temporal features of the environment, but also innovatively introduces the CCA algorithm to quantify the deep coupling relationship between soil and vegetation. This enables the generated feature matrix to reflect the ecological feedback mechanism that is difficult to observe with the naked eye, providing a key basis for subsequent identification of hidden sensitive areas.

[0009] Preferably, step S3 includes the following steps: S31. Randomly extract a subset of samples from the high-dimensional ecological background feature matrix and construct multiple isolated trees; S32. Traverse each isolated tree in the multiple isolated trees, calculate the path length from the root node to the leaf node for each data point in the high-dimensional ecological background feature matrix, and take the average value to obtain the average path length. S33. Calculate the anomaly score based on the average path length; mark the data points with anomaly scores greater than the preset sensitivity threshold as anomalies, and obtain the anomaly point set; S34. Perform spatial clustering and morphological closing operations on all marked outliers in the outlier set to generate a binarized potential sensitive area mask file. This invention utilizes an improved isolated forest algorithm to automatically discover potential sensitive areas; it can quickly locate ecologically degraded areas by analyzing the anomalies in data distribution without relying on a large number of manually labeled negative samples; at the same time, it combines morphological processing to remove noise, significantly reducing the computational scope of subsequent complex models and achieving an efficiency improvement from flood irrigation to precision drip irrigation.

[0010] Preferably, step S4 includes the following steps: S41. Using the potential sensitive area mask file, perform mask extraction on the spatiotemporal ecological data cube to obtain the sensitive area dataset; S42. Based on the sensitive area dataset, the weights of soil erosion factor, topographic factor and habitat quality factor are calculated using principal component analysis to obtain the first dynamic weight set; the ecological vulnerability index is obtained by weighted summation of the first dynamic weight set. S43. Based on the sensitive area dataset, the weights of nighttime light intensity, impermeable surface ratio, and road network density are calculated using the analytic hierarchy process to obtain the second dynamic weight set; the human activity stress index is obtained by weighted summation of the second dynamic weight set. S44. Construct a capacity coupling formula; calculate the coupling degree between the ecological vulnerability index and the human activity pressure index using the capacity coupling formula; introduce a comprehensive evaluation index to correct the coupling degree and obtain the coupling coordination degree. S45. Combine the ecological vulnerability index, the human activity pressure index, and the coupling coordination degree to form an ecological coupling feature vector; This invention quantifies the complex feedback relationship between the natural background and human activities by constructing a vulnerability-pressure bidirectional coupling model. It overcomes the shortcomings of the traditional superposition method, which only performs simple summation, and accurately distinguishes between "high-sensitivity high-pressure (actual conflict zone)" and "high-sensitivity low-pressure (potential risk zone)" through coupling coordination degree, so that the assessment results can better reflect the true carrying capacity of the ecosystem and the level of risk it faces.

[0011] Preferably, step S5 includes the following steps: S51. Obtain the ecological coupling feature vector sequence at several historical time steps to obtain the historical ecological coupling feature vector sequence. S52. Construct a long short-term memory network prediction model; train the long short-term memory network prediction model and optimize the model parameters using the historical ecological coupling feature vector sequence to obtain the optimized long short-term memory network prediction model. S53. Input the current ecological coupling feature vector into the optimized long short-term memory network prediction model to predict the coupling coordination degree at future time steps, and obtain the predicted coupling coordination degree. S54. Statistical analysis of the predicted coupling coordination degree is performed using the natural breakpoint method to determine the grading threshold, and the target area is divided into different levels to generate a dynamic ecological sensitivity grading map. This invention achieves a forward-looking assessment of ecological sensitivity through LSTM time-series prediction and dynamic classification; it breaks through the limitations of traditional methods that can only assess the "past" or "present" state, and can provide early warning of potential ecological degradation areas, reserving a valuable intervention window for management departments and achieving the effect of preventing problems before they occur.

[0012] Preferably, step S52 includes the following steps: S521. Divide the historical ecological coupling feature vector sequence into training data and test data; use the training data to train the long short-term memory network prediction model, and adjust the learning rate of the initial long short-term memory network prediction model by manual parameter tuning to obtain the trained long short-term memory network prediction model. S522. Set the test accuracy threshold; input the test data into the trained long short-term memory network prediction model to obtain the training model prediction data, calculate the similarity between the training model prediction data and the label data in the test data, and obtain the test accuracy; determine whether the test accuracy is ≥ the test accuracy threshold. If it is greater, then the trained long short-term memory network prediction model is used as the optimized long short-term memory network prediction model. Otherwise, the parameters of the trained Long Short-Term Memory Network (LSTM) prediction model are optimized using a genetic algorithm to obtain the optimal solution; the optimal solution is then used as the parameters of the trained LSTM prediction model to obtain the optimized LSTM prediction model. This invention introduces a genetic algorithm to adaptively optimize the hyperparameters of the LSTM model, solving the problems of low efficiency and easy getting trapped in local optima in manual parameter tuning (such as learning rate and number of hidden layer nodes). It ensures the generalization ability and prediction accuracy of the prediction model in different ecological scenarios, thereby improving the reliability of early warning of future ecological risks.

[0013] Preferably, the optimal solution obtained by optimizing the parameters of the long short-term memory network prediction model trained by the genetic algorithm in step S522 includes the following steps: S5221. Construct a chromosome population; each chromosome in the chromosome population serves as a parameter for a randomly trained Long Short-Term Memory (LSTM) network prediction model; set the maximum number of optimization iterations. S5222. By maximizing the test accuracy, select, crossover, and mutate chromosomes in the chromosome population; repeat the iteration until the set maximum number of iterations is reached or the test accuracy is greater than or equal to the test accuracy threshold, and obtain the optimal solution. This invention uses the execution logic of a genetic algorithm; by simulating the biological evolution mechanism, it quickly finds the optimal parameter combination that minimizes the model's prediction error in the global search space, avoiding the computational redundancy of the traditional grid search method, and significantly improving the model training efficiency and the final prediction accuracy.

[0014] Preferably, step S6 includes the following steps: S61. Calculate the water conservation and soil retention of the target area based on the InVEST model, and generate an ecosystem service value distribution map. S62. Spatially overlay the dynamic ecological sensitivity grading map with the ecosystem service value distribution map to construct a two-dimensional decision matrix that includes sensitivity and value; S63. Based on the quadrant position of the two-dimensional decision matrix, match the preset strategy library and output protection priority suggestion instructions; the instructions include level one red line protection instructions, level two ecological restoration instructions, level three restricted development instructions and level four moderate utilization instructions; This invention achieves differentiated and precise control by constructing a two-dimensional decision matrix of "sensitivity-value". It avoids a one-size-fits-all protection model, protects high-value and high-risk core areas, and leaves room for the reasonable development of low-risk areas, thus achieving a scientific balance between ecological protection and economic development.

[0015] Preferably, a smart identification and assessment system for ecologically sensitive areas based on big data analysis is provided to implement the aforementioned smart identification and assessment method for ecologically sensitive areas based on big data analysis; the system includes: The multi-source data sensing and reconstruction module is responsible for connecting to satellite data interfaces and IoT gateways, performing Savitzky-Golay filtering and Kriging interpolation operations, and outputting a spatiotemporal ecological data cube. The deep feature mining module, with its built-in CNN network structure and CCA algorithm library, is responsible for extracting high-dimensional features from the cube. The automatic sensitive area identification module runs the isolated forest algorithm and outputs a mask of potential sensitive areas. The coupling evaluation and prediction module integrates the physical coupling model and the LSTM prediction unit, and is responsible for calculating the coupling coordination degree and predicting future trends. The hierarchical decision output module stores the decision rule base and outputs the final protection strategy instructions based on the InVEST model results.

[0016] (III) Beneficial Effects The present invention has the following beneficial effects: This invention achieves deep fusion and latent feature mining of multi-source heterogeneous data; it breaks through the limitations of traditional methods that rely solely on remote sensing macroscopic images, integrating meteorological, soil microphysicochemical properties, and human activity trajectory data; through network extraction and canonical correlation analysis, it not only extracts explicit spatial texture features, but also mines the temporal evolution features of environmental parameters and the deep nonlinear correlation between soil and vegetation; this enables the evaluation model to capture ecological feedback mechanisms that are difficult to observe with the naked eye, significantly improving the accuracy of identifying latent ecologically sensitive areas.

[0017] This invention constructs a dynamic coupling prediction model, achieving a leap from "static assessment" to "trend early warning." It abandons the traditional static weighted superposition method, innovatively constructing a two-way coupling formula of "ecological vulnerability - human activity pressure," and introduces the concept of physical capacity coupling to calculate the coupling coordination degree, accurately distinguishing between existing conflict zones and potential risk zones. By introducing a long short-term memory network optimized by a genetic algorithm, it uses historical sequences to predict future ecological states in a temporal sequence. This enables managers to anticipate ecological deterioration trends within future time windows, thereby transforming conservation efforts from passive post-event management to proactive pre-event prevention.

[0018] This invention establishes a two-dimensional decision-making mechanism based on "sensitivity-value," providing differentiated and precise management strategies. Combining the ecosystem service value calculated using the InVEST model, this invention constructs a two-dimensional decision matrix, avoiding the traditional one-size-fits-all protection approach. The system can automatically match and output differentiated instructions based on the region's sensitivity level and ecological value, including Level 1 red line protection, Level 2 ecological restoration, Level 3 restricted development, or Level 4 moderate utilization. This refined decision support effectively protects core ecological assets while reserving reasonable space for the green development of the regional economy, achieving synergistic optimization of ecological and socio-economic benefits.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to the present invention. Figure 2This is a schematic diagram of the modules of the intelligent identification and assessment system for ecologically sensitive areas based on big data analysis of the present invention. Figure 3 This is a flowchart illustrating the ecological coupling feature vector of the intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to the present invention. Figure 4 This is a schematic diagram of the process for obtaining the predicted coupling coordination degree in the intelligent identification and assessment method for ecologically sensitive areas based on big data analysis of the present invention. Figure 5 This is a schematic diagram of the process for obtaining a dynamic ecological sensitivity classification map in the intelligent identification and assessment method for ecologically sensitive areas based on big data analysis of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0023] Please see Figure 1 This invention discloses an intelligent identification and assessment method for ecologically sensitive areas based on big data analysis, comprising the following steps: S1. Acquire remote sensing image data, meteorological and hydrological monitoring data, ground soil sampling data and human activity trajectory data of the target area respectively, and perform spatiotemporal benchmark unification, noise reduction and interpolation reconstruction preprocessing to construct a spatiotemporal ecological data cube. The above embodiment S1 includes the following steps: S11. Use satellite remote sensing platform to acquire multispectral images, and calculate and generate vegetation index time series based on red light band and near-infrared band. Obtain rainfall and evaporation data from ground meteorological stations. Obtain soil organic matter content and soil erosion modulus data through IoT sensors and manual sampling. Obtain nighttime light data and road network density data to obtain the initial multi-source heterogeneous dataset. In specific implementation, the above embodiment S11 is as follows: This embodiment selects an ecological barrier area (approximately 2000 square kilometers) in a certain watershed as the target area; acquires Landsat-8 OLI satellite imagery with a time resolution of 16 days to obtain remote sensing data; accesses data from 15 automatic weather stations distributed within the area to collect daily rainfall (mm), daily average temperature (°C), and wind speed (m / s) to obtain meteorological data; acquires real-time soil moisture content through 50 deployed LoRaWAN soil moisture sensors; combines annual manual field sampling to obtain soil organic matter content (%), soil texture (ratio of sand / silt / clay), and K factor (soil erodibility factor) from 300 sampling points to obtain soil data; acquires NPP-VIIRS monthly composite nighttime light products (500m resolution) and OpenStreetMap road network vector data to obtain human activity data; the time span of all data is the past 5 years; S12. The Savitzky-Golay filtering algorithm is used to smoothly reconstruct the vegetation index time series in the initial multi-source heterogeneous dataset and remove cloud and fog noise; the Kriging interpolation algorithm is used to map the discrete soil sampling data into continuous raster data to obtain a soil attribute raster map. In specific implementation, the above embodiment S12 is as follows: First, vegetation index reconstruction is performed by calculating the normalized vegetation index NDVI = (NIR − Red) / (NIR + Red); where NDVI is used to reflect vegetation greenness and coverage; NIR represents the reflectance of the near-infrared band; Red represents the reflectance of the red band; due to cloud cover, the original NDVI time series curve has a large number of abrupt low values ​​(noise); noise is processed by Savitzky-Golay filter; after processing, the abnormal drop in NDVI of a certain forest pixel from 0.7 to 0.1 caused by rainy clouds in June is corrected to 0.68, restoring the true vegetation growth trend; Further soil data interpolation was performed; for 300 discrete soil sampling points, the ordinary kriging algorithm was used to fit the spherical model parameters (nuclear value, sill value, range); the fitted spherical model parameters were used to estimate the unsampled area and generate a soil property raster map with a resolution of 30m×30m (including organic matter distribution map and K factor distribution map), realizing the data expansion from point to surface; S13. Establish a unified projection coordinate system and time reference axis, and resample and spatially register the multispectral images, reconstructed vegetation index, meteorological data, soil attribute raster map and nighttime light data to form a spatiotemporal ecological data cube containing time dimension, spatial dimension and attribute dimension. In specific implementation, the above embodiment S13 specifically involves: uniformly adopting WGS84 UTM projection; using the 30m resolution of Landsat imagery as a benchmark, bilinear interpolation resampling of the 500m resolution nighttime light data and interpolated meteorological data to 30m; on the time axis, aggregating all data into monthly values ​​to obtain a spatiotemporal ecological data cube; the spatiotemporal ecological data cube is a four-dimensional tensor D∈R T×H×W×C Where R indicates that the tensor elements are real numbers, T=60 (time step), H and W are the number of rows and columns of the regional raster, respectively, and C=10 (representing the number of channels, including red band, NIR band, NDVI, rainfall, temperature, soil organic matter, soil K factor, nighttime lights, road network distance, and impervious surfaces); the spatiotemporal ecological data cube provides unified data for all subsequent analyses; S2. Based on the spatiotemporal ecological data cube, extract the spatial texture features of land cover and the temporal evolution features of environmental parameters, and mine the nonlinear correlation between features to generate a high-dimensional ecological background feature matrix. The above embodiment S2 includes the following steps: S21. Input the image layer in the spatiotemporal ecological data cube into the residual network to extract spatial texture features; input the meteorological and soil layers in the spatiotemporal ecological data cube into the convolutional network to extract temporal evolution features; In specific implementation, the above embodiment S21 specifically involves: extracting the latest remote sensing image slice from the cube and inputting it into the ResNet-18 network; extracting the spatial texture feature vector F reflecting surface fragmentation and connectivity. spa ∈R 64 This feature can distinguish between "whole forest" (uniform texture) and "fragmented forest patches" (chaotic texture). The ResNet-18 in this embodiment is an 18-layer residual network. Its core structure includes: an initial 7x7 convolutional layer (stride 2) and a max-pooling layer for initial downsampling; followed by four residual module groups (each containing two residual blocks with 64, 128, 256, and 512 channels respectively), using skip connections to address the vanishing gradient problem; finally, a global average pooling layer outputs a 512-dimensional feature vector (adjusted to 64-dimensional spatial texture features via a fully connected layer). Historical meteorological (rainfall, temperature) and soil moisture sequences (length 60) for corresponding pixel locations are extracted and input into a 3-layer 1D-CNN network. The 1D-CNN network has a convolution kernel size of 3, and extracts the temporal evolution feature vector F reflecting the periodic fluctuations of the environment. tem ∈R 32 This feature can capture the trend of "reduced rainfall leading to a lag in soil moisture reduction"; S22. Extract soil attribute data from the spatiotemporal ecological data cube as the first variable group, and extract vegetation index data as the second variable group. Calculate the canonical correlation coefficients of the first and second variable groups and their corresponding canonical variables to obtain the bio-geochemical association characteristics; In specific implementation, the above embodiment S22 is as follows: Ecological sensitivity often stems from an imbalance in the soil-plant relationship; the first variable group X = [soil organic matter, soil pH, soil K factor] and the second variable group Y = [NDVI, EVI, LAI] are defined; where EVI represents the ecological vulnerability index, used to correct for atmospheric and soil background influences, and is suitable for high biomass areas; LAI represents the leaf area index, which is the proportion of total plant leaf area per unit surface area, characterizing the vegetation canopy structure; canonical correlation analysis (CCA) is used to find projection vectors a and b such that the canonical variable U = a T X and V = b T The correlation coefficient ρ of Y is the largest; the largest ρ value obtained by calculation is the bio-geochemical correlation characteristic; if ρ>0.8 in a certain area, it indicates that the vegetation growth in that area is extremely dependent on soil conditions (such as karst areas), and once the soil is damaged, the vegetation will degrade irreversibly, which belongs to a highly sensitive mechanism; if ρ<0.3, it indicates that the vegetation growth is mainly controlled by climate and has little soil restriction; this characteristic reveals a deep ecological driving mechanism. S23. Spatial texture features, temporal evolution features and bio-geochemical correlation features are spliced ​​and normalized to construct a high-dimensional ecological background feature matrix; In specific implementation, the above embodiment S23 specifically involves: calculating all pixels in the entire target area and concatenating the above features to obtain a high-dimensional ecological background feature matrix M∈R. N×M Where N is the total number of pixels and M is the total dimension of the features (in this embodiment, M = 64 + 32 + 1 = 97 dimensions). S3. Input the high-dimensional ecological background feature matrix into the improved isolated forest algorithm model, calculate the average path length and anomaly score of the data points, and generate a potential sensitive area mask file; The above embodiment S3 includes the following steps: S31. Randomly extract a subset of samples from the high-dimensional ecological background feature matrix and construct multiple isolated trees; S32. Traverse each isolated tree in the multiple isolated trees, calculate the path length from the root node to the leaf node for each data point in the high-dimensional ecological background feature matrix, and take the average value to obtain the average path length. S33. Calculate the anomaly score based on the average path length; mark the data points with anomaly scores greater than the preset sensitivity threshold as anomalies, and obtain the anomaly point set; In specific implementation, the above embodiments S31-S33 are specifically: using isolated forests for unsupervised anomaly detection; in this embodiment, most of the area is stable forest or farmland (background data), while ecologically sensitive areas (such as areas with severe soil erosion or mining development areas) appear as outliers in the feature space; Construct 100 isolated trees; for a point in the feature matrix x i ,like x i The features (such as low NDVI, fragmented texture, and extremely high soil-vegetation correlation) differ greatly from the background and will be quickly cut and separated during tree construction, resulting in... x i Path length h( x i The score is very short; the anomaly score formula is: S( x i , n =2−E(h(x)) / c(n); where n represents the subsample size of the training isolated tree; E(h(x)) represents the average value of h(x) across all trees in the isolated forest; c(n) represents the normalization factor for the average path length; the calculation results in this embodiment show that the score of the stable forest area is about 0.3, while the score of a certain landslide hazard area is as high as 0.85; the preset sensitivity threshold in this embodiment is 0.7, and all pixels with abnormal scores >0.7 are marked as "1" (abnormal), and the rest are "0"; S34. Perform spatial clustering and morphological closing operations on all marked outliers in the outlier set to generate a binarized potential sensitive area mask file. In specific implementation, the above embodiment S34 is as follows: The original identification result (set of anomalies) in S33 may contain sporadic noise; therefore, morphological "closing operation" (dilation followed by erosion) is used to connect adjacent anomalies, fill internal holes, and remove isolated noise with an area of ​​less than 5 pixels; the final generated potential sensitive area mask file accurately delineates the areas that need further key evaluation (in this embodiment, the areas that need key evaluation account for about 18% of the total area), and this embodiment S34 significantly reduces the computational load of the subsequent S4 steps; S4. Extract the corresponding data of the area covered by the potential sensitive area mask file from the spatiotemporal ecological data cube, and calculate the coupling coordination degree between the ecological vulnerability index and the human activity pressure index to obtain the ecological coupling feature vector. Please see Figure 3 The above embodiment S4 includes the following steps: S41. Using the potential sensitive area mask file, perform mask extraction on the spatiotemporal ecological data cube to obtain the sensitive area dataset; S42. Based on the sensitive area dataset, the weights of soil erosion factor, topographic factor and habitat quality factor are calculated using principal component analysis to obtain the first dynamic weight set; the ecological vulnerability index is obtained by weighted summation of the first dynamic weight set. In specific implementation, the above embodiment S42 is as follows: calculations are performed only for the area within the Mask; selected indicators include: soil erodibility K (from the interpolation results of S12), topographic relief LS (from DEM), and habitat quality HQ (the habitat quality is the ability of ecological environment quality to support biodiversity); weights are determined using PCA (by constructing the covariance matrix of the indicator data, calculating eigenvalues ​​and eigenvectors, selecting principal components with a cumulative variance contribution rate greater than 85%, and normalizing the eigenvectors as weights); the ecological vulnerability index EVI = 0.45 × K + 0.35 × LS + 0.20 × (1 − HQ); the higher the EVI value, the more fragile the natural background and the worse the resistance to interference; S43. Based on the sensitive area dataset, the weights of nighttime light intensity, impermeable surface ratio, and road network density are calculated using the analytic hierarchy process to obtain the second dynamic weight set; the human activity stress index is obtained by weighted summation of the second dynamic weight set. In specific implementation, the above embodiment S43 specifically involves: selecting indicators: Normalized Nighttime Light Level (NTL), Impermeable Surface Area Ratio (ISA), and Distance from Main Road (Dist); Human Activity Pressure Index (HAI) = 0.5 × NTL + 0.3 × ISA + 0.2 × (1 / ISA); the higher the HAI value, the greater the intensity of external human interference; the specific steps for calculating the weights using the Analytic Hierarchy Process (AHP) are as follows: In this embodiment, a discriminant matrix is ​​constructed, and five ecological experts score the pairwise importance of the three indicators, calculate the maximum eigenvalue and its corresponding eigenvector, and through consistency testing, finally determine the weight vector as [0.5, 0.3, 0.2]. S44. Construct a capacity coupling formula; calculate the coupling degree between the ecological vulnerability index and the human activity pressure index using the capacity coupling formula; introduce a comprehensive evaluation index to correct the coupling degree and obtain the coupling coordination degree. In specific implementation, the above embodiment S44 is as follows: Further, in S42 and S43, a high EVI or HAI alone does not necessarily indicate sensitivity; the interaction (coupling) between the two is key. The coupling degree C is calculated using the concept of capacity coupling in physics. When both EVI and HAI are high, the system is in a "high-level resonance" state, which is extremely prone to collapse. When EVI is high but HAI is low, the system is in a "potential risk" state. To distinguish between these two states, a coupling coordination degree D is introduced. D is calculated using the capacity coupling formula: D = C × T. T = α × EVI + β × HAI; where D is the coupling coordination degree, C is the coupling degree, T is the comprehensive evaluation index; k is the adjustment coefficient, and α and β are undetermined weight coefficients; in this embodiment, k = 3 (to enhance the distinguishability). In this embodiment, α and β are set to 0.6 and 0.4 respectively, emphasizing ecological protection. In specific implementation, α and β can be dynamically set according to the needs of emphasizing ecological protection and economic development; the calculated values ​​for a certain mining area are EVI = 0.8, HAI = 0.8, further C ≈ 1, T ≈ 0.9 (extremely imbalanced); and for a certain deep mountain forest area, EVI = 0.8, HAI = 0.1, further C ≈ 0.3, T ≈ 0.4 (potentially vulnerable). S45. Combine the ecological vulnerability index, the human activity pressure index, and the coupling coordination degree to form an ecological coupling feature vector; In specific implementation, the above embodiment S45 specifically involves: generating an ecological coupling feature vector V for each pixel within the Mask. couple =[EVI,HAI,D]; Ecological coupling feature vector, which encapsulates the current physical examination report of this area; S5. Using a long short-term memory network, the ecological coupling feature vector within the historical time window is temporally predicted to obtain the prediction result; based on the prediction result, dynamic classification is performed to generate a dynamic ecological sensitivity classification map. Please see Figure 4 The above embodiment S5 includes the following steps: S51. Obtain the ecological coupling feature vector sequence at several historical time steps to obtain the historical ecological coupling feature vector sequence. S52. Construct a long short-term memory network prediction model; train the long short-term memory network prediction model and optimize the model parameters using the historical ecological coupling feature vector sequence to obtain the optimized long short-term memory network prediction model. The above embodiment S52 includes the following steps: S521. Divide the historical ecological coupling feature vector sequence into training data and test data; The long short-term memory network prediction model is trained using training data, and the learning rate of the initial long short-term memory network prediction model is adjusted by manual parameter tuning to obtain the trained long short-term memory network prediction model. S522. Set the test accuracy threshold; input the test data into the trained long short-term memory network prediction model to obtain the training model prediction data, calculate the similarity between the training model prediction data and the label data in the test data, and obtain the test accuracy; determine whether the test accuracy is ≥ the test accuracy threshold. If it is greater, then the trained long short-term memory network prediction model is used as the optimized long short-term memory network prediction model. Otherwise, the parameters of the trained Long Short-Term Memory Network (LSTM) prediction model are optimized using a genetic algorithm to obtain the optimal solution; the optimal solution is then used as the parameters of the trained LSTM prediction model to obtain the optimized LSTM prediction model. In the above embodiment S522, optimizing the parameters of the long short-term memory network prediction model trained by a genetic algorithm to obtain the optimal solution includes the following steps: S5221. Construct a chromosome population; each chromosome in the chromosome population serves as a parameter for a randomly trained Long Short-Term Memory (LSTM) network prediction model; set the maximum number of optimization iterations. S5222. By maximizing the test accuracy, select, crossover, and mutate chromosomes in the chromosome population; repeat the iteration until the set maximum number of iterations is reached or the test accuracy is greater than or equal to the test accuracy threshold, and obtain the optimal solution. In specific implementation; 5221 and 5222 above specifically refer to: constructing a chromosome population, setting the size of the chromosome population as... r This embodiment sets r If the value is 60, then the chromosome population is represented as follows: ,in, R i Represents the 1st chromosome in the population. Each chromosome in the chromosome population is used as a parameter combination of a randomly trained long short-term memory network prediction model within the search space; the maximum number of optimization iterations is set to 300. The process begins with iteration. In each iteration, the similarity between the model's predicted data and the labeled data in the test data is calculated to obtain the fitness value of each chromosome in the chromosome population, resulting in a set of chromosome fitness values. Based on the chromosome fitness values ​​in the set, chromosomes are selected from the chromosome population in descending order of fitness value, resulting in a selected chromosome population. Crossover and mutation operations are performed on the chromosomes in the selected chromosome population to obtain the manipulated chromosome population. In each iteration, the current best combination is selected based on the chromosome performance values ​​in the set of chromosome performance values. The crossover is calculated using the mean of the parent values, and the mutation is Gaussian mutation. Repeat the iteration until the maximum number of iterations is reached or the test accuracy is greater than or equal to the test accuracy threshold, then stop the iteration and take the current best combination of parameters as the optimal solution. After optimization using a genetic algorithm, the final LSTM model parameters determined in this embodiment are as follows: 64 LSTM units in the first layer and a Dropout ratio of 0.2; 32 LSTM units in the second layer; a learning rate of 0.005; and the Adam optimizer. This parameter combination achieved a prediction accuracy of 0.92 on the test set, with the root mean square error reduced to below 0.05, demonstrating its ability to accurately fit the temporal fluctuations of ecological indicators. S53. Input the current ecological coupling feature vector into the optimized long short-term memory network prediction model to predict the coupling coordination degree at future time steps, and obtain the predicted coupling coordination degree. In specific implementation, S51-S53 of the above embodiments are as follows: Ecological sensitivity is dynamically changing; therefore, continuous prediction is required; using the historical data of the cube in S13, the ecological coupling feature vector sequence of the past 60 months is calculated; a long short-term memory network prediction model is constructed, with the vector sequence from the past year to the current time as the input and the coupling coordination degree D for the next six months as the output. pred ; In this embodiment, although the current D value of a certain region is within a reasonable range, the HAI has been rising from 0.2 to 0.5 over the past year. The LSTM predicts that the future D value will deteriorate significantly, thus achieving a leap from static assessment to trend warning. The labeled data used in the training phase are the actual calculated D values ​​at historical moments, that is, the coupling coordination degree calculated using historical ecological data. S54. Statistical analysis of the predicted coupling coordination degree is performed using the natural breakpoint method to determine the grading threshold, and the target area is divided into different levels to generate a dynamic ecological sensitivity grading map. For specific implementation details, please refer to [link / reference]. Figure 5 Specifically, in the above embodiment S54, the predicted global domain D is processed as follows: pred The values ​​were statistically analyzed; Jenks' natural breakpoint method was used to find natural breakpoints in the data distribution, minimizing within-class variance and maximizing between-class variance; in this embodiment, the breakpoints were determined to be: 0.35, 0.55, 0.75, and further divided into levels: 0 <D pred ≤0.35: Low sensitivity (green); 0.35 <D pred ≤0.55: Moderately sensitive (yellow); 0.55 <D pred ≤0.75: Highly sensitive (orange); D pred >0.75: Extremely sensitive (red); Generate a dynamic ecological sensitivity grading map, which not only reflects the current situation but also includes predictions of future trends; S6. Calculate the ecosystem service value distribution map of the target area; based on the dynamic ecological sensitivity grading map and the ecosystem service value distribution map, construct a two-dimensional decision matrix and output priority suggestions; The above embodiment S6 includes the following steps: S61. Calculate the water conservation and soil retention of the target area based on the InVEST model, and generate an ecosystem service value distribution map. In specific implementation, the above embodiment S61 is as follows: First, data preprocessing is performed, the LULC layer is reclassified into an InVEST model recognizable code, the evapotranspiration coefficient and root depth in the biophysical parameter table are corrected according to local literature, and the rainfall erosivity factor is calculated using monthly rainfall data. The rainfall and soil data obtained from S1 and the preprocessed LULC data are input into the InVEST model to calculate the water conservation and soil retention of each pixel and obtain the ecosystem service value. The output results are normalized by range and divided into three levels (high, medium and low) according to the quantile method to generate an ecosystem service value distribution map.

[0024] S62. Spatially overlay the dynamic ecological sensitivity grading map with the ecosystem service value distribution map to construct a two-dimensional decision matrix that includes sensitivity and value; S63. Based on the quadrant position of the two-dimensional decision matrix, match the preset strategy library and output protection priority suggestion instructions; the instructions include level one red line protection instructions, level two ecological restoration instructions, level three restricted development instructions and level four moderate utilization instructions; In specific implementation, the above embodiments S62-S63 are as follows: the map (Risk) of S54 and the map of S61 are spatially superimposed to construct the decision logic: Scenario 1: A certain area is located in the "extremely sensitive area (red)" and the ESV is "high value (water source)"; it is determined that the risk is high and the value is great, belonging to the core protection object; the output instruction (first-level red line protection) is "immediately delineate it into the ecological red line, implement closed management, prohibit all human development, and deploy high-frequency monitoring equipment"; Scenario 2: A certain area is located in a "highly sensitive area (orange)" but its ESV is "low value (abandoned mine)". It is classified as sensitive because of severe damage, but currently has no service value. Output instruction (secondary ecological restoration): "Implement artificial assisted regeneration projects (such as topsoil spraying), focus on controlling soil erosion, and downgrade management after ecological functions are restored". Scenario 3: A certain area is located in the "low-sensitivity zone (green)" and the ESV is "medium value (farmland)"; the system is determined to be stable and has production function; output instruction (level 4 moderate utilization): "Green agriculture and ecotourism are allowed, the amount of chemical fertilizers and pesticides used is restricted, and soil quality is monitored regularly"; Example 2:

[0025] For further details, please refer to Figure 2 A smart identification and assessment system for ecologically sensitive areas based on big data analysis is provided to implement the aforementioned smart identification and assessment method for ecologically sensitive areas based on big data analysis. The system includes: The multi-source data sensing and reconstruction module is responsible for connecting to satellite data interfaces and IoT gateways, performing Savitzky-Golay filtering and Kriging interpolation operations, and outputting a spatiotemporal ecological data cube. The deep feature mining module, with its built-in CNN network structure and CCA algorithm library, is responsible for extracting high-dimensional features from the cube. The automatic sensitive area identification module runs the isolated forest algorithm and outputs a mask of potential sensitive areas. The coupling evaluation and prediction module integrates the physical coupling model and the LSTM prediction unit, and is responsible for calculating the coupling coordination degree and predicting future trends. The hierarchical decision output module stores the decision rule base and outputs the final protection strategy instructions based on the InVEST model results.

[0026] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0027] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for intelligent identification and assessment of ecologically sensitive areas based on big data analysis, characterized in that, Includes the following steps: S1. Acquire remote sensing image data, meteorological and hydrological monitoring data, ground soil sampling data and human activity trajectory data of the target area respectively, and perform spatiotemporal benchmark unification, noise reduction and interpolation reconstruction preprocessing to construct a spatiotemporal ecological data cube. S2. Based on the spatiotemporal ecological data cube, extract the spatial texture features of land cover and the temporal evolution features of environmental parameters, and mine the nonlinear correlations between features to generate a high-dimensional ecological background feature matrix; the nonlinear correlations are specifically the bio-geochemical correlation features between soil properties and vegetation indices. S3. Input the high-dimensional ecological background feature matrix into the improved isolated forest algorithm model, calculate the average path length and anomaly score of the data points, and generate a potential sensitive area mask file; S4. Extract the corresponding data of the area covered by the potential sensitive area mask file from the spatiotemporal ecological data cube, and calculate the coupling coordination degree between the ecological vulnerability index and the human activity pressure index to obtain the ecological coupling feature vector formed by the combination of the ecological vulnerability index, the human activity pressure index and the coupling coordination degree. S5. Using a long short-term memory network, the ecological coupling feature vector within the historical time window is temporally predicted to obtain the prediction result; based on the prediction result, dynamic classification is performed to generate a dynamic ecological sensitivity classification map. S6. Calculate the distribution map of ecosystem service value in the target area; based on the dynamic ecological sensitivity grading map and the distribution map of ecosystem service value, construct a two-dimensional decision matrix and output priority suggestions.

2. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 1, characterized in that, S1 includes the following steps: S11. Use satellite remote sensing platform to acquire multispectral images, and calculate and generate vegetation index time series based on red light band and near-infrared band. Obtain rainfall and evaporation data from ground meteorological stations. Obtain soil organic matter content and soil erosion modulus data through IoT sensors and manual sampling. Obtain nighttime light data and road network density data to obtain the initial multi-source heterogeneous dataset. S12. The Savitzky-Golay filtering algorithm is used to smoothly reconstruct the vegetation index time series in the initial multi-source heterogeneous dataset and remove cloud and fog noise; the Kriging interpolation algorithm is used to map the discrete soil sampling data into continuous raster data to obtain a soil attribute raster map. S13. Establish a unified projection coordinate system and time reference axis, and resample and spatially register multispectral images, reconstructed vegetation indices, meteorological data, soil attribute raster maps and nighttime light data to form a spatiotemporal ecological data cube containing time dimension, spatial dimension and attribute dimension.

3. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 2, characterized in that, S2 includes the following steps: S21. Input the image layer in the spatiotemporal ecological data cube into the residual network to extract spatial texture features; input the meteorological and soil layers in the spatiotemporal ecological data cube into the convolutional network to extract temporal evolution features; S22. Extract soil attribute data from the spatiotemporal ecological data cube as the first variable group, and extract vegetation index data as the second variable group; Calculate the canonical correlation coefficients of the first and second variable groups and their corresponding canonical variables to obtain the bio-geochemical association characteristics; S23. Spatial texture features, temporal evolution features and bio-geochemical correlation features are spliced ​​and normalized to construct a high-dimensional ecological background feature matrix.

4. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 3, characterized in that, S3 includes the following steps: S31. Randomly extract a subset of samples from the high-dimensional ecological background feature matrix and construct multiple isolated trees; S32. Traverse each isolated tree in the multiple isolated trees, calculate the path length from the root node to the leaf node for each data point in the high-dimensional ecological background feature matrix, and take the average value to obtain the average path length. S33. Calculate the anomaly score based on the average path length; mark the data points with anomaly scores greater than the preset sensitivity threshold as anomalies, and obtain the anomaly point set; S34. Perform spatial clustering and morphological closing operations on all marked outliers in the outlier set to generate a binarized potential sensitive area mask file.

5. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 4, characterized in that, S4 includes the following steps: S41. Using the potential sensitive area mask file, perform mask extraction on the spatiotemporal ecological data cube to obtain the sensitive area dataset; S42. Based on the sensitive area dataset, the weights of soil erosion factor, topographic factor and habitat quality factor are calculated using principal component analysis to obtain the first dynamic weight set; the ecological vulnerability index is obtained by weighted summation of the first dynamic weight set. S43. Based on the sensitive area dataset, the weights of nighttime light intensity, impermeable surface ratio, and road network density are calculated using the analytic hierarchy process to obtain the second dynamic weight set; the human activity stress index is obtained by weighted summation of the second dynamic weight set. S44. Construct a capacity coupling formula; calculate the coupling degree between the ecological vulnerability index and the human activity pressure index using the capacity coupling formula; introduce a comprehensive evaluation index to correct the coupling degree and obtain the coupling coordination degree. S45. Combine the ecological vulnerability index, the human activity pressure index, and the coupling coordination degree to form an ecological coupling feature vector.

6. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 5, characterized in that, S5 includes the following steps: S51. Obtain the ecological coupling feature vector sequence at several historical time steps to obtain the historical ecological coupling feature vector sequence. S52. Construct a long short-term memory network prediction model; train the long short-term memory network prediction model and optimize the model parameters using the historical ecological coupling feature vector sequence to obtain the optimized long short-term memory network prediction model. S53. Input the current ecological coupling feature vector into the optimized long short-term memory network prediction model to predict the coupling coordination degree at future time steps, and obtain the predicted coupling coordination degree. S54. Statistical analysis of the predicted coupling coordination degree is performed using the natural breakpoint method to determine the grading threshold, and the target area is divided into different levels to generate a dynamic ecological sensitivity grading map.

7. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 6, characterized in that, S52 includes the following steps: S521. Divide the historical ecological coupling feature vector sequence into training data and test data; The long short-term memory network prediction model is trained using training data, and the learning rate of the initial long short-term memory network prediction model is adjusted by manual parameter tuning to obtain the trained long short-term memory network prediction model. S522. Set the test accuracy threshold; input the test data into the trained long short-term memory network prediction model to obtain the training model prediction data, calculate the similarity between the training model prediction data and the label data in the test data, and obtain the test accuracy; determine whether the test accuracy is ≥ the test accuracy threshold. If it is greater, then the trained long short-term memory network prediction model is used as the optimized long short-term memory network prediction model. Otherwise, the parameters of the trained Long Short-Term Memory (LSTM) network prediction model are optimized using a genetic algorithm to obtain the optimal solution; the optimal solution is then used as the parameters of the trained LSTM network prediction model to obtain the optimized LSTM network prediction model.

8. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 7, characterized in that, The optimal solution obtained by optimizing the parameters of the long short-term memory network prediction model trained by the genetic algorithm in S522 includes the following steps: S5221. Construct a chromosome population; each chromosome in the chromosome population serves as a parameter for a randomly trained Long Short-Term Memory (LSTM) network prediction model; set the maximum number of optimization iterations. S5222. By maximizing the test accuracy, select, crossover, and mutate chromosomes in the chromosome population; repeat the iteration until the set maximum number of iterations is reached or the test accuracy is greater than or equal to the test accuracy threshold, and obtain the optimal solution.

9. The intelligent identification and assessment method for ecologically sensitive areas based on big data analysis according to claim 6, characterized in that, S6 includes the following steps: S61. Calculate the water conservation and soil retention of the target area based on the InVEST model, and generate an ecosystem service value distribution map. S62. Spatially overlay the dynamic ecological sensitivity grading map with the ecosystem service value distribution map to construct a two-dimensional decision matrix that includes sensitivity and value; S63. Based on the quadrant position of the two-dimensional decision matrix, match the preset strategy library and output protection priority suggestion instructions; the instructions include level one red line protection instructions, level two ecological restoration instructions, level three restricted development instructions and level four moderate utilization instructions.

10. A smart identification and assessment system for ecologically sensitive areas based on big data analysis, used to implement the smart identification and assessment method for ecologically sensitive areas based on big data analysis as described in any one of claims 1-9; the system comprises: The multi-source data sensing and reconstruction module is responsible for connecting to satellite data interfaces and IoT gateways, performing Savitzky-Golay filtering and Kriging interpolation operations, and outputting a spatiotemporal ecological data cube. The deep feature mining module, with its built-in CNN network structure and CCA algorithm library, is responsible for extracting high-dimensional features from the cube. The automatic sensitive area identification module runs the isolated forest algorithm and outputs a mask of potential sensitive areas. The coupling evaluation and prediction module integrates the physical coupling model and the LSTM prediction unit, and is responsible for calculating the coupling coordination degree and predicting future trends. The hierarchical decision output module stores the decision rule base and outputs the final protection strategy instructions based on the InVEST model results.

Images