Dynamic monitoring method and system of ecological quality of marine protected areas based on GEE platform

Abstract

The present application provides a method and system for dynamic monitoring of the ecological quality of a marine protected area based on a GEE platform, comprising: S1: calling remote sensing data within a first preset time period and a preset sea area through the GEE platform, extracting ecological indicators from the remote sensing data to form an ecological monitoring indicator set, and constructing a time series of ecological indicators by pixel point unit; S2: constructing a fluctuation characteristic vector of the pixel point for the ecological indicator time series corresponding to each pixel point, and inputting the fluctuation characteristic vector of each pixel point into a clustering algorithm to divide the sensitivity level area of ​​the preset sea area; S3: constructing a regional indicator weight adjustment model, calculating the degree of influence of each ecological indicator in each sensitive level area on the ecological quality, and generating a regional adaptive weight table; S4: extracting the ecological indicator corresponding to each pixel point and generating a fusion index value of the pixel point according to a preset fusion method, and inputting the fusion index value into a fuzzy neural network model to map the pixel point to an ecological quality level.

Description

Technical Field

[0001] The present invention relates to the field of ecological environment monitoring, and specifically discloses a method and system for dynamic monitoring of ecological quality of marine protected areas based on a GEE platform. Background Art

[0002] Traditional marine ecological quality monitoring methods rely on on-site sampling and experimental analysis, which are characterized by long cycles, limited coverage, and poor real-time performance. These methods struggle to meet the demand for continuous, dynamic, and multi-indicator monitoring of large-scale marine ecological quality changes. In recent years, the development of remote sensing technology and cloud computing platforms has made big data-based marine ecological monitoring possible. However, existing methods suffer from the following challenges: They lack regionally adaptive integration of ecological indicators and dynamic weight modeling; they lack the ability to automatically identify sensitive areas, preventing them from effectively focusing on areas experiencing significant ecological change; and they often rely on static assessment methods, lacking predictive and early warning capabilities. Summary of the Invention

[0003] The purpose of this application is to provide a method and system for dynamic monitoring of the ecological quality of marine protected areas based on the GEE platform. By constructing a remote sensing indicator system and combining time-series image processing with intelligent evaluation models, it can achieve long-term, automatic, and visual analysis and evaluation of marine ecological quality.

[0004] To achieve the above objectives, one aspect of this application provides a method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform, comprising the following steps:

[0005] S1: calling remote sensing data within a first preset time period and a preset sea area through the GEE platform, extracting ecological indicators from the remote sensing data to form an ecological monitoring indicator set, and constructing a time series of ecological indicators by pixel point unit;

[0006] S2: Calculate the fluctuation statistical characteristics of the ecological indicator time series corresponding to each pixel point to construct the fluctuation characteristic vector of the pixel point. Input the fluctuation characteristic vector of each pixel point into the clustering algorithm to divide the sensitive level area of ​​the preset sea area;

[0007] S3: constructing a regional indicator weight adjustment model based on the sensitivity level areas, calculating the impact of each ecological indicator on ecological quality in each sensitivity level area, and generating a regional adaptive weight table;

[0008] S4: Extracting the ecological index corresponding to each pixel point and generating a fusion index value of the pixel point according to a preset fusion method, and inputting the fusion index value into a fuzzy neural network model, and mapping the pixel point to an ecological quality grade through the fuzzy neural network model.

[0009] Preferably, the fluctuation characteristic vector includes at least two items of range, standard deviation, seasonal index, trend slope, and time mutation frequency.

[0010] Preferably, S2 includes:

[0011] S21: extracting ecological indicators of the pixel points within a second preset time period;

[0012] S22: Calculate at least two of the range, standard deviation, seasonal index, trend slope, and time mutation frequency of the ecological indicator;

[0013] S23: The calculation results are combined into a fluctuation feature vector and input into a clustering algorithm for use in sensitive level area partitioning.

[0014] Preferably, the clustering algorithm divides the preset sea area into at least three sensitivity levels according to the fluctuation characteristic vector, and constructs regional ecological indicator weight adjustment parameters in each sensitivity level area as input to the regional ecological indicator weight adjustment model.

[0015] Preferably, S3 includes:

[0016] S31: sampling a preset number of pixel points in each sensitivity level area;

[0017] S32: Construct the correlation distribution between each ecological indicator and ecological quality level;

[0018] S33: Use the grey correlation method to calculate the influence coefficient of each indicator, and set the adjustment function according to the sensitivity level to perform quadratic weighting on it;

[0019] S34: Generate a regional adaptive weight table for the fuzzy neural network.

[0020] Preferably, the preset fusion method includes: weighted synthesis of the ecological index corresponding to each pixel point and the weight in the index weight table corresponding to the sensitive level area where it is located, so as to form a fusion index value reflecting the ecology of the pixel point.

[0021] Preferably, S4 further includes: introducing regional weights as control parameters when constructing the fuzzy neural network model to adapt to areas of different sensitivity levels.

[0022] Preferably, the method further includes S5: performing statistical analysis on the ecological quality levels of the pixel points within each sensitive level area to determine the overall ecological quality level of each sensitive level area.

[0023] Preferably, the ecological indicators include one or more of chlorophyll concentration, suspended particulate matter concentration, sea surface temperature, water transparency and water index.

[0024] Another aspect of the present application is to provide a method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform, including:

[0025] a remote sensing data acquisition module configured to call remote sensing data within a first preset time period and a preset sea area through the GEE platform, extract multiple ecological indicators to form an ecological monitoring indicator set, and construct a time series of ecological indicators by pixel point unit;

[0026] The sensitivity level classification module is configured to calculate the fluctuation statistical characteristics of the ecological indicator time series corresponding to each pixel point to construct a fluctuation feature vector of the pixel point, and input the fluctuation feature vector into a clustering algorithm to classify the sensitivity level area of ​​the preset sea area;

[0027] A regional weight modeling module is configured to construct a regional indicator weight adjustment model based on the sensitivity level areas, calculate the impact of each ecological indicator on ecological quality in each sensitivity level area, and generate a regional adaptive weight table;

[0028] And an ecological grade assessment module is configured to extract the ecological indicators corresponding to each pixel point, generate a fusion index value of the pixel point according to a preset fusion method, and input the fusion index value into a fuzzy neural network model, and map the pixel point to an ecological quality grade through the fuzzy neural network model.

[0029] The technical solution provided by this application can achieve the following technical effects:

[0030] 1. This application divides the monitored sea area into areas of different sensitivity levels by introducing fluctuation characteristic vectors and clustering algorithms, and constructs an adaptive ecological indicator weight model in each type of area. It can effectively reflect the differences in the responses of different areas to ecological characteristics, making the ecological quality assessment more spatially adaptable.

[0031] 2. This application achieves the mapping of ecological levels by weighted fusion of current ecological index values ​​and regional sensitivity weights to generate fused index values ​​as model input, and introduces regional control parameters in the fuzzy neural network construction process.

[0032] 3. Based on the output pixel-level ecological quality grade, this application further derives the overall ecological grade of each sensitive level area through statistical analysis methods to support regional ecological risk zoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0034] Figure 1This is a flow chart of a method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform provided in an embodiment of the present application;

[0035] Figure 2 This is a flow chart of the method for calculating the fluctuation characteristic vector provided by the embodiment of the present application;

[0036] Figure 3 This is a flow chart of a method for generating a regional adaptive weight table provided in an embodiment of the present application;

[0037] Figure 4 Schematic diagram of the structure of the dynamic monitoring method of the ecological quality of marine protected areas based on the GEE platform provided in the embodiment of the present application. DETAILED DESCRIPTION

[0038] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0039] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0040] This application provides a method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform, which includes the following steps:

[0041] S1: calling remote sensing data within a first preset time period and a preset sea area through the GEE platform, extracting ecological indicators from the remote sensing data to form an ecological monitoring indicator set, and constructing a time series of ecological indicators in pixel units.

[0042] In one embodiment of the present application, S1 is used to complete the acquisition of remote sensing data and the construction of ecological indicators for a preset sea area within a first preset period of time, forming an ecological monitoring indicator set that supports subsequent analysis and processing. The preset sea area at least includes the target monitoring sea area.

[0043] First, set the area and time range of the preset sea area. Specifically, in a preset control system, set the latitude and longitude range of the preset sea area, for example:

[0044] R=[λ min ,λ max ]×[Φ min , Φ max ];

[0045] Among them, λ represents longitude, λ min Indicates the minimum longitude of the preset sea area, λ max Indicates the maximum longitude of the preset sea area; Φ indicates latitude, Φ min Indicates the minimum latitude of the preset sea area, Φ max Indicates the maximum latitude of the preset sea area.

[0046] At the same time, set the first preset time period:

[0047] T=[t start ,t end ];

[0048] Wherein, T represents the first preset period, t start Indicates the start time of analysis, t end Indicates the end time of the analysis. For example, T can be set from January 1, 2020 to December 31, 2024.

[0049] Through the GEE platform interface, a multi-source remote sensing image dataset covering the preset sea area R and the first preset time period T is called. The GEE platform can access historical remote sensing data from MODIS, Sentinel-2 MSI, and Landsat series satellites. Among the multi-source remote sensing image datasets, the optimal dataset is automatically selected based on image resolution and availability, and image scheduling is performed.

[0050] From the acquired remote sensing data, a preset remote sensing inversion model is used to calculate multiple relevant ecological indicators, including but not limited to: chlorophyll a concentration (Chl-a), suspended particulate matter concentration (TSM), sea surface temperature (SST), water transparency and water quality index (NDWI). Among them, the inversion formula for chlorophyll a concentration is as follows:

[0051] ;

[0052] Where, , is the logarithm of the ratio of the two bands.

[0053] The inversion formula for sea surface temperature is as follows:

[0054] ;

[0055] Where, T 31 With T 32 is the brightness temperature in the thermal infrared band, and a, b, and c are empirical coefficients.

[0056] The inversion formula of water index is as follows:

[0057] ;

[0058] Where, ρ green is the reflectivity of the green band, ρ NIR is the reflectivity in the near-infrared band, which can be derived from the B3 band (560 nm) and the B8 band (842 nm) of Sentinel-2.

[0059] It should be noted that each ecological indicator is resampled with a uniform spatial resolution, and the grid structure is generated using pixels as units.

[0060] For each pixel point P ij (i-th row, j-th column), construct the ecological indicator time series within the first preset period T, with the indicator f k represents the kth ecological indicator (such as Chl-a, NDWI, etc.), then the pixel point P ij At time t, the value of a certain indicator is f k (ij) (t).

[0061] Furthermore, for the pixel point P ij , its complete indicator time series is expressed as:

[0062] ;

[0063] The above structure constitutes a three-dimensional dataset: pixel space dimension (i, j), indicator dimension (k), and time dimension (t).

[0064] It is conceivable that to address missing data due to cloud cover and outliers in remote sensing data, interpolation algorithms (such as linear interpolation and KNN interpolation) or band synthesis methods can be introduced to complete data restoration. Once all processing is completed, a set of ecological monitoring indicators can be obtained, which can be used for subsequent feature extraction and model training.

[0065] Through the above S1, the remote sensing data acquisition, ecological indicator extraction, pixel-level time series construction and preprocessing process of the preset sea area within a given first preset time period were completed based on the GEE platform, providing a data basis for the subsequent sensitive area division and ecological level evaluation.

[0066] S2: Calculate the fluctuation statistical characteristics of the ecological indicator time series corresponding to each pixel point to construct the fluctuation characteristic vector of the pixel point, and input the fluctuation characteristic vector of each pixel point into the clustering algorithm to divide the sensitive level area of ​​the preset sea area.

[0067] In one embodiment of the present application, S2 is used to extract the fluctuation characteristics of the ecological indicator sequence of each pixel point in the time dimension, and perform clustering division of sensitive level areas based on this.

[0068] Following the processing result of S1, for any pixel point P in the preset sea area ij , which is in the first preset time period T=[t1,t2,...,t n ] constitutes the various ecological indicators of F (ij) Structure, in F (ij) Where, f k (ij) (t) represents the pixel point P ij The observed value of the kth ecological indicator at time t, where K represents the number of ecological indicators.

[0069] Furthermore, for each pixel point P ij Calculate the time series fluctuation characteristics of each ecological indicator, including but not limited to the following statistics: range, standard deviation, seasonal index, maximum change slope, and mutation frequency. Among them, standard deviation is used to indicate the intensity of fluctuation, maximum change slope is used to indicate rate characteristics, seasonal index is used to indicate periodic differences, and mutation frequency is used to indicate instability characteristics.

[0070] Specifically, the standard deviation is calculated as follows:

[0071] ;

[0072] Where, is the pixel point P ij The standard deviation of the time series on the kth ecological indicator, is the pixel point P ij The value of the kth ecological indicator at time t, is the pixel point P ij The time series mean of the kth ecological indicator, n is the length of the time series, that is, the number of observation time points in the second preset period, t∈{1,2,...,n}, t represents each observation moment in the time series.

[0073] The calculation formula for the maximum change slope is as follows:

[0074] ;

[0075] Where, is the pixel point P ij The maximum change slope corresponding to the kth ecological indicator in the entire time series, Δt is the time interval between two adjacent time points, t∈{1,2,...,n−1}.

[0076] The seasonal difference is calculated as follows:

[0077] ;

[0078] Where, is the pixel point P ij The absolute difference between the summer mean and the winter mean of the kth ecological indicator, is the pixel point P ij The average value of the observed values ​​of the kth ecological indicator in all summer months, It is the average value of the observed values ​​of the kth ecological indicator in all winter months. The summer average and winter average are calculated by monthly grouping, for example, summer is June to August and winter is December to February.

[0079] The calculation formula for mutation frequency is as follows:

[0080] The remote sensing sequence mutation detection algorithm BFAST is used to perform trend mutation analysis on each indicator sequence, and the number of mutations is recorded as:

[0081] ;

[0082] is the pixel point P ij The number of mutations detected in the kth ecological indicator time series. CountBreaks represents the operation function used to detect structural mutations in time series.

[0083] The above statistics are spliced ​​according to the factor dimension to form the pixel point P ij The fluctuation eigenvector of :

[0084] ;

[0085] Where V (ij) is the pixel point P ij The fluctuation characteristic vector of is used as the input of the clustering algorithm. is the standard deviation, is the seasonal difference, is the mutation frequency.

[0086] To ensure consistency in clustering results, it is recommended to normalize each statistical feature, such as min-max normalization or Z-score normalization.

[0087] The fluctuation characteristic vector V of all pixel points (ij) As input, spatial clustering is performed using an unsupervised clustering algorithm, including but not limited to: K-means, DBSCAN and GMM.

[0088] In one example, the K-means clustering algorithm is selected, and the clustering process is as follows:

[0089] ;

[0090] Where V (ij) is the pixel point P ij The fluctuation characteristic vector, μ cij For c ij Class center vector, c ij ∈{1,2,...,N} is the pixel point P ij The sensitivity level category number.

[0091] The classification results of all pixels are reorganized into a sensitivity level layer by spatial location. Each pixel is assigned a category label. For example, the labels are set to "0, 1, 2," where "0" represents a low-sensitivity area, "1" represents a medium-sensitivity area, and "2" represents a high-sensitivity area. The sensitivity level layer serves as the basis for subsequent regional weight modeling and can be exported to spatial data formats such as GeoTIFF and KML.

[0092] S2 models the fluctuation characteristics of pixel-level multi-index time series data, constructs fluctuation characteristic vectors, and clusters spatially sensitive areas based on this, thereby achieving automated division of differences in the dynamic response of marine ecological systems.

[0093] S3: Construct a regional indicator weight adjustment model based on the sensitive level areas, calculate the impact of each ecological indicator on ecological quality in each sensitive level area, and generate a regional adaptive weight table.

[0094] In one embodiment of the present application, S3 is used to construct a regional indicator weight adjustment model for subsequent ecological quality assessment based on the sensitive level areas divided by S2, and calculate the degree of influence of each ecological indicator on the ecological quality in each type of sensitive area, and generate a regional adaptive weight table.

[0095] For each sensitive level area Z l (where l∈{1,2,...,L}, such as high, medium and low sensitivity areas), randomly select several pixel points {P ij} as training samples.

[0096] For each pixel point, obtain its ecological index value at the current moment , and assign a corresponding ecological quality grade label y to it based on the survey results or historical assessment results (ij) .

[0097] For each sensitive area Z l To construct the correlation measurement between ecological indicators and ecological levels in the region, the grey correlation analysis method can be used. First, the sequence of each ecological indicator is standardized. } and the corresponding ecological level sequence {y (ij)}; For each indicator k, calculate the correlation coefficient between it and the level vector. The calculation formula is as follows:

[0098] ;

[0099] ;

[0100] Where, is the absolute value of the difference between the indicator value and the grade value, is the pixel point P ij The kth ecological indicator value, y (ij) is the pixel point P ij The ecological quality level, is the grey correlation coefficient between the kth ecological indicator and its ecological quality grade, is the minimum difference between the ecological index and the ecological level, is the maximum value of the difference between the ecological index and the ecological level, ρ∈(0,1) is the discrimination coefficient, and the value can be 0.5;

[0101] Calculate the average grey relational degree of each indicator as the original score of the indicator weight. The formula is as follows:

[0102] ;

[0103] Where, For sensitive level area Z l The average grey correlation coefficient of the kth ecological indicator is used as the original impact weight score of the indicator, N l Sensitive level area Z l The number of pixel point samples included.

[0104] Furthermore, the calculated indicator correlation scores are normalized to obtain the region Z l Normalized weights of ecological indicators within , satisfying the following formula:

[0105] ;

[0106] Where, is the kth ecological indicator in the sensitive level area Z l The normalized weight value in is , and K is the total number of ecological indicators. The final generated regional adaptive weight table is shown in Table 1 below:

[0107] Table 1 Regional adaptive weight table

[0108]

[0109] The generated weight table of each sensitive level area is stored in the database or configuration file in the form of a dictionary or key-value pair for subsequent fusion index generation or neural network model call. ij When the sensitivity level zone Z l Dynamically call the corresponding weight table.

[0110] Through S3, an independent ecological indicator weight adjustment model was constructed in each sensitive level area, and a regional adaptive weight table was generated based on the degree of correlation between pixels and ecological levels.

[0111] S4: Extracting the ecological index corresponding to each pixel point and generating a fusion index value of the pixel point according to a preset fusion method, and inputting the fusion index value into a fuzzy neural network model, and mapping the pixel point to an ecological quality grade through the fuzzy neural network model.

[0112] In one embodiment of the present application, S4 is used to fuse the ecological indicators of each pixel point at the current moment according to the weight information of the corresponding sensitive area, generate a fusion index value, and pass it as input into the fuzzy neural network model (FNN) to realize the evaluation of the ecological quality level of the pixel point.

[0113] For any pixel point, obtain the ecological indicator value set at the current moment, which is expressed as follows:

[0114] ;

[0115] in, Represents pixel point P ij The current value of the kth ecological characteristic, K is the total number of indicators.

[0116] According to P ij Find the sensitive level area Z to which it belongs l , and obtain the corresponding weight vector from the regional adaptive weight table:

[0117] .

[0118] In one example, a weighted summation method is used to generate a fusion index value, and the formula is as follows:

[0119] .

[0120] The fuzzy neural network consists of the following main structures: input layer, fuzzy membership layer, rule reasoning layer, weighted output layer and output layer. Specifically, the input layer receives the fusion index value F (ij) , or the original indicator vector; the fuzzy membership layer converts the input data into a fuzzy set, using triangular, trapezoidal or Gaussian membership functions, and the formula is as follows:

[0121] ;

[0122] Where μ k,j (x k ) is the input value x k The membership degree of the jth fuzzy subset belonging to the kth ecological indicator, c k，j is the center position of the Gaussian function, σ k，j is the standard deviation of the Gaussian function;

[0123] The rule inference layer is based on IF-THEN fuzzy rules, for example: "IF Chl-a is high AND SST is high THEN grade = poor"; the weighted output layer aggregates the rule outputs and calculates the final score; the output layer outputs the ecological quality grade label, for example: 1-excellent, 2-good, 3-medium, 4-poor).

[0124] The fuzzy neural network model is trained using existing labeled samples, and the parameters of each layer are optimized using gradient descent method or genetic algorithm.

[0125] After the model training is completed, the fusion index value F (ij) Input model, output level label y (ij) ∈{1,2,3,4}, that is:

[0126] ;

[0127] Where θ is the neural network parameter.

[0128] The ecological quality levels of all pixel points are summarized and reorganized into a spatial distribution layer to form a grade classification table, as shown in Table 2 below:

[0129] Table 2 Ecological quality classification table

[0130]

[0131] The above-mentioned hierarchical layers can be exported to GeoTIFF, KML and other formats for visualization analysis and ecological risk management.

[0132] Through S4, based on the current indicator status of the pixel point and the weight of the sensitive area to which it belongs, a closed-loop modeling of fusion indicator generation and grade mapping is realized, which has regional adaptive capabilities and effectively improves the accuracy of ecological quality assessment.

[0133] In one example, S2 includes:

[0134] S21: Extracting ecological indicators of the pixel points within a second preset time period.

[0135] Specifically, based on the remote sensing data obtained by the GEE platform, for each pixel point P ij Extract the n ], for example, the extracted ecological indicator series include: chlorophyll a concentration (Chl-a) , Water Body Index (NDWI) and sea surface temperature (SST) wait.

[0136] The time series structure is represented as:

[0137] ;

[0138] The time series of each ecological indicator was used for subsequent calculation of fluctuation characteristics.

[0139] S22: Calculate at least two of the range, standard deviation, seasonal index, trend slope, and time mutation frequency of the ecological indicator.

[0140] Specifically, for each ecological indicator sequence at each pixel point, at least two of the following fluctuation statistical characteristics are calculated to construct the feature space, where the range calculation formula is as follows:

[0141] ;

[0142] Where, is the pixel point P ij The time series of the kth ecological indicator is extremely poor. is the pixel point P ij The maximum observed value of the kth ecological indicator, is the pixel point P ij The difference between the minimum observed values ​​of the kth ecological indicator.

[0143] The least squares method is used to fit the linear trend of the indicator series. The calculation formula for the trend slope is as follows:

[0144] ;

[0145] Where, is the pixel point P ij The trend slope of the kth ecological indicator time series is one of the fluctuation characteristics, and ε is the residual term, that is, the error between the actual observation value and the fitted value.

[0146] S23: The calculation results are combined into a fluctuation feature vector and input into a clustering algorithm for use in sensitive level area partitioning.

[0147] Specifically, the fluctuation eigenvalues ​​obtained in S22 are combined into a eigenvector to form the pixel point P ij The fluctuation eigenvector of :

[0148] ;

[0149] Where, is the range of the kth ecological indicator at the pixel point, is the standard deviation, is the seasonal difference, is the trend slope, is the mutation frequency.

[0150] To ensure clustering accuracy, the features of each dimension are normalized, such as using Z-score normalization or Min-Max normalization.

[0151] The feature vector set of all pixel points is input into the K-means clustering algorithm to divide the entire preset sea area into sensitive level areas.

[0152] Finally, each pixel point will be labeled with a sensitivity level category label c (ij) ∈{0,1,...,N−1}, and use this to generate a sensitivity level space layer.

[0153] Through S21-S23, the fluctuation characteristics of ecological indicators at each pixel point within the second preset time period were quantified, and sensitive areas were divided based on the multi-indicator fluctuation vector. This method avoids subjective zoning, supports dynamic and differentiated spatial ecological assessment schemes, and improves the overall adaptability and regional performance of the model.

[0154] In one example, the clustering algorithm divides the preset sea area into at least three sensitivity levels according to the fluctuation characteristic vector, and constructs regional ecological indicator weight adjustment parameters in each sensitivity level area as input to the regional ecological indicator weight adjustment model.

[0155] Specifically, the set of fluctuation eigenvectors of all pixel points is expressed as follows:

[0156] ;

[0157] Where, is the set of fluctuation characteristic vectors of all pixel points in the preset sea area A, is the pixel point P ij The fluctuation characteristic vector of .

[0158] Will The K-means clustering algorithm is input for spatial clustering, and the number of clusters NNN is set to 3, corresponding to the sensitivity level division: high sensitivity area Z1, medium sensitivity area Z2 and low sensitivity area Z3.

[0159] The clustering objective function is to minimize the intra-class square error, and the calculation formula is as follows:

[0160] ;

[0161] Where μ l is the center vector of the lth class, is the square of the Euclidean distance between the pixel point feature vector and its cluster center, is the sum of squared errors between the feature vectors of all pixels belonging to the lth cluster and the center of the cluster.

[0162] Each pixel point P ij Get a category label c (ij) ∈{1,2,3}, the clustering results are reorganized into a sensitivity level layer.

[0163] In each sensitive level area Z l Independently carry out ecological indicator weight modeling, in each area Z l Select a representative pixel point set within:

[0164] ;

[0165] Where, From sensitive level area Z l The representative pixel point set selected from Indicates that the collection is a sensitive level area Z l A subset of .

[0166] For each pixel point, obtain its current ecological indicator value vector:

[0167] ;

[0168] Where, is the pixel point P ij The ecological indicator value vector at the current moment, is the pixel point P ij The current value of the kth ecological indicator.

[0169] and its known or simulated ecological quality grade labels(ij) ∈{1,2,3,4}.

[0170] Calculate the grey correlation between each ecological indicator k and the grade label, normalize the original weight score of each indicator, and generate the region Z l The weight parameter set w (Zl) , weight vector w (Zl) It is used in the following two ways: first, it is used for the current moment indicator fusion (generating the fusion indicator value in S4); second, it is used as the fuzzy neural network model structure parameter for weight configuration in training and inference.

[0171] Based on this, the monitored sea area is divided into three sensitive level areas using clustering algorithm, and exclusive ecological indicator weight parameters are constructed in each area as input items of the regional ecological weight model.

[0172] In one example, S3 includes:

[0173] S31: Sampling a preset number of pixel points in each sensitivity level area.

[0174] Specifically, for each sensitive level area Z l , according to the preset spatial balance or random sampling strategy, select N from the region s Representative pixel points As weighted modeling samples.

[0175] The data of each sample includes: the ecological indicator value vector at the current moment: ; Corresponding ecological quality grade label y (ij) ∈{1,2,3,4}. This sampling set constitutes the region Z l The training sample set is used for subsequent indicator weight analysis.

[0176] S32: Construct the correlation distribution between each ecological indicator and ecological quality level.

[0177] Specifically, in region Z l For each ecological indicator f k The relationship between the quality level label y constructs a set of correlation data series and standardizes the index value of each sample and the grade value y (ij) , to eliminate the dimension difference and form the following data set:

[0178] ;

[0179] Each It represents the distribution of the kth ecological indicator and grade value in a certain area, and prepares for the calculation of the grey correlation coefficient.

[0180] S33: The grey correlation method is used to calculate the influence coefficient of each indicator, and the adjustment function is set according to the sensitivity level to perform quadratic weighting on it.

[0181] Specifically, the correlation coefficient calculation formula is and initial weight values This has been given in the above examples and will not be repeated here.

[0182] Considering the differences in sensitivity between different regions, this example introduces a regional weight adjustment function, whose formula is as follows:

[0183] ;

[0184] Where, Φ(Z l ) is the regional sensitivity adjustment factor. For example, the high-sensitivity area can be set to Φ=1.2, the medium-sensitivity area to 1.0, and the low-sensitivity area to 0.8.

[0185] For all Normalize and get the normalized weight value as follows:

[0186] .

[0187] S34: Generate a regional adaptive weight table for the fuzzy neural network.

[0188] Specifically, the normalized weight value is used as the region Z l The ecological indicator weight configuration items are used to generate a regional adaptive weight table, as shown in Table 3 below:

[0189] Table 3 Regional adaptive weight table

[0190]

[0191] The regional adaptive weight table can be stored in the format of JSON object, database record or configuration file for dynamic calling by subsequent fuzzy neural network model.

[0192] In one example, the preset fusion method includes: weighted synthesis of the ecological index corresponding to each pixel point and the weight in the index weight table corresponding to the sensitive level area where it is located to form a fusion index value reflecting the ecology of the pixel point.

[0193] Specifically, for any pixel point, obtain its ecological index value vector f at the target time t0 from the monitoring index set of S1 (ij) , according to the sensitivity level layer divided by clustering in S2, determine the pixel point P ij Sensitive area Z l, read the weight vector w of the region from the regional adaptive weight table generated in S3 (Zl) , using the linear weighted synthesis method, the current ecological index value is combined with its regional weight to calculate the fusion index value. The calculation formula is as follows:

[0194] .

[0195] This example also provides a data structure, for example, the current pixel point index value is: f (ij) =[1.8,26.5,0.42,34.0,0.16], the corresponding weight table for area Z1 (highly sensitive area) is: w (Z1) =[0.35,0.15,0.20,0.18,0.12], then the fusion index value is 1.8⋅0.35+26.5⋅0.15+0.42⋅0.20+34.0⋅0.18+0.16⋅0.12=10.778, which will be used as the main input signal for model evaluation.

[0196] The fusion index value F of each pixel point (ij) The data is input into the fuzzy neural network model (FNN), trained with historical evaluation labels, and finally the pixel ecological quality grade is output.

[0197] In one example, S4 further includes: introducing regional weights as control parameters when constructing the fuzzy neural network model to adapt to areas of different sensitivity levels.

[0198] Specifically, for the membership function layer of the fuzzy neural network model, the control function of the regional weight is to adjust the shape of the membership function. The regional weight is used as an adjustment parameter to dynamically control the membership function width of each input indicator. The formula is as follows:

[0199] ;

[0200] Where σ0 is the default width, is the kth index in Z l The weight of the region, β is the adjustment sensitivity coefficient, and the control effect of the regional weight makes the indicator with a larger weight have a wider fuzzy response range.

[0201] During the training process of the fuzzy neural network model, the following contents are input simultaneously: current ecological index value or fusion value, regional weight vector w (Zl) Based on the training results, the fuzzy neural network model will learn how to automatically adjust the judgment path based on the weight distribution in different regions, thus realizing the automation of regional ecological quality assessment.

[0202] In one example, the step further includes S5: performing statistical analysis on the ecological quality levels of the pixel points within each sensitive level area to determine the overall ecological quality level of each sensitive level area.

[0203] Specifically, the following two types of input data are obtained: the ecological quality level layer output by the S4 fuzzy neural network model and the sensitivity level area layer output by the S2 clustering algorithm. ij Assigned an eco-grade label (ij) ∈{1,2,3,4}, representing "excellent, good, medium, poor" respectively. Each pixel point corresponds to a sensitivity level area number Z l For example, Z1 represents a high-sensitivity area, Z2 represents a medium-sensitivity area, and Z3 represents a low-sensitivity area. The area number is used as the classification basis, and the pixel level values ​​are grouped and counted.

[0204] The majority statistics method can be used to calculate the sensitivity level of each area Z. l , collect the ecological level label sets of all pixel points within it, expressed as follows:

[0205] ;

[0206] The level corresponding to the largest number of pixels in the region is taken as the overall level, which is expressed as follows:

[0207] ;

[0208] Where Mode(⋅) is the mode function, For region Z l The ecological quality level set of all pixel points in .

[0209] Alternatively, a weighted average grade method can be used, where the grade labels are treated as numerical values ​​(excellent = 1, poor = 4), the average grade is calculated, and the average grade is rounded down to the nearest integer as the regional grade, as shown below:

[0210] ;

[0211] Where N l It is area Z l The total number of pixels in the image.

[0212] You can also use the risk-sensitive method to set risk rules. For example, if the proportion of "poor" pixels in the area exceeds the set threshold (such as 30%), it will be evaluated as "poor", which is expressed as follows: if (count(poor) / N l > 0.3) → G(Z l ) = bad.

[0213] Each sensitive area Z lThe statistical results of G(Z l ) as its overall ecological quality grade, and generate a regional grade layer. Each region in this layer has only one grade label, which expresses the overall ecological status of the sensitive area. Its visual layer can be shown in Table 4 below:

[0214] Table 4 Sensitive area level layer

[0215]

[0216] These results can be exported as vector or raster images for loading into mapping systems, used to generate reports, heat maps, or used as a basis for prioritizing governance. Regional ratings can serve as a reference for early warning levels for ecological governance and can be compared with historical regional ratings to identify ecological degradation.

[0217] In one example, a method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform is also provided, including a remote sensing data acquisition module 10 , a sensitivity level classification module 20 , a regional weight modeling module 30 and an ecological level assessment module 40 .

[0218] Specifically, the remote sensing data acquisition module 10 is configured to call the remote sensing data within the first preset time period and the preset sea area through the GEE platform, extract multiple ecological indicators to form an ecological monitoring indicator set, and construct a time series of ecological indicators by pixel point unit; the sensitivity level division module 20 is configured to calculate the fluctuation statistical characteristics of the ecological indicator time series corresponding to each pixel point to construct a fluctuation characteristic vector of the pixel point, and input the fluctuation characteristic vector into the clustering algorithm to divide the sensitivity level area of ​​the preset sea area; the regional weight modeling module 30 is configured to construct a regional indicator weight adjustment model according to the sensitive level area, calculate the degree of influence of each ecological indicator in each sensitive level area on the ecological quality, and generate a regional adaptive weight table; the ecological level assessment module 40 is configured to extract the ecological indicator corresponding to each pixel point, generate the fusion index value of the pixel point according to the preset fusion method, and input the fusion index value into the fuzzy neural network model, and map the pixel point to the ecological quality level through the fuzzy neural network model.

[0219] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform, characterized by: The following steps are involved: S1: calling remote sensing data within a first preset time period and a preset sea area through the GEE platform, extracting ecological indicators from the remote sensing data to form an ecological monitoring indicator set, and constructing a time series of ecological indicators by pixel point unit; S2: Calculate the fluctuation statistical characteristics of the ecological indicator time series corresponding to each pixel point to construct the fluctuation characteristic vector of the pixel point. Input the fluctuation characteristic vector of each pixel point into the clustering algorithm to divide the sensitive level area of ​​the preset sea area; S3: constructing a regional indicator weight adjustment model based on the sensitivity level areas, calculating the impact of each ecological indicator on ecological quality in each sensitivity level area, and generating a regional adaptive weight table; S4: extracting the ecological index corresponding to each pixel point and generating a fusion index value of the pixel point according to a preset fusion method, and inputting the fusion index value into a fuzzy neural network model, and mapping the pixel point to an ecological quality grade through the fuzzy neural network model; Among them, the ecological indicators of each pixel point at the current moment are fused according to the weight information of the corresponding sensitive area to generate a fusion index value.

2. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1, characterized in that: The fluctuation characteristic vector includes at least two items of the range, standard deviation, seasonal index, trend slope, and time mutation frequency.

3. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 2, wherein S2 include: S21: extracting ecological indicators of the pixel points within a second preset time period; S22: Calculate at least two of the range, standard deviation, seasonal index, maximum change slope, and mutation frequency of the ecological indicator; S23: The calculation results are combined into a fluctuation feature vector and input into a clustering algorithm for use in sensitive level area partitioning.

4. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1 or 3, characterized in that: The clustering algorithm divides the preset sea area into at least three sensitivity levels according to the fluctuation characteristic vector, and constructs regional ecological indicator weight adjustment parameters in each sensitivity level area to serve as input of the regional ecological indicator weight adjustment model.

5. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1, wherein S3 include: S31: sampling a preset number of pixel points in each sensitivity level area; S32: Construct the correlation distribution between each ecological indicator and ecological quality level; S33: Use the grey correlation method to calculate the influence coefficient of each indicator, and set the adjustment function according to the sensitivity level to perform quadratic weighting on it; S34: Generate a regional adaptive weight table for the fuzzy neural network.

6. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1 or 5, characterized in that: The preset fusion method includes: weighted synthesis of the ecological index corresponding to each pixel point and the weight in the index weight table corresponding to the sensitive level area where the pixel point is located, to form a fusion index value reflecting the ecology of the pixel point.

7. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1, characterized in that: S4 also includes: introducing regional weights as control parameters when constructing the fuzzy neural network model.

8. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1, characterized in that: It also includes S5: conducting statistical analysis on the ecological quality levels of the pixel points in each sensitive level area to determine the overall ecological quality level of each sensitive level area.

9. The method for dynamic monitoring of ecological quality of marine protected areas according to claim 1, characterized in that: The ecological indicators include one or more of chlorophyll concentration, suspended particulate matter concentration, sea surface temperature, water transparency and water index.

10. A method for dynamic monitoring of ecological quality of marine protected areas based on the GEE platform, using the method according to any one of claims 1 to 9, characterized in that: include: a remote sensing data acquisition module configured to call remote sensing data within a first preset time period and a preset sea area through the GEE platform, extract multiple ecological indicators to form an ecological monitoring indicator set, and construct a time series of ecological indicators by pixel point unit; The sensitivity level classification module is configured to calculate the fluctuation statistical characteristics of the ecological indicator time series corresponding to each pixel point to construct a fluctuation feature vector of the pixel point, and input the fluctuation feature vector into a clustering algorithm to classify the sensitivity level area of ​​the preset sea area; A regional weight modeling module is configured to construct a regional indicator weight adjustment model based on the sensitivity level areas, calculate the impact of each ecological indicator on ecological quality in each sensitivity level area, and generate a regional adaptive weight table; And an ecological grade assessment module is configured to extract the ecological indicators corresponding to each pixel point, generate a fusion index value of the pixel point according to a preset fusion method, and input the fusion index value into a fuzzy neural network model, and map the pixel point to an ecological quality grade through the fuzzy neural network model.

Images