Island reef vegetation index prediction method, device and equipment

Abstract

The application provides an atoll vegetation index prediction method, which can be applied to the technical fields of geographic information and deep learning. The method comprises the following steps: obtaining monthly normalized vegetation index data of a target atoll, corresponding marine environment data and meteorological data, and constructing a vegetation growth dataset; calculating the partial correlation of the marine environment data, the meteorological data and the normalized vegetation index change, so as to determine the key environmental data affecting the normalized vegetation index change; determining the lag response period of the normalized vegetation index to each key marine environment data; based on the lag response period, combining attention mechanism and double-time-scale information mining, constructing a double-time-phase convolution long short-term memory neural network model considering the climate time lag effect of the atoll, wherein the neural network model cell structure comprises convolution operation; training the model by using the vegetation growth dataset to obtain a trained model; and predicting the future short-term atoll vegetation normalized vegetation index change by using the trained model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic information and deep learning, and more particularly to an island reef vegetation index prediction method, device and equipment. BACKGROUND

[0002] Vegetation provides information sources for the structure, function and dynamic change of regional ecosystems, and its healthy growth is the basis for maintaining the stability of regional ecological environment and species diversity. Normalized Difference Vegetation Index (NDVI) evaluates the health status and coverage dynamics of vegetation through band operation of remote sensing images, and has been widely used in regional vegetation scientific research and ecological environment assessment. Under the background of globalization, island reefs have become increasingly prominent in terms of resource value and research significance due to their unique offshore location advantage and rich biological resources endowment. Island reefs are far away from the mainland, and the closed environment caused by their inherent boundaries makes the island ecosystem vulnerable, and the vegetation on the island shows abnormal fluctuations different from the distribution pattern on the land. Therefore, accurately grasping the spatial law of long-term vegetation dynamic change of island reefs is of great significance for the sustainable development and ecological protection of island resources.

[0003] Compared with the data stationarity and linearity emphasized by traditional statistical models in time series prediction problems, the emerging deep learning model has the advantages of automatic hierarchical ability for large data sets and adaptability to spatio-temporal non-stationary data, allowing non-linear problems to be modeled from a data-driven perspective, and can capture the complex trends of NDVI spatio-temporal data. At present, in response to the application demand of vegetation NDVI spatio-temporal prediction, domestic scholars have focused on the spatial correlation of NDVI data, calculated the spatial distance matrix and modeled the spatial comprehensive weight of each pixel, added spatial weight information to the neural network to achieve high-precision prediction of NDVI. However, the feature correlation in time series is often ignored, and there is currently a lack of vegetation spatio-temporal prediction research for island reefs. Due to the relative isolation of island reefs from the mainland and their dispersed distribution, the area is usually small and often affected by adverse weather such as typhoons, heavy rain and thick fog. Satellites have difficulty in penetrating through thick clouds to obtain clear island reef images, resulting in a reduction in the effective observation range and time. How to extract the periodic feature information of vegetation in time series from limited remote sensing data to realize intelligent management of the dynamic change of island reef vegetation is an important part of the island reef vegetation NDVI prediction problem. At present, intelligent models for vegetation NDVI spatio-temporal prediction focus on the extraction of universal rules for model improvement, and this improvement method is difficult to adapt to the special vegetation growth cycle rules and growth driving mechanisms caused by island reef types.

[0004] Existing NDVI prediction models based on convolutional long short-term memory neural networks rely solely on historical time-series data for NDVI prediction, lacking a direct response mechanism to environmental factors at the prediction time. In actual vegetation growth environments, NDVI changes are not only affected by the accumulation of historical data, but real-time environmental data also significantly influence NDVI values.

[0005] Existing NDVI spatiotemporal prediction models based on convolutional long short-term memory neural networks focus on capturing the distribution differences and connections between geographical locations from a spatial perspective. They only utilize simple correlations between adjacent time steps and lack the ability to mine periodic and seasonal characteristics of vegetation NDVI time series. For island and reef spatial objects, it is required to extract as many deep temporal variation features as possible from limited time series information and mine long-range dependencies across time periods in the time series.

[0006] Based on the above situation, there is an urgent need for an intelligent method that can accurately extract spatiotemporal variation characteristics of island and reef vegetation growth for island and reef NDVI prediction. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] To address at least one of the aforementioned problems, this invention provides a method, apparatus, and device for predicting island and reef vegetation index. By combining historical NDVI data of island and reef vegetation with climate and marine environmental data, it can quickly predict the evolution pattern of NDVI of island and reef vegetation over a future period, thus compensating for the problem of insufficient extraction of data temporal features caused by the short time span of island and reef remote sensing data, and achieving accurate short-term prediction of NDVI of island and reef vegetation.

[0009] (II) Technical Solution

[0010] To address the aforementioned technical problems, embodiments of the present invention provide a method, apparatus, and equipment for predicting island and reef vegetation indices.

[0011] According to a first aspect of the present application, a reef vegetation index prediction method is provided, comprising: obtaining monthly normalized vegetation index data of a target reef and ocean environment data and meteorological data in a corresponding time range to construct a vegetation growth dataset; using a Pearson partial correlation analysis method, calculating the partial correlation of the ocean environment data and meteorological data with the normalized vegetation index change to determine the key environmental data affecting the normalized vegetation index change, including key ocean environment data and key meteorological data; for the key environmental data, sequentially calculating the partial correlation coefficient of the normalized vegetation index and different lag months in a long time series on a per-pixel basis; based on the maximum value of the partial correlation coefficient and the corresponding lag response month length, determining the lag response period of the normalized vegetation index to each key ocean environment data; based on the lag response period, combining attention mechanism and double time scale information mining, constructing a double phase convolution long short-term memory neural network model considering the time lag effect of reef climate, wherein the double phase convolution long short-term memory neural network model cell structure includes convolution operation; using the vegetation growth dataset to complete the training, verification and testing of the double phase convolution long short-term memory neural network model, obtaining a trained model; and using the trained model to predict future short-term reef vegetation normalized vegetation index change.

[0012] In some exemplary embodiments, the method for obtaining normalized vegetation index data comprises: obtaining initial satellite remote sensing image data; preprocessing the initial satellite remote sensing image data to remove remote sensing images with a cloud layer probability exceeding a preset value to obtain target satellite remote sensing image data; based on the target satellite remote sensing image data, using a monthly mean synthesis method to generate monthly normalized vegetation index data of the target reef in the research year; wherein the format of the normalized vegetation index data includes raster field data.

[0013] In some exemplary embodiments, the ocean environment data and meteorological data include seawater salinity, sea surface wind speed, sea surface temperature, precipitation, and solar radiation, wherein the format of the ocean environment data and meteorological data includes raster field data.

[0014] In some exemplary embodiments, constructing the vegetation growth dataset includes normalizing the ocean environment data and meteorological data to obtain normalized data values.

[0015] In some example embodiments, based on a lag response period, combined with attention mechanism and double time scale information mining, constructing a double time phase convolution long short-term memory neural network model comprises: dividing data windows in time sequence order and setting window sliding step; inputting normalized vegetation index data and key environmental data required for current time step prediction into the input gate of the initial convolution long short-term memory network cell structure, obtaining a cell structure calculation formula, wherein the cell structure comprises an input gate, a forget gate, and an output gate; adding a convolution operation after the output gate, inputting key environmental data corresponding to the prediction time step into the cell structure unit, obtaining a double time phase convolution long short-term memory neural network cell structure; inputting normalized vegetation index data and corresponding key environmental data of the previous M months and key environmental data of the given prediction month, constructing a monthly change prediction sub-model containing two layers of networks, obtaining a monthly scale prediction feature vector; retrieving corresponding normalized vegetation index data and key environmental data of the same period of each year from the data set according to the prediction time step matching time, constructing an annual cycle prediction sub-model containing two layers of networks, obtaining an annual scale prediction feature vector; constructing an attention mechanism fusion model for attention weighting of the monthly scale prediction feature vector and the annual scale prediction feature vector, generating a final normalized vegetation index prediction value, wherein M is the month corresponding to the lag response period.

[0016] In some example embodiments, dividing data windows in time sequence order and setting window sliding step comprises: based on time series, input window is composed of normalized vegetation index data of consecutive M months, corresponding key environmental data, and key environmental data of the prediction month; the normalized vegetation index data of the prediction month is taken as the output window; sliding division is performed according to the preset step, forming an “input-label” data pair.

[0017] In some example embodiments, training, verifying, and testing of the double time phase convolution long short-term memory neural network model are completed using a vegetation growth data set, and the trained model comprises: dividing the vegetation growth data set into a training set, a verification set, and a test set according to a preset ratio; taking mean square error as a loss function, using adaptive moment estimation optimization algorithm, and using training set data to perform multiple rounds of iterative training on the double time phase convolution long short-term memory neural network model until the model converges, obtaining the trained double time phase convolution long short-term memory neural network model; using at least one of the following indicators: mean absolute error, root mean square error, determination coefficient, and Pearson correlation coefficient, using the verification set and the test set to evaluate the prediction performance of the trained double time phase convolution long short-term memory neural network model, obtaining the trained model.

[0018] In some example embodiments, further comprising: performing multiple iterations on the trained model to predict future medium and long term island vegetation normalized vegetation index changes.

[0019] According to a second aspect of the present application, a reef vegetation index prediction device is provided, comprising: an acquisition module for acquiring monthly normalized vegetation index data of a target reef and ocean environment data and meteorological data in a corresponding time range to construct a vegetation growth data set; a first calculation module for calculating the partial correlation of the ocean environment data and meteorological data and the normalized vegetation index change using a Pearson partial correlation analysis method to determine key environmental data affecting the normalized vegetation index change, including key ocean environment data and key meteorological data; a second calculation module for sequentially calculating the partial correlation coefficient of the normalized vegetation index and different lag months under long time series for each key environmental data; a determination module for determining the lag response period of the normalized vegetation index to each key ocean environment data based on the maximum value of the partial correlation coefficient and the corresponding lag response month length; a model construction module for constructing a double-phase convolution long short-term memory neural network model considering the time lag effect of the reef climate based on the lag response period combined with attention mechanism and double time scale information mining, wherein the double-phase convolution long short-term memory neural network model cell structure includes convolution operation; a model training module for completing the training, verification and testing of the double-phase convolution long short-term memory neural network model using the vegetation growth data set to obtain a trained model; and a prediction module for predicting future short-term reef vegetation normalized vegetation index change using the trained model.

[0020] According to a third aspect of the present application, an electronic device is provided, comprising: one or more processors; a memory for storing one or more computer programs, the one or more processors executing the one or more computer programs to implement the steps of any of the above methods.

[0021] (Three) beneficial effects

[0022] As can be seen from the above technical solutions, the reef vegetation index prediction method, device and equipment provided by the embodiments of the present application have at least the following beneficial effects:

[0023] (1) The present application adds convolution operation in the convolution long short-term memory network (ConvLSTM) model cell structure, considers the lag response of NDVI change to environmental data and the direct influence of the corresponding environmental data of the superimposed prediction month on NDVI change, and thus more truly simulates the action mechanism of environmental data on reef vegetation NDVI change.

[0024] (2) The present application uses two sub-model structures of monthly change prediction and annual same period prediction, and integrates the sub-model output feature vectors using attention mechanism, overcomes the problem of limited time series information of reef remote sensing images, and mines the periodicity, seasonal fluctuation and long-term dependence relationship of reef NDVI time series from the time dimension, so as to realize high-precision intelligent prediction of short-term NDVI. Attached Figure Description

[0025] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0026] Figure 1 A flowchart illustrating a method for predicting island and reef vegetation indices according to an embodiment of the present invention is shown.

[0027] Figure 2 The illustration shows a flowchart of constructing a dual-temporal convolutional long short-term memory neural network model that considers the climate lag effect of islands and reefs, based on the lag response period combined with attention mechanism and dual-timescale information mining according to an embodiment of the present invention.

[0028] Figure 3 The schematic diagram illustrates a cell structure of a dual-temporal convolutional long short-term memory neural network according to an embodiment of the present invention.

[0029] Figure 4 The diagram illustrates the structure of a dual-temporal convolutional long short-term memory neural network model that considers the climate lag effect of islands and reefs according to an embodiment of the present invention.

[0030] Figure 5 The flowchart illustrates the process of training, validating, and testing a dual-temporal convolutional long short-term memory neural network model using a vegetation growth dataset according to an embodiment of the present invention, resulting in a trained model.

[0031] Figure 6 A flowchart illustrating another method for predicting island and reef vegetation indices according to an embodiment of the present invention is shown schematically.

[0032] Figure 7 An island and reef vegetation index prediction device according to an embodiment of the present invention is illustrated schematically.

[0033] Figure 8 A block diagram of an electronic device for predicting island and reef vegetation index according to an embodiment of the present invention is shown schematically. Detailed Implementation

[0034] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0035] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the terms "comprises", "comprising", "includes", "including" and the like are specifically intended to be open-ended and to mean that other features, steps, operations, and / or components can be added.

[0036] All terms used herein including technical and scientific terms have the meanings commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the terms used herein are defined as having meanings that are consistent with the context of the specification in which they are used and should not be interpreted in an overly idealized or overly formal way.

[0037] In the case of using expressions similar to "at least one of A, B, and C, etc.", it should generally be interpreted to include at least one of A, B, or C, in other words, to include systems that have at least one of A, B, or C, as well as systems that have two or more of A, B, and C.

[0038] Figure 1 A flowchart of a method for predicting a reef vegetation index according to an embodiment of the present application is schematically shown.

[0039] As shown in Figure 1 A method for predicting a reef vegetation index according to an embodiment of the present application includes steps S110-S170.

[0040] In step S110, monthly normalized vegetation index data of a target reef and marine environment data and weather data within a corresponding time range are acquired, and a vegetation growth dataset is constructed.

[0041] In some exemplary embodiments, the NDVI data is obtained by simple band operation from remote sensing images, and a method for obtaining normalized vegetation index data includes: acquiring initial satellite remote sensing image data; pre-processing the initial satellite remote sensing image data to remove remote sensing images with a cloud layer probability exceeding a preset value, to obtain target satellite remote sensing image data; based on the target satellite remote sensing image data, using a monthly mean synthesis method to generate monthly normalized vegetation index data of the target reef in the research year; wherein the format of the normalized vegetation index data includes raster field data.

[0042] For example, after acquiring remote sensing image data, images with a cloud layer probability exceeding 60% are removed according to the quality band, and a monthly NDVI data of the reef in the research year is generated using a monthly mean synthesis method, wherein the calculation method of NDVI is as follows:

[0043] (1)

[0044] wherein, Representing the near-infrared band, Represents the infrared band.

[0045] In some exemplary embodiments, the marine environmental data and meteorological data include seawater salinity, sea surface wind speed, sea surface temperature, precipitation, and solar radiation, wherein the format of the marine environmental data and meteorological data includes raster field data.

[0046] In some exemplary embodiments, considering the wide and uneven distribution of environmental data values, in order to reduce the complexity and uncertainty of the subsequent neural network training process, the original data values ​​of each environmental data are normalized and mapped sequentially. The calculation method for format conversion is as follows:

[0047] (2)

[0048] in The data values ​​after normalization mapping. The data value to be processed. This corresponds to the minimum value in the original dataset of the environmental data. This represents the maximum value in the original dataset.

[0049] In step S120, the Pearson partial correlation analysis method is used to calculate the partial correlation between marine environmental data, meteorological data and changes in the normalized vegetation index, so as to identify the key environmental data affecting the changes in the normalized vegetation index, including key marine environmental data and key meteorological data.

[0050] For example, using Pearson partial correlation analysis, the partial correlation coefficients between each environmental variable and NDVI are calculated sequentially, eliminating the interference of other irrelevant variables. For instance, when discussing the partial correlation between sea surface wind speed and vegetation NDVI, the influence of three environmental variables—seawater salinity, dew point temperature, and net solar radiation at the land surface—on NDVI changes is eliminated. The formula for calculating the partial correlation coefficient is:

[0051] (3)

[0052] in, For NDVI data, To select environmental variables for calculating partial correlation, For the control variables that need to be removed, for The correlation coefficient between them for The correlation coefficient between them for The correlation coefficient between them This represents the partial correlation coefficient between NDVI and the selected environmental variable.

[0053] In step S130, for the key environmental data, the partial correlation coefficients of the normalized vegetation index and different lag months in a long time sequence are calculated in sequence pixel by pixel.

[0054] In step S140, the lag response period of the normalized vegetation index to each key marine environmental data is determined based on the maximum value of the partial correlation coefficient and the corresponding lag response month length.

[0055] For example, for the key environmental data affecting the island NDVI change, the partial correlation coefficients of the island NDVI and the current month environmental data, the previous 1 month environmental data, the previous 2 month environmental data, the previous 3 month environmental data, the previous 4 month environmental data, the previous 5 month environmental data and the previous 6 month environmental data in a long time sequence are calculated in sequence pixel by pixel, the maximum partial correlation coefficient and the corresponding lag response month length in the 7 combinations are determined pixel by pixel according to the environmental data category, and the lag response period of the island NDVI to each key environmental data is determined. The lag response period of the island NDVI to each key environmental data determined by the present application is 3 months.

[0056] In step S150, based on the lag response period, the attention mechanism and the dual-time scale information mining are combined to construct a dual-temporal convolutional long short-term memory neural network model (Dual-Temporal Climate-Lag Convolutional Long Short-Term Memory model, DTCL-ConvLSTM) considering the time lag effect of island climate, wherein the dual-temporal convolutional long short-term memory neural network model cell structure includes convolution operation.

[0057] In some exemplary embodiments, step S150 includes steps S151-S156, see Figure 2 .

[0058] In step S151, the data window is divided in time sequence order, and the window sliding step is set.

[0059] In some exemplary embodiments, based on the time sequence, the normalized vegetation index data of the previous M months, the corresponding key environmental data and the key environmental data of the predicted month form an input window; the normalized vegetation index data of the predicted month is taken as an output window; and sliding division is performed according to a preset step, forming an "input-label" data pair, wherein M is the month corresponding to the lag response period.

[0060] For example, based on the lag analysis results in step S140, the lag response period is 3 months. Using the NDVI and key environmental data of the previous 3 months, as well as the key environmental data corresponding to the forecast month, the NDVI data for the current forecast month is output. Following the chronological order, the NDVI and key environmental data of the previous 3 months and the key environmental data corresponding to the forecast month form one input window, and the NDVI data of one month form the output window, using a sliding window. The window sliding step is set to 1, and all available data are traversed sequentially along the time series to form complete "input-label" data pairs.

[0061] In step S152, the normalized vegetation index data and key environmental data required for the current time step prediction are input into the input gate of the initial convolutional long short-term memory network (ConvLSTM) cell structure to obtain the cell structure calculation formula, wherein the cell structure includes an input gate, a forget gate, and an output gate.

[0062] For example, the ConvLSTM algorithm includes an input gate, a forget gate, and an output gate. The input gate takes the vegetation NDVI data and environmental data needed for prediction at the current time step. The cell structure calculation formula is as follows:

[0063] (4)

[0064] (5)

[0065] (6)

[0066] (7)

[0067] in, It's a convolution operation. It is Hadamaji. This represents the sigmoid activation function. This represents the hyperbolic tangent activation function. These represent the input gate, forget gate, and output gate, respectively. These represent the memory units at the previous time step t-1 and the current time step t, respectively. Input for the current time step. It was a step a while ago The hidden state. For different convolution operations The weights, similarly, yes The weight, This is the bias value used in the model calculation.

[0068] In step S153, a convolution operation is added after the output gate, and the key environmental data at the time corresponding to the predicted time step is input to the cell structure unit to obtain a double-time-phase convolution long short-term memory neural network (DTCL-ConvLSTM) cell structure. The model cell structure constructed according to the embodiment of the application is shown in FIG. 6. Figure 3 .

[0069] For example, a convolution operation with a convolution kernel size of 3*3 is added after the output gate of the ConvLSTM cell structure, and the environmental data at the time corresponding to the predicted time step is input to the cell unit, that is, the influence of the current time step environmental data on the predicted time step NDVI change is superimposed in the time step prediction to update the cell hidden state, and the DTCL-ConvLSTM cell structure is obtained. For the time , the hidden state calculation formula of the DTCL-ConvLSTM cell structure is as follows:

[0070] (8)

[0071] wherein, is the output of the DTCL-ConvLSTM model unit, is the weight of the convolution operation, is the second input of the model unit, including the environmental data at the time corresponding to the current predicted time step, is the output gate, represents the hyperbolic tangent activation function, represents the memory unit of the current time step t, is the bias term in the convolution operation.

[0072] In step S154, the normalized vegetation index data of the previous M months and the corresponding key environmental data and the key environmental data of the given prediction month are input to construct a monthly change prediction sub-model containing a two-layer network to obtain a monthly scale prediction feature vector.

[0073] In step S155, the normalized vegetation index data and the key environmental data corresponding to the same period of each year in the data set are retrieved according to the prediction time step matching time to construct an annual cycle prediction sub-model containing a two-layer network to obtain an annual scale prediction feature vector.

[0074] In step S156, an attention mechanism fusion model is constructed for attention weighting of the monthly scale prediction feature vector and the annual scale prediction feature vector to generate a final normalized vegetation index prediction value.

[0075] For example, the DTCL-ConvLSTM monthly variation prediction sub-model and the ConvLSTM annual cycle prediction sub-model output the monthly scale prediction feature vector and the annual scale prediction feature vector, and the vector dot product operation is performed to calculate the feature vector attention score. The calculated attention score is converted into a probability distribution through a softmax normalization function, and the two outputs are weighted and fused according to the weight to obtain the final NDVI prediction feature vector. The specific calculation formula is as follows:

[0076] (9)

[0077] (10)

[0078] (11)

[0079] wherein, are the feature vectors output by the DTCL-ConvLSTM monthly variation prediction sub-model and the ConvLSTM annual cycle prediction sub-model, is an attention query vector, is an attention score of the feature vector, is a weight of the output feature vector after normalization processing, is the final output prediction NDVI feature vector after attention processing.

[0080] In step S160, the training, verification and testing of the dual-time convolution long short-term memory neural network model are completed by using the vegetation growth dataset, and a trained model is obtained. The model structure diagram constructed according to the embodiment of the application is shown in Figure 4 .

[0081] In some exemplary embodiments, step S160 includes steps S161-S163, as shown in Figure 5 .

[0082] In step S161, the vegetation growth dataset is divided into a training set, a verification set and a test set according to a preset proportion.

[0083] In step S162, the mean square error is used as a loss function, the adaptive moment estimation optimization algorithm is adopted, and the training set data is used to perform multiple rounds of iterative training on the dual-time convolution long short-term memory neural network model until the model converges, and the trained dual-time convolution long short-term memory neural network model is obtained.

[0084] In step S163, at least one of the mean absolute error, the root mean square error, the determination coefficient and the Pearson correlation coefficient is used to evaluate the prediction performance of the trained dual-time convolution long short-term memory neural network model by using the verification set and the test set, and the trained model is obtained.

[0085] In some exemplary embodiments, the error between the DTCL-ConvLSTM model output result and the true value and the benchmark ConvLSTM model output result is calculated, and the specific calculation formula is as follows:

[0086] (12)

[0087] (13)

[0088] (14)

[0089] (15)

[0090] wherein, is the number of image pixels in the study area, is the prediction value of the DTCL-ConvLSTM model at pixel , is the true value or the prediction value of the benchmark ConvLSTM model at pixel , is the average of all prediction values of the DTCL-ConvLSTM model on the image, is the average of all true values or prediction values of the benchmark ConvLSTM model on the image. In step S170, the trained model is used to predict the future short-term change of the normalized vegetation index of the island vegetation.

[0091] A flowchart of another island vegetation index prediction method according to an embodiment of the present application is schematically shown.

[0092] Figure 6 A flowchart of another island vegetation index prediction method according to an embodiment of the present application is schematically shown.

[0093] As shown in Figure 6 , another island vegetation index prediction method according to an embodiment of the present application further comprises step S180 on the basis of the method shown in Figure 1 .

[0094] In step S180, the trained model is iterated multiple times to predict the future medium and long-term change of the normalized vegetation index of the island vegetation.

[0095] Figure 1 ​​The shown prediction method is a short-term intelligent prediction method for island reef NDVI changes. If medium and long-term change prediction of NDVI is to be realized, the DTCL-ConvLSTM model needs to be iterated multiple times. The future environmental data required for model prediction is provided by future climate change scenario data. Since the model prediction error rises with the accumulation of iterations, the number of iterations of the DTCL-ConvLSTM model provided by the embodiment of the present application is limited and is subject to the prediction error constraint.

[0096] Figure 7 An island reef vegetation index prediction device according to an embodiment of the present application is schematically shown.

[0097] As Figure 7 shown, the island reef vegetation index prediction device 800 of this embodiment includes an acquisition module 810, a first calculation module 820, a second calculation module 830, a determination module 840, a model construction module 850, a model training module 860, and a prediction module 870.

[0098] The acquisition module 810 is configured to acquire monthly normalized vegetation index data of a target island reef and marine environmental data and meteorological data in a corresponding time range, and construct a vegetation growth data set.

[0099] The first calculation module 820 is configured to calculate the partial correlation of the marine environmental data and the meteorological data with the normalized vegetation index change by using a Pearson partial correlation analysis method, so as to determine key environmental data, including key marine environmental data and key meteorological data, that affect the normalized vegetation index change.

[0100] The second calculation module 830 is configured to sequentially calculate, for the key environmental data, a partial correlation coefficient between the normalized vegetation index in a long time sequence and different lag months on a pixel-by-pixel basis.

[0101] The determination module 840 is configured to determine, based on the maximum value of the partial correlation coefficient and the corresponding lag response month length, a lag response period of the normalized vegetation index to each key marine environmental data.

[0102] The model construction module 850 is configured to construct, based on the lag response period, a double-phase convolution long short-term memory neural network model that considers the climate time-lag effect of the island reef by combining attention mechanism and double-time scale information mining, wherein the double-phase convolution long short-term memory neural network model cell structure includes convolution operation.

[0103] The model training module 860 is configured to complete training, verification, and testing of the double-phase convolution long short-term memory neural network model by using the vegetation growth data set, and obtain a trained model.

[0104] The prediction module 870 is configured to predict future short-term island reef vegetation normalized vegetation index change by using the trained model.

[0105] According to embodiments of the present invention, any plurality of modules among the acquisition module 810, the first calculation module 820, the second calculation module 830, the determination module 840, the model building module 850, the model training module 860, and the prediction module 870 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules can be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the acquisition module 810, the first calculation module 820, the second calculation module 830, the determination module 840, the model building module 850, the model training module 860, and the prediction module 870 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 810, the first calculation module 820, the second calculation module 830, the determination module 840, the model building module 850, the model training module 860, and the prediction module 870 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0106] Figure 8 A block diagram of an electronic device for predicting island and reef vegetation index according to an embodiment of the present invention is shown schematically.

[0107] like Figure 8 As shown, an electronic device 900 according to an embodiment of the present invention includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0108] In the RAM 903, various programs and data required for the operation of the electronic device 900 are stored. The processor 901, the ROM 902, and the RAM 903 are connected to each other via the bus 904. The processor 901 performs various operations of the method flow according to the embodiments of the present application by executing the programs in the ROM 902 and / or the RAM 903. It should be noted that the programs can also be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 can also perform various operations of the method flow according to the embodiments of the present application by executing the programs stored in the one or more memories.

[0109] According to the embodiments of the present application, the electronic device 900 can further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 can further include one or more of the following components connected to the input / output (I / O) interface 905: an input part 906 including a keyboard, a mouse, etc.; an output part 907 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 908 including a hard disk, etc.; and a communication part 909 including a network interface card such as a LAN card, a modem, etc. The communication part 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as necessary. A removable recording medium 911 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 910 as necessary, so that a computer program read therefrom is installed in the storage part 908 as necessary.

[0110] The present application also provides a computer readable storage medium, which can be included in the device / apparatus / system described in the above embodiments; or can exist separately without being assembled into the device / apparatus / system. The above computer readable storage medium carries one or more programs, when the one or more programs are executed, the method according to the embodiments of the present application is implemented.

[0111] According to embodiments of the present application, the computer readable storage medium can be a non-transitory computer readable storage medium, which can include, but is not limited to, a portable computer diskette, a hard disk, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or Flash memory), a portable compact disc read only memory (CD-ROM), optical storage, magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium can be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the present application, a computer readable storage medium can include one or more of the above-described ROM 902 and / or RAM 903, and / or one or more memory other than the ROM 902 and the RAM 903.

[0112] It will be appreciated by persons skilled in the art that features of the various embodiments described herein can be combined and / or substituted into each other's embodiments, even if this is not explicitly stated in the present disclosure. In particular, features of the various embodiments described herein can be combined and / or substituted into each other's embodiments without departing from the spirit and teachings of the present application. All such combinations and / or substitutions are within the scope of the present application.

Claims

1. A method for predicting island and reef vegetation indices, characterized in that, include: Obtain monthly normalized vegetation index data for the target islands and reefs, along with corresponding marine environmental and meteorological data for the time range, and construct a vegetation growth dataset. Using Pearson partial correlation analysis, the partial correlation between the marine environmental data, meteorological data and the change in the normalized vegetation index is calculated to identify the key environmental data affecting the change in the normalized vegetation index, including key marine environmental data and key meteorological data. For the key environmental data, the partial correlation coefficients between the normalized vegetation index and different lag months were calculated pixel by pixel for the long-term series. Based on the maximum value of the partial correlation coefficient and its corresponding lag response month length, the lag response period of the normalized vegetation index to each key marine environmental data is determined. Based on the aforementioned lag response period combined with attention mechanism and dual time scale information mining, a dual temporal convolutional long short-term memory neural network model considering the climate lag effect of islands and reefs is constructed, wherein the cell structure of the dual temporal convolutional long short-term memory neural network model includes convolution operation; The vegetation growth dataset was used to train, validate, and test the dual-temporal convolutional long short-term memory neural network model, resulting in a fully trained model; and Using the trained model, predict the short-term changes in the normalized vegetation index of island and reef vegetation. The construction of a dual-temporal convolutional long short-term memory neural network model based on the lag response period combined with attention mechanism and dual-timescale information mining includes: Divide the data into windows according to the time series order and set the window sliding step size; Input the normalized vegetation index data and key environmental data required for the current time step prediction into the input gate of the initial convolutional long short-term memory network cell structure to obtain the cell structure calculation formula, where the cell structure includes an input gate, a forget gate, and an output gate; A convolution operation is added after the output gate, and key environmental data corresponding to the predicted time step are input into the cell structure unit to obtain the dual-temporal convolutional long short-term memory neural network cell structure. Input the normalized vegetation index data and corresponding key environmental data for the previous M months, as well as the key environmental data for the given prediction month, and construct a monthly change prediction sub-model with a two-layer network to obtain the monthly scale prediction feature vector. Based on the predicted time step matching time, the normalized vegetation index data and key environmental data corresponding to the same period of each year are retrieved from the dataset. An annual cycle prediction sub-model with two-layer network is constructed to obtain the annual scale prediction feature vector. An attention-based fusion model is constructed to perform attention-weighted summation on the monthly-scale and annual-scale predicted feature vectors, generating the final normalized vegetation index (NZVI) prediction value. Where M represents the month corresponding to the lag response period.

2. The method according to claim 1, characterized in that, The method for obtaining the normalized vegetation index data includes: Acquire initial satellite remote sensing image data; The initial satellite remote sensing image data is preprocessed to remove remote sensing images with a cloud probability exceeding a preset value, thereby obtaining the target satellite remote sensing image data; Based on the target satellite remote sensing image data, the monthly normalized vegetation index data of the target islands and reefs in the study year are generated using the monthly mean synthesis method. The normalized vegetation index data includes raster field data.

3. The method according to claim 1, characterized in that, The marine environmental and meteorological data include seawater salinity, sea surface wind speed, sea surface temperature, precipitation, and solar radiation. The format of the marine environmental and meteorological data includes raster field data.

4. The method according to claim 1, characterized in that, The construction of the vegetation growth dataset includes normalizing and mapping the marine environmental data and meteorological data to obtain normalized data values.

5. The method according to claim 1, characterized in that, The step of dividing the data window according to the time series order and setting the window sliding step size includes: Based on time series, the input window consists of normalized vegetation index data for M consecutive months, corresponding key environmental data, and key environmental data for the predicted month. The normalized vegetation index data for the predicted month is used as the output window; The data is divided into "input-label" pairs by sliding according to a preset step size.

6. The method according to claim 1, characterized in that, The process of training, validating, and testing the dual-temporal convolutional long short-term memory neural network model using the vegetation growth dataset, resulting in a trained model, includes: The vegetation growth dataset is divided into a training set, a validation set, and a test set according to a preset ratio; Using mean squared error as the loss function, an adaptive moment estimation optimization algorithm is adopted, and the dual-temporal convolutional long short-term memory neural network model is trained in multiple rounds of iterative training using training set data until the model converges, thus obtaining the trained dual-temporal convolutional long short-term memory neural network model. The predictive performance of the trained biphase convolutional long short-term memory neural network model is evaluated using at least one of the following metrics: mean absolute error, root mean square error, coefficient of determination, and Pearson correlation coefficient, with a validation set and a test set, to obtain the trained model.

7. The method according to claim 1, characterized in that, Also includes: The trained model is iterated multiple times to predict the changes in the normalized vegetation index of island and reef vegetation in the medium and long term.

8. A device for predicting the vegetation index of islands and reefs, characterized in that, The device includes: The acquisition module is used to acquire monthly normalized vegetation index data of the target islands and reefs, as well as marine environmental and meteorological data within the corresponding time range, and to construct a vegetation growth dataset. The first calculation module is used to calculate the partial correlation between the marine environmental data and meteorological data and the change in the normalized vegetation index using the Pearson partial correlation analysis method, so as to determine the key environmental data affecting the change in the normalized vegetation index, including key marine environmental data and key meteorological data. The second calculation module is used to calculate the normalized vegetation index and the partial correlation coefficient of different lag months for the key environmental data on a pixel-by-pixel basis over a long time series. The determination module is used to determine the lag response period of the normalized vegetation index to each key marine environmental data based on the maximum value of the partial correlation coefficient and its corresponding lag response month length. The model building module is used to construct a dual-temporal convolutional long short-term memory neural network model that considers the climate lag effect of islands and reefs based on the lag response period combined with attention mechanism and dual-timescale information mining. The cell structure of the dual-temporal convolutional long short-term memory neural network model includes convolution operation. The model training module is used to train, validate, and test the dual-temporal convolutional long short-term memory neural network model using the vegetation growth dataset, thereby obtaining the trained model; and The prediction module is used to predict the short-term changes in the normalized vegetation index of island and reef vegetation using the trained model. The construction of a dual-temporal convolutional long short-term memory neural network model based on the lag response period combined with attention mechanism and dual-timescale information mining includes: Divide the data into windows according to the time series order and set the window sliding step size; Input the normalized vegetation index data and key environmental data required for the current time step prediction into the input gate of the initial convolutional long short-term memory network cell structure to obtain the cell structure calculation formula, where the cell structure includes an input gate, a forget gate, and an output gate; A convolution operation is added after the output gate, and key environmental data corresponding to the predicted time step are input into the cell structure unit to obtain the dual-temporal convolutional long short-term memory neural network cell structure. Input the normalized vegetation index data and corresponding key environmental data for the previous M months, as well as the key environmental data for the given prediction month, and construct a monthly change prediction sub-model with a two-layer network to obtain the monthly scale prediction feature vector. Based on the predicted time step matching time, the normalized vegetation index data and key environmental data corresponding to the same period of each year are retrieved from the dataset. An annual cycle prediction sub-model with two-layer network is constructed to obtain the annual scale prediction feature vector. An attention-based fusion model is constructed to perform attention-weighted summation on the monthly-scale and annual-scale predicted feature vectors, generating the final normalized vegetation index (NZVI) prediction value. Where M represents the month corresponding to the lag response period.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

Images