A vegetation index interpolation method and system based on LSTM model

Abstract

The present invention relates to the field of vegetation index processing technology, and specifically discloses a vegetation index interpolation method and system based on an LSTM model, the method comprising acquiring remote sensing data of a vegetation area according to a preset time period, calculating a vegetation index according to the remote sensing data; arranging the calculated vegetation index according to a time sequence, determining a feature time window and a label time window according to the number of vegetation indexes, selecting features based on the feature time window, selecting labels based on the label time window, and constructing a sample set; training an LSTM model based on a new sample set, performing data prediction on a data missing position based on the LSTM model, obtaining a predicted value, and inserting the predicted value into the data missing position. The present invention obtains vegetation indexes through remote sensing data to obtain an index sequence, trains an LSTM model based on the index sequence, and when faced with missing data, the missing data can be completed with the help of a trained LSTM model, with extremely high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of vegetation index processing, and in particular to a vegetation index interpolation method and system based on an LSTM model. Background Art

[0002] LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) designed to solve the long-term dependency problem in long sequence data. Unlike traditional RNN, LSTM retains long-term information by introducing "memory units" and "gating mechanisms", effectively avoiding the gradient vanishing or gradient exploding problems, making it perform better in processing time series data and sequence prediction tasks.

[0003] The vegetation index is a data that changes with time. Each moment has a corresponding vegetation index. In practical applications, there are many cases where data is missing. For example, the data obtained at certain moments is obviously invalid, or the remote sensing data acquisition at certain moments fails. Correspondingly, the vegetation index cannot be calculated. These missing data affect the integrity of the vegetation index. How to interpolate the missing vegetation index and improve the data integrity is the technical problem that the technical solution of the present invention wants to solve. Summary of the invention

[0004] The purpose of the present invention is to provide a vegetation index interpolation method and system based on the LSTM model to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A vegetation index interpolation method based on an LSTM model, the method comprising:

[0007] Acquiring remote sensing data of a vegetation area according to a preset time period, and calculating a vegetation index according to the remote sensing data;

[0008] Arrange the calculated vegetation indices in chronological order, determine the feature time window and label time window according to the number of vegetation indices, select features based on the feature time window, select labels based on the label time window, and construct a sample set;

[0009] Select data according to a preset length in the exponential sequence, perform fast Fourier transform on the selected data to obtain frequency component information, insert the frequency component information into the selected features as new features, and obtain a new sample set;

[0010] The LSTM model is trained based on the new sample set, and data is predicted at the data missing position based on the LSTM model to obtain a predicted value, which is then inserted into the data missing position.

[0011] As a further solution of the present invention: the step of acquiring remote sensing data of a vegetation area according to a preset time period and calculating a vegetation index according to the remote sensing data comprises:

[0012] Receive the time step set by the administrator, and obtain remote sensing data of the vegetation area every time the time step passes;

[0013] Extract reflectance parameters from remote sensing data and calculate vegetation index based on the reflectance parameters;

[0014] The calculation process of the vegetation index includes:

[0015] ;

[0016] ;

[0017] In the formula, Represents the first vegetation index, and its value is represents the reflectivity of near-infrared light, Indicates the reflectivity of red light;

[0018] represents the second vegetation index, Indicates the reflectivity of blue light.

[0019] As a further solution of the present invention: the vegetation indices calculated according to the time sequence are arranged, the feature time window and the label time window are determined according to the number of vegetation indices, features are selected based on the feature time window, and labels are selected based on the label time window. The steps of constructing a sample set include:

[0020] Query the acquisition time of the remote sensing data corresponding to the vegetation index, arrange the vegetation index based on the acquisition time, and obtain an index sequence;

[0021] The total number of elements of the exponential sequence is obtained, and a characteristic time window is determined according to the total number of elements; the characteristic time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence;

[0022] Receive a time span input by a user as a label time window; the label time window is represented by a numerical value, and the value of the numerical value is a positive integer, representing the number of elements in the exponential sequence;

[0023] Select a subsequence in the exponential sequence based on the characteristic time window as a feature;

[0024] Based on the label time window, a subsequence connected to the tail of each feature is selected as the label of the corresponding feature;

[0025] Count the features and their corresponding labels to get the sample set.

[0026] As a further solution of the present invention: the step of selecting data according to a preset length in the exponential sequence, performing fast Fourier transform on the selected data to obtain frequency component information, inserting the frequency component information into the selected feature as a new feature, and obtaining a new sample set includes:

[0027] Selecting data according to a preset length in the exponential sequence, performing fast Fourier transform on the selected data, and obtaining frequency component information;

[0028] selecting a frequency component from the frequency component information based on the amplitude;

[0029] Insert the frequency component into the selected feature as a new feature;

[0030] The process of selecting frequency components is as follows:

[0031] Selecting a preset number of frequency components in descending order from the frequency component information based on the amplitude, synchronously calculating the difference ratio of the selected frequency components, and screening the frequency components according to the difference ratio;

[0032] The process of inserting frequency components into the selected features is:

[0033] Assume that the exponential sequence is The selected frequency components are , the corresponding amplitude is , the feature after insertion is ;in, Represents the total number of elements in the sequence, represents the total number of selected frequency components, Indicates the index sequence index.

[0034] As a further solution of the present invention: the steps of training the LSTM model based on the new sample set, predicting the data at the data missing position based on the LSTM model, obtaining the predicted value, and inserting the predicted value into the data missing position include:

[0035] Train the LSTM model based on the new sample set;

[0036] When a data missing position is retrieved, a label time window containing the data missing position is determined;

[0037] Querying a feature time window based on the tag time window;

[0038] Read the data in the feature time window, input the trained LSTM model, and obtain the predicted output;

[0039] Calculate the numerical mean of all predicted outputs at the data missing position, and insert it into the data missing position as the predicted value.

[0040] As a further solution of the present invention: the method further comprises:

[0041] Classify vegetation areas according to their locational relationships;

[0042] For similar vegetation areas, the trained LSTM model corresponding to the vegetation area is compared regularly;

[0043] Determine the abnormal model based on the comparison results, read the sample set corresponding to the abnormal model, and send it to the verification end.

[0044] The technical solution of the present invention also provides a vegetation index interpolation system based on the LSTM model, the system comprising:

[0045] A vegetation index calculation module, used to obtain remote sensing data of a vegetation area according to a preset time period, and calculate a vegetation index according to the remote sensing data;

[0046] A sample set construction module is used to arrange the calculated vegetation indices in chronological order, determine the feature time window and the label time window according to the number of vegetation indices, select features based on the feature time window, select labels based on the label time window, and construct a sample set;

[0047] A sample set updating module is used to select data according to a preset length in the exponential sequence, perform fast Fourier transform on the selected data to obtain frequency component information, insert the frequency component information into the selected features as new features, and obtain a new sample set;

[0048] The data insertion module is used to train the LSTM model based on the new sample set, perform data prediction on the data missing position based on the LSTM model, obtain the predicted value, and insert the predicted value into the data missing position.

[0049] As a further solution of the present invention: the vegetation index calculation module includes:

[0050] A data acquisition unit is used to receive a time step set by an administrator, and acquire remote sensing data of a vegetation area once each time step passes;

[0051] A calculation execution unit, used for extracting reflectance parameters from remote sensing data and calculating vegetation index based on the reflectance parameters;

[0052] The calculation process of the vegetation index includes:

[0053] ;

[0054] ;

[0055] In the formula, Represents the first vegetation index, and its value is represents the reflectivity of near-infrared light, Indicates the reflectivity of red light;

[0056] represents the second vegetation index, Indicates the reflectivity of blue light.

[0057] As a further solution of the present invention: the sample set construction module includes:

[0058] An index sequence generating unit is used to query the acquisition time of the remote sensing data corresponding to the vegetation index, and arrange the vegetation index based on the acquisition time to obtain an index sequence;

[0059] A characteristic time window generating unit is used to obtain the total number of elements of the exponential sequence and determine the characteristic time window according to the total number of elements; the characteristic time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence;

[0060] A label time window generating unit is used to receive a time span input by a user as a label time window; the label time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence;

[0061] A feature generation unit, used for selecting a subsequence in the exponential sequence as a feature based on a feature time window;

[0062] A label generation unit, used for selecting a subsequence connected to the tail of each feature based on the label time window as a label of the corresponding feature;

[0063] The data statistics unit is used to count the features and their corresponding labels to obtain a sample set.

[0064] As a further solution of the present invention: the sample set updating module includes:

[0065] A frequency conversion unit, used to select data in the exponential sequence according to a preset length, perform fast Fourier transform on the selected data, and obtain frequency component information;

[0066] A component selection unit, used for selecting a frequency component from the frequency component information based on the amplitude;

[0067] An update execution unit, used for inserting the frequency component into the selected feature as a new feature;

[0068] The process of selecting frequency components is as follows:

[0069] Selecting a preset number of frequency components in descending order from the frequency component information based on the amplitude, synchronously calculating the difference ratio of the selected frequency components, and screening the frequency components according to the difference ratio;

[0070] The process of inserting frequency components into the selected features is:

[0071] Assume that the exponential sequence is The selected frequency components are , the corresponding amplitude is , the feature after insertion is ;in, Represents the total number of elements in the sequence, represents the total number of selected frequency components, Indicates the index sequence index.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention obtains vegetation index through remote sensing data to obtain an index sequence, and trains an LSTM model based on the index sequence. When faced with missing data, the missing data can be supplemented with the help of the trained LSTM model with extremely high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0074] Figure 1 This is a flowchart of the vegetation index interpolation method based on the LSTM model.

[0075] Figure 2 This is the first sub-process flowchart of the vegetation index interpolation method based on the LSTM model.

[0076] Figure 3 This is the second sub-process flowchart of the vegetation index interpolation method based on the LSTM model.

[0077] Figure 4 This is the third sub-process flowchart of the vegetation index interpolation method based on the LSTM model.

[0078] Figure 5 This is the fourth sub-process flowchart of the vegetation index interpolation method based on the LSTM model.

[0079] Figure 6 This is a structural block diagram of the vegetation index interpolation system based on the LSTM model. DETAILED DESCRIPTION

[0080] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0081] Figure 1 The flowchart of the vegetation index interpolation method based on the LSTM model is shown in FIG. 1 . In an embodiment of the present invention, a vegetation index interpolation method based on the LSTM model includes:

[0082] Step S100: acquiring remote sensing data of a vegetation area according to a preset time period, and calculating a vegetation index according to the remote sensing data;

[0083] The subject of acquiring remote sensing data is the satellite. The execution subject of this method acquires remote sensing data by sending a data acquisition request to the satellite, and then the satellite feeds back the remote sensing data. In this process, there is a time interval for acquiring remote sensing data. This time interval is the time period in the above content, which can be one hour, half a day, one day, etc. The reflectivity of near-infrared, visible light bands and blue light is extracted from the remote sensing data, and the vegetation index can be calculated based on the reflectivity.

[0084] Step S200: Arrange the calculated vegetation indices in chronological order, determine the feature time window and the label time window according to the number of vegetation indices, select features based on the feature time window, select labels based on the label time window, and construct a sample set;

[0085] The remote sensing data at each moment corresponds to a vegetation index. The remote sensing data contains a time tag, so the vegetation index also contains a time tag. The vegetation index calculated based on the time sequence of the time tag is arranged to obtain an index sequence, which is actually an array composed of vegetation indices.

[0086] Get the total number of elements in the exponential sequence. The more elements there are, the more data there is. At this time, the number of elements used as features (dependent variables) can be slightly larger, represented by the feature time window, which represents the number of feature elements. Furthermore, it is also necessary to determine the number of label elements (label time window). The label represents the dependent variable, and the number of elements is generally in the single digit.

[0087] For the above features and labels, assuming that the features include 30 vegetation indices and the labels include 2 vegetation indices, the function of the trained model is: input 30 vegetation indices and predict 2 vegetation indices.

[0088] Features and labels are read in the index sequence. One feature and one label are regarded as a sample. Since there are a lot of data in the index sequence, many samples can be constructed to obtain a sample set.

[0089] It is worth mentioning that the corresponding features and labels are continuous in time. 30 vegetation indices are input and 2 vegetation indices are predicted, with a total of 32 data, which are continuous in time.

[0090] Step S300: selecting data in the exponential sequence according to a preset length, performing fast Fourier transform on the selected data to obtain frequency component information, inserting the frequency component information into the selected feature as a new feature, and obtaining a new sample set;

[0091] The feature itself is a time domain signal. By performing fast Fourier transform on the selected features, we can obtain information in the frequency domain, which is called frequency component information. Frequency component information is data at various frequencies. The main consideration is the amplitude at different frequencies. In the frequency component information, components with sufficiently high amplitudes are selected and inserted into the original features as additional features to obtain new features, and then a new sample set is obtained.

[0092] In summary, the above content is to convert the time series data into the frequency domain through the fast Fourier transform (FFT) to obtain the frequency components of the data. Then, you can select the components of specific frequencies as additional features, or concatenate the frequency features with the original time series data and input them into the LSTM model to help the model capture periodic or repetitive patterns.

[0093] It should be noted that if the number of features is small, it is meaningless to perform fast Fourier transform on the selected features. This requires fast Fourier transform on data within a larger time span to obtain frequency component information; the larger time span is generally several times or dozens of times the number of features, which is still a relatively small value relative to the total number of data in the exponential sequence.

[0094] Step S400: training an LSTM model based on a new sample set, performing data prediction on a data missing position based on the LSTM model to obtain a predicted value, and inserting the predicted value into the data missing position;

[0095] The LSTM model is trained based on the new sample set. After the LSTM model is trained, data interpolation can be performed for some data missing positions based on the LSTM model. For a data missing position, the previous data (with the same dimension as the feature) is read, and the data is input into the trained LSTM model. The predicted value can be used as the value to be inserted.

[0096] Figure 2 The first sub-flow chart of the vegetation index interpolation method based on the LSTM model is as follows: the step of acquiring remote sensing data of the vegetation area according to a preset time period and calculating the vegetation index according to the remote sensing data includes:

[0097] Step S101: receiving a time step set by an administrator, and acquiring remote sensing data of a vegetation area once each time step passes;

[0098] Step S102: extracting reflectance parameters from the remote sensing data, and calculating vegetation index based on the reflectance parameters;

[0099] The calculation process of the vegetation index includes:

[0100] ;

[0101] ;

[0102] In the formula, Represents the first vegetation index, and its value is represents the reflectivity of near-infrared light, Indicates the reflectivity of red light;

[0103] represents the second vegetation index, Indicates the reflectivity of blue light.

[0104] The interval for acquiring remote sensing data is set by the administrator. The administrator sets a time step. Every time a time step passes, the execution subject of this method acquires remote sensing data of the vegetation area, extracts reflectivity parameters from the remote sensing data, and calculates the vegetation index based on the reflectivity parameters.

[0105] Figure 3 The second sub-flow diagram of the vegetation index interpolation method based on the LSTM model is as follows. The vegetation index calculated according to the time sequence is arranged, the feature time window and the label time window are determined according to the number of vegetation indexes, features are selected based on the feature time window, and labels are selected based on the label time window. The steps of constructing a sample set include:

[0106] Step S201: querying the acquisition time of remote sensing data corresponding to the vegetation index, and arranging the vegetation index based on the acquisition time to obtain an index sequence;

[0107] Step S202: Obtain the total number of elements in the exponential sequence, and determine a characteristic time window according to the total number of elements; the characteristic time window is represented by a numerical value, and the value of the numerical value is a positive integer, representing the number of elements in the exponential sequence;

[0108] Step S203: receiving a time span input by a user as a label time window; the label time window is represented by a numerical value, and the value of the numerical value is a positive integer, representing the number of elements in the exponential sequence;

[0109] Step S204: selecting a subsequence in the exponential sequence as a feature based on the feature time window;

[0110] Step S205: selecting a subsequence connected to the tail of each feature based on the label time window as a label of the corresponding feature;

[0111] Step S206: Count the features and their corresponding labels to obtain a sample set.

[0112] The above content limits the creation process of the sample set. The remote sensing data corresponding to each vegetation index contains a label. Therefore, each vegetation index corresponds to a time, called the acquisition moment. The vegetation indices are sorted based on the order of the acquisition moments to obtain an index sequence. The total number of elements in the index sequence is read, and the characteristic time window is determined based on the positive proportion of the total number of elements to represent the dimension of the independent variable (also called input or feature). The time span input by the user is received as the dimension of the dependent variable (also called output or label).

[0113] After the feature time window and label time window are determined, features and labels can be selected in the index sequence. It should be noted that the corresponding features and labels are continuous. Therefore, there is a sequence in the feature and label selection process. First, features are selected based on the feature time window, and then labels are selected immediately afterwards. The features and their corresponding labels are counted to obtain a sample set.

[0114] It is worth mentioning that the relationship between the total number of elements and the characteristic time window generally adopts a logarithmic-based composite function, whose function is that as the total number of random elements increases, the change in the characteristic time window corresponding to the same change in the total number of elements gradually decreases, that is, the derivative of the characteristic time window and the total number of elements is inversely proportional.

[0115] Figure 4 The third sub-flow diagram of the vegetation index interpolation method based on the LSTM model is as follows. The steps of selecting data according to a preset length in the index sequence, performing fast Fourier transform on the selected data, obtaining frequency component information, inserting the frequency component information into the selected features as new features, and obtaining a new sample set include:

[0116] Step S301: selecting data in the exponential sequence according to a preset length, performing fast Fourier transform on the selected data, and obtaining frequency component information;

[0117] Step S302: selecting a frequency component from the frequency component information based on the amplitude;

[0118] Step S303: inserting the frequency component into the selected feature as a new feature;

[0119] The process of selecting frequency components is as follows:

[0120] Selecting a preset number of frequency components in descending order from the frequency component information based on the amplitude, synchronously calculating the difference ratio of the selected frequency components, and screening the frequency components according to the difference ratio;

[0121] The process of inserting frequency components into the selected features is:

[0122] Assume that the exponential sequence is The selected frequency components are , the corresponding amplitude is , the feature after insertion is ;in, Represents the total number of elements in the sequence, represents the total number of selected frequency components, Indicates the index sequence index.

[0123] The above content provides a specific sample update scheme. In practical applications, even if the sample set is not updated, the subsequent model training process can be carried out. This method does not take into account the periodicity of the data. In fact, vegetation data will have obvious periodicity. Based on this, the present application selects data according to a preset length in the exponential sequence (generally taking the latest time as the starting point, and reversely obtaining data of the preset length), and performs a fast Fourier transform on the selected data to obtain amplitudes at multiple frequencies, called frequency components. The frequency components are screened to select several relatively important frequency components, and the frequency components are inserted into the selected features to obtain new features, that is, inserting the original features as additional features.

[0124] Figure 5 The fourth sub-flow diagram of the vegetation index interpolation method based on the LSTM model, wherein the steps of training the LSTM model based on the new sample set, predicting the data at the data missing position based on the LSTM model, obtaining the predicted value, and inserting the predicted value into the data missing position include:

[0125] Step S401: training an LSTM model based on a new sample set;

[0126] Step S402: when the data missing position is retrieved, determining the tag time window containing the data missing position;

[0127] Step S403: querying a feature time window based on the tag time window;

[0128] Step S404: read the data in the feature time window, input the trained LSTM model, and obtain the prediction output;

[0129] Step S405: Calculate the numerical mean of all predicted outputs at the data missing position, use it as the predicted value, and insert it into the data missing position.

[0130] In an example of the technical solution of the present invention, the model training and application process are described in detail. First, the LSTM model is trained based on a new sample set. When the data missing position is retrieved, the next step is to interpolate at the data missing position. Specifically, a label time window is first determined based on the data missing position. For example, the label time window contains two data. Then two label time windows can be determined based on the data missing position. In these two label time windows, the data missing position is used as the first data and the second data, respectively. The two label time windows correspond to two features. The features are input into the trained LSTM model to obtain predicted values. Different features can obtain different predicted values. The mean of the predicted values ​​is calculated to obtain the final predicted value, and the data missing position can be inserted.

[0131] As a preferred embodiment of the technical solution of the present invention, the method further includes:

[0132] Classify vegetation areas according to their locational relationships;

[0133] For similar vegetation areas, the trained LSTM model corresponding to the vegetation area is compared regularly;

[0134] Determine the abnormal model based on the comparison results, read the sample set corresponding to the abnormal model, and send it to the verification end.

[0135] In one example of the technical solution of the present invention, vegetation areas are classified according to their positional relationships. Vegetation areas with similar environmental parameters can be classified into one category, or vegetation areas that are sufficiently close can be classified into one category. The area classification process is not complicated. There are many intelligent classification schemes in the prior art, which will not be described in detail in this application.

[0136] For similar vegetation areas, the trained LSTM models obtained should also be similar. Therefore, by comparing similar LSTM models, the similarity between each LSTM model and other LSTM models is queried. According to the similarity, the most abnormal LSTM model, called the abnormal model, can be determined. The sample set corresponding to the abnormal model is read and sent to the verification end. The verification end is a port for verifying the sample set. It is generally an artificial end containing a verification module, which is used to screen the sample set. For the technical solution of the present invention, it is sufficient to interact with the verification end.

[0137] Figure 6 The structure block diagram of the vegetation index interpolation system based on the LSTM model is shown in FIG. 1 . In an embodiment of the present invention, a vegetation index interpolation system based on the LSTM model is provided. The system 10 includes:

[0138] The vegetation index calculation module 11 is used to obtain remote sensing data of the vegetation area according to a preset time period, and calculate the vegetation index according to the remote sensing data;

[0139] The sample set construction module 12 is used to arrange the calculated vegetation indices in time order, determine the feature time window and the label time window according to the number of vegetation indices, select features based on the feature time window, select labels based on the label time window, and construct a sample set;

[0140] The sample set updating module 13 is used to select data according to a preset length in the exponential sequence, perform fast Fourier transform on the selected data to obtain frequency component information, insert the frequency component information into the selected feature as a new feature, and obtain a new sample set;

[0141] The data insertion module 14 is used to train the LSTM model based on the new sample set, perform data prediction on the data missing position based on the LSTM model, obtain the predicted value, and insert the predicted value into the data missing position.

[0142] Furthermore, the vegetation index calculation module 11 includes:

[0143] A data acquisition unit is used to receive a time step set by an administrator, and acquire remote sensing data of a vegetation area once each time step passes;

[0144] A calculation execution unit, used for extracting reflectance parameters from remote sensing data and calculating vegetation index based on the reflectance parameters;

[0145] The calculation process of the vegetation index includes:

[0146] ;

[0147] ;

[0148] In the formula, Represents the first vegetation index, and its value is represents the reflectivity of near-infrared light, Indicates the reflectivity of red light;

[0149] represents the second vegetation index, Indicates the reflectivity of blue light.

[0150] Specifically, the sample set construction module 12 includes:

[0151] An index sequence generating unit is used to query the acquisition time of the remote sensing data corresponding to the vegetation index, and arrange the vegetation index based on the acquisition time to obtain an index sequence;

[0152] A characteristic time window generating unit is used to obtain the total number of elements of the exponential sequence and determine the characteristic time window according to the total number of elements; the characteristic time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence;

[0153] A label time window generating unit is used to receive a time span input by a user as a label time window; the label time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence;

[0154] A feature generation unit, used for selecting a subsequence in the exponential sequence as a feature based on a feature time window;

[0155] A label generation unit, used for selecting a subsequence connected to the tail of each feature based on the label time window as a label of the corresponding feature;

[0156] The data statistics unit is used to count the features and their corresponding labels to obtain a sample set.

[0157] In addition, the sample set updating module 13 includes:

[0158] A frequency conversion unit, used to select data in the exponential sequence according to a preset length, perform fast Fourier transform on the selected data, and obtain frequency component information;

[0159] A component selection unit, used for selecting a frequency component from the frequency component information based on the amplitude;

[0160] An update execution unit, used for inserting the frequency component into the selected feature as a new feature;

[0161] The process of selecting frequency components is as follows:

[0162] Selecting a preset number of frequency components in descending order from the frequency component information based on the amplitude, synchronously calculating the difference ratio of the selected frequency components, and screening the frequency components according to the difference ratio;

[0163] The process of inserting frequency components into the selected features is:

[0164] Assume that the exponential sequence is The selected frequency components are , the corresponding amplitude is , the feature after insertion is ;in, Represents the total number of elements in the sequence, represents the total number of selected frequency components, Indicates the index sequence index.

[0165] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A vegetation index interpolation method based on LSTM model, characterized in that: The method comprises: Acquiring remote sensing data of a vegetation area according to a preset time period, and calculating a vegetation index according to the remote sensing data; Arrange the calculated vegetation indices in chronological order, determine the feature time window and label time window according to the number of vegetation indices, select features based on the feature time window, select labels based on the label time window, and construct a sample set; Select data according to a preset length in the exponential sequence, perform fast Fourier transform on the selected data to obtain frequency component information, insert the frequency component information into the selected features as new features, and obtain a new sample set; Training an LSTM model based on the new sample set, performing data prediction on the data missing position based on the LSTM model to obtain a predicted value, and inserting the predicted value into the data missing position; The steps of arranging the calculated vegetation indices in time order, determining the feature time window and the label time window according to the number of vegetation indices, selecting features based on the feature time window, and selecting labels based on the label time window, and constructing a sample set include: Query the acquisition time of the remote sensing data corresponding to the vegetation index, arrange the vegetation index based on the acquisition time, and obtain an index sequence; The total number of elements of the exponential sequence is obtained, and a characteristic time window is determined according to the total number of elements; the characteristic time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence; Receive a time span input by a user as a label time window; the label time window is represented by a numerical value, and the value of the numerical value is a positive integer, representing the number of elements in the exponential sequence; Select a subsequence in the exponential sequence based on the characteristic time window as a feature; Based on the label time window, a subsequence connected to the tail of each feature is selected as the label of the corresponding feature; Count the features and their corresponding labels to get the sample set.

2. The vegetation index interpolation method based on the LSTM model according to claim 1 is characterized in that: The step of acquiring remote sensing data of the vegetation area according to a preset time period and calculating the vegetation index according to the remote sensing data comprises: Receive the time step set by the administrator, and obtain remote sensing data of the vegetation area every time the time step passes; Extract reflectance parameters from remote sensing data and calculate vegetation index based on the reflectance parameters; The calculation process of the vegetation index includes: ； ； In the formula, Represents the first vegetation index, and its value is , represents the reflectivity of near-infrared light, Indicates the reflectivity of red light; represents the second vegetation index, , Indicates the reflectivity of blue light.

3. The vegetation index interpolation method based on the LSTM model according to claim 1 is characterized in that: The step of selecting data according to a preset length in the exponential sequence, performing fast Fourier transform on the selected data to obtain frequency component information, inserting the frequency component information into the selected feature as a new feature, and obtaining a new sample set includes: Selecting data according to a preset length in the exponential sequence, performing fast Fourier transform on the selected data, and obtaining frequency component information; selecting a frequency component from the frequency component information based on the amplitude; Insert the frequency component into the selected feature as a new feature; The process of selecting frequency components is as follows: Selecting a preset number of frequency components in descending order from the frequency component information based on the amplitude, synchronously calculating the difference ratio of the selected frequency components, and screening the frequency components according to the difference ratio; The process of inserting the frequency components into the selected features is: Assume that the exponential sequence is The selected frequency components are , the corresponding amplitude is , the feature after insertion is ;in, Represents the total number of elements in the sequence, represents the total number of selected frequency components, Indicates the index sequence index.

4. The vegetation index interpolation method based on the LSTM model according to claim 1, characterized in that: The steps of training the LSTM model based on the new sample set, predicting the data at the data missing position based on the LSTM model, obtaining the predicted value, and inserting the predicted value into the data missing position include: Train the LSTM model based on the new sample set; When a data missing position is retrieved, a label time window containing the data missing position is determined; Querying a feature time window based on the tag time window; Read the data in the feature time window, input the trained LSTM model, and obtain the predicted output; Calculate the numerical mean of all predicted outputs at the data missing position, and insert it into the data missing position as the predicted value.

5. The vegetation index interpolation method based on the LSTM model according to claim 1, characterized in that: The method further comprises: Classify vegetation areas according to their locational relationships; For similar vegetation areas, the trained LSTM model corresponding to the vegetation area is compared regularly; Determine the abnormal model based on the comparison results, read the sample set corresponding to the abnormal model, and send it to the verification end.

6. A vegetation index interpolation system based on LSTM model, characterized in that: The system comprises: A vegetation index calculation module, used to obtain remote sensing data of a vegetation area according to a preset time period, and calculate a vegetation index according to the remote sensing data; A sample set construction module is used to arrange the calculated vegetation indices in chronological order, determine the feature time window and the label time window according to the number of vegetation indices, select features based on the feature time window, select labels based on the label time window, and construct a sample set; A sample set updating module is used to select data according to a preset length in the exponential sequence, perform fast Fourier transform on the selected data to obtain frequency component information, insert the frequency component information into the selected features as new features, and obtain a new sample set; A data insertion module is used to train an LSTM model based on a new sample set, perform data prediction on a data missing position based on the LSTM model, obtain a predicted value, and insert the predicted value into the data missing position; The sample set construction module includes: An index sequence generating unit is used to query the acquisition time of the remote sensing data corresponding to the vegetation index, and arrange the vegetation index based on the acquisition time to obtain an index sequence; A characteristic time window generating unit is used to obtain the total number of elements of the exponential sequence and determine the characteristic time window according to the total number of elements; the characteristic time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence; A label time window generating unit is used to receive a time span input by a user as a label time window; the label time window is represented by a numerical value, and the value of the numerical value is a positive integer, which represents the number of elements in the exponential sequence; A feature generation unit, used for selecting a subsequence in the exponential sequence as a feature based on a feature time window; A label generation unit, used for selecting a subsequence connected to the tail of each feature based on the label time window as a label of the corresponding feature; The data statistics unit is used to count the features and their corresponding labels to obtain a sample set.

7. The vegetation index interpolation system based on the LSTM model according to claim 6 is characterized in that: The vegetation index calculation module includes: A data acquisition unit is used to receive a time step set by an administrator, and acquire remote sensing data of a vegetation area once each time step passes; A calculation execution unit, used for extracting reflectance parameters from remote sensing data and calculating vegetation index based on the reflectance parameters; The calculation process of the vegetation index includes: ； ； In the formula, Represents the first vegetation index, and its value is , represents the reflectivity of near-infrared light, Indicates the reflectivity of red light; represents the second vegetation index, , Indicates the reflectivity of blue light.

8. The vegetation index interpolation system based on the LSTM model according to claim 6, characterized in that: The sample set updating module comprises: A frequency conversion unit, used to select data in the exponential sequence according to a preset length, perform fast Fourier transform on the selected data, and obtain frequency component information; A component selection unit, used for selecting a frequency component from the frequency component information based on the amplitude; An update execution unit, used for inserting the frequency component into the selected feature as a new feature; The process of selecting frequency components is as follows: Selecting a preset number of frequency components in descending order from the frequency component information based on the amplitude, synchronously calculating the difference ratio of the selected frequency components, and screening the frequency components according to the difference ratio; The process of inserting the frequency components into the selected features is: Assume that the exponential sequence is The selected frequency components are , the corresponding amplitude is , the feature after insertion is ;in, Represents the total number of elements in the sequence, represents the total number of selected frequency components, Indicates the index sequence index.

Images