Grassland vegetation index time sequence analysis and classification method based on SAR remote sensing image

Abstract

The invention relates to a steppe vegetation index time sequence analysis and classification method based on an SAR remote sensing image. The method comprises the steps of obtaining meteorological data, Sentinel-1 SAR data and topographic data of a research area; utilizing the backscattering coefficient to construct a radar vegetation index; constructing and visualizing a time sequence; the time sequence analysis at least comprises trend analysis and mutation point analysis; and decision tree classification is completed. According to the embodiment of the invention, the grassland is subjected to ground feature classification according to the time sequence characteristics, and the problem of data missing caused by long period and weather influence is solved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of SAR remote sensing image technology, and particularly to a classification method for time series analysis of grassland vegetation index based on SAR remote sensing images. Background Art

[0002] In the prior art, the time series analysis method of vegetation index has the following main limitations:

[0003] High requirement for data continuity: Time series analysis requires long-term SAR image data, but the acquisition of SAR data is affected by factors such as satellite revisit period and data archiving restrictions, which may lead to data loss or discontinuity and affect the reliability of analysis results;

[0004] Soil moisture interference: SAR signals are sensitive to soil moisture, and changes in soil moisture may mask vegetation information, resulting in errors in vegetation indices;

[0005] Influence of meteorological conditions: Although SAR is not affected by clouds, extreme meteorological conditions (such as heavy rain and snow cover) may change the scattering characteristics of the ground surface and affect the extraction of vegetation indices. Summary of the Invention

[0006] The present disclosure aims to provide a classification method for time series analysis of grassland vegetation index based on SAR remote sensing images, which classifies ground objects of grassland according to time series characteristics and overcomes the problem of data loss caused by long cycle and weather influence.

[0007] According to one of the solutions of the present disclosure, the classification method for time series analysis of grassland vegetation index based on SAR remote sensing images includes:

[0008] Obtain meteorological data, Sentinel-1 SAR data, and terrain data of the study area;

[0009] Construct a radar vegetation index using the backscattering coefficient;

[0010] Construct and visualize the time series;

[0011] Time series analysis, including at least trend analysis and breakpoint analysis;

[0012] Complete decision tree classification.

[0013] In some embodiments, obtaining meteorological data, Sentinel-1 SAR data, and terrain data of the study area includes:

[0014] Taking the sampling point location as the core, select typical sample plots to record key information including the longitude, latitude, altitude, and main vegetation types of each sample plot;

[0015] Perform processing including meteorological interpolation, data splicing, and cropping;

[0016] Load SAR image data;

[0017] Preprocessing of time-series dual-polarized Sentinel-1 data, including radiometric calibration, multi-look processing, registration, filtering, geocoding, and terrain correction.

[0018] In some embodiments,

[0019] Radiometric calibration includes: using the radiometric calibration information provided by the Sentinel-1 satellite to convert the original radar signal into the radiance information of the ground object;

[0020] Multi-look processing includes: weakening or smoothing the speckle noise effect in the data by integrating SAR data of multiple pixels;

[0021] Filtering includes: using the DeGrandi filter to process the noise and speckle in the time-series SAR image;

[0022] Geocoding includes: converting the slant range coordinate system of the SAR data into a geographic coordinate system.

[0023] In some embodiments, constructing a radar vegetation index using the backscatter coefficient includes:

[0024] Constructing a radar vegetation index based on two polarization modes of the backscatter coefficient, where:

[0025] The radar vegetation index is sensitive to vegetation structure and biomass; the positive value of the dual-polarization vegetation index indicates the vegetation area, and the negative value indicates the non-vegetation area, which is applicable to areas with low vegetation coverage.

[0026] In some embodiments, based on two polarization modes of the backscatter coefficient, includes:

[0027] VV polarization means that both the transmitted and received electromagnetic waves of the radar are vertically polarized;

[0028] VH polarization means that the radar transmits vertically polarized electromagnetic waves and receives horizontally polarized echoes.

[0029] In some embodiments, constructing and visualizing the time series includes:

[0030] Constructing the time series and extracting time series features, and exporting a CSV file;

[0031] Using the chart tool of the Google Earth Engine platform to plot the time series chart and visualize the spatial distribution of the vegetation index in this area over a period of time.

[0032] In some embodiments, trend analysis includes:

[0033] The analysis results of the trend direction and vegetation conditions are obtained through a linear regression equation.

[0034] In some embodiments, the mutation point analysis includes:

[0035] Based on decomposing the time series into a trend component, a seasonal component, and a residual component, analyze the trend and seasonal variations.

[0036] In some embodiments, the completion of decision tree classification includes:

[0037] Use a machine learning algorithm based on a binary tree to establish rules for object-oriented decision trees and automatic threshold decision trees, implement decision tree classification, and evaluate the accuracy of the classification results.

[0038] In some embodiments, the completion of decision tree classification includes:

[0039] Compare each pixel with the training samples and classify it into similar sample classes according to different judgment rules to complete the supervised classification of the entire image;

[0040] Combine with ENVI remote sensing image processing software for object-oriented decision tree classification;

[0041] Use the ENVI software to automatically obtain the extended tool RuleGen based on the decision tree rules of the CART algorithm to implement the construction of decision tree rules.

[0042] The method for classifying the time series analysis of grassland vegetation indices based on SAR remote sensing images in various embodiments of the present disclosure at least obtains meteorological data, Sentinel-1 SAR data, and terrain data of the study area; constructs radar vegetation indices using the backscattering coefficient; constructs and visualizes the time series; the time series analysis includes at least trend analysis and mutation point analysis; completes decision tree classification, mainly using the sensitivity of SAR images to vegetation, combines with time series analysis techniques to classify ground objects in the grassland, so as to use the Google Earth Engine platform for time series analysis and an improved method for ground object classification, uses Google Earth Engine to implement the time series analysis of grassland vegetation indices based on SAR remote sensing images, and classifies ground objects in the grassland according to the time series characteristics.

[0043] It should be understood that the foregoing general description and the following detailed description are merely exemplary and explanatory, and are not restrictive of the claimed present disclosure. Brief Description of the Drawings

[0044] In the accompanying drawings, which are not necessarily drawn to scale, like reference numerals in different views may represent like components. Like reference numerals with alphabetic suffixes or like reference numerals with different alphabetic suffixes may represent different instances of like components. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and are used in conjunction with the specification and the claims to explain the disclosed embodiments.

[0045] Figure 1 The schematic diagram of the data preprocessing process according to an embodiment of the present disclosure is shown;

[0046] Figures 2 to 3 The schematic diagram of the radiometric correction operation according to an embodiment of the present disclosure is shown;

[0047] Figure 4 The schematic diagram of the multi-view processing operation according to an embodiment of the present disclosure is shown;

[0048] Figure 5 The schematic diagram of the filtering operation according to an embodiment of the present disclosure is shown;

[0049] Figure 6 The schematic diagram of the geocoding operation according to an embodiment of the present disclosure is shown;

[0050] Figures 7 to 8 The schematic diagram of constructing and visualizing time series according to an embodiment of the present disclosure is shown;

[0051] Figure 9 The schematic diagram of the supervised classification technical route according to an embodiment of the present disclosure is shown;

[0052] Figure 10 The schematic diagram of the decision tree classification result according to an embodiment of the present disclosure is shown. Detailed implementation manners

[0053] In order to make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present disclosure. Obviously, the described embodiments are some but not all of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present disclosure without creative efforts shall fall within the scope of protection of the present disclosure.

[0054] Synthetic Aperture Radar (SAR), as a microwave remote sensing technology that can perform all-weather and all-day detection, can penetrate certain obstacles for imaging by selecting an appropriate wavelength. Moreover, the imaging resolution is independent of the wavelength and the radar operating distance, overcoming the inherent defects of traditional optical methods, making the SAR technology have greater flexibility and can be applied to more fields.

[0055] In recent years, significant progress has been made in the research on SAR-based vegetation indices. For example: RVI (Radar Vegetation Index): Based on multi-polarization SAR data, it can effectively characterize vegetation coverage and biomass; DVI (Dual Polarization Vegetation Index): Based on dual-polarization SAR data, it is suitable for distinguishing vegetation and non-vegetation areas. These indices provide new technical means for the detection of grassland vegetation.

[0056] Combined with the content recorded in the background art part above, the present disclosure exemplarily records corresponding solutions in the form of embodiments to solve the defects existing in the prior art, but does not limit the scope of patent protection required by the present disclosure.

[0057] As one of the solutions, the embodiment of the present disclosure provides a method for classifying the time series analysis of grassland vegetation indices based on SAR remote sensing images, including:

[0058] Obtain meteorological data, Sentinel-1 SAR data, and terrain data of the study area;

[0059] Construct radar vegetation indices using the backscattering coefficient;

[0060] Construct and visualize the time series;

[0061] Time series analysis, including at least trend analysis and breakpoint analysis;

[0062] Complete decision tree classification.

[0063] In view of the above-mentioned content and the characteristics of SAR images, the present disclosure proposes an improved method for time series analysis and ground object classification using the Google Earth Engine platform. Using Google Earth Engine to implement the time series analysis of grassland vegetation indices based on SAR remote sensing images, and classifying ground objects on the grassland according to the time series characteristics. The purpose of the present disclosure is to utilize the powerful computing resources and massive geospatial datasets of the Google Earth Engine platform to overcome the problems of long cycle and data loss caused by weather impacts. Define the study area using geometric tools, set the data type and perform preprocessing. By calculating radar vegetation indices such as RVI and DVI, visualize and then construct a time series, and extract features from the time series, such as mean, variance, seasonal amplitude, trend, etc. Combine with the ENVI remote sensing image processing software, establish the rules of object-oriented decision tree and automatic threshold decision tree, realize decision tree classification, and evaluate the accuracy of the classification results.

[0064] Steps of the embodiments of the present disclosure. Exemplarily, the following takes the example of applying the method of the present disclosure to the time series analysis of the grassland vegetation index in Xiwu Banner Grassland. Taking steps S1 to S5 as examples, the classification method for the time series analysis of the grassland vegetation index based on SAR remote sensing images of the present disclosure is further described.

[0065] In some specific implementation manners, the present disclosure may be to obtain meteorological data, Sentinel-1 SAR data, and topographic data of the research area, including:

[0066] Taking the sampling point location as the core, select typical plots to record key information including the longitude, latitude, altitude, and main vegetation types of each plot.

[0067] Perform processing including meteorological interpolation, data mosaicking, and cropping.

[0068] Load SAR image data.

[0069] Preprocessing of time series dual-polarization Sentinel-1 data includes radiometric calibration, multi-look processing, registration, filtering processing, geocoding, and topographic correction.

[0070] Step S1: Data preparation. Obtain meteorological data, Sentinel-1 SAR data, and topographic data of the research area.

[0071] Step S11: Field investigation of the grassland. Taking the sampling point location as the core, select typical plots with uniform species distribution, flat terrain, and sufficient area. Then, record in detail key information such as the longitude, latitude, altitude, and main vegetation types of each plot to ensure the representativeness of the plots and the traceability of the data.

[0072] Step S12: The topographic data includes: elevation, slope, and aspect. The elevation data comes from COPERNICUS / DEM / GLO30 with a spatial resolution of 30 m and is provided by the cloud computing platform GEE. The slope and aspect data are generated by calculating the elevation data.

[0073] Step S14: Load SAR image data. The data loaded by the method of the present disclosure is of the Sentinel-1 AGRD type, and the spatial range, polarization mode, and time are filtered. All the following times are from January 1, 2020 to December 30, 2020.

[0074] In some specific implementation manners, such as Figures 1 to 6An example of data preprocessing in an embodiment of the present disclosure is shown. The present disclosure can be radiometric calibration, including: using the radiometric calibration information provided by the Sentinel-1 satellite to convert the original radar signal into the radiometric intensity information of the ground object; multi-looking processing, including: weakening or smoothing the speckle noise effect in the data by integrating the SAR data of multiple pixels; filtering processing, including: using the DeGrandi filter to process the noise and speckle in the temporal SAR image; geocoding, including: converting the slant range coordinate system of the SAR data into a geographic coordinate system.

[0075] Step S15: Data preprocessing. The preprocessing of temporal dual-polarization Sentinel-1 data is a key step in extracting polarization features, including radiometric calibration, multi-looking processing, registration, filtering processing, geocoding, and terrain correction.

[0076] The preprocessing process using the SNAP software is as follows:

[0077] Radiometric calibration: As Figures 2 to 3 shown, radiometric calibration is the process of converting the original data into the backscattering coefficient (σ 0 or β 0 ), including:

[0078] In SNAP, select the Radiometric Calibration tool (search in the toolbar or through the menu: Radar->Radiometric->Calibration);

[0079] Set parameters: Select the output band (such as Sigma0 or Beta0); Select the polarization channel (such as VV, VH);

[0080] Click "Run" to perform radiometric calibration.

[0081] In this embodiment of radiometric calibration, the radiometric calibration information provided by the Sentinel-1 satellite is used to convert the original radar signal into the radiometric intensity information of the ground object. That is, converting the SAR data from the original amplitude value (usually a dimensionless number) into a radiometric intensity or reflectivity with physical units helps to ensure that the temporal SAR data at different times and locations has a consistent radiometric scale, and converting the DN value into the backscattering coefficient (σ 0 ).

[0082] Multi-looking processing: As Figure 4 shown, it includes:

[0083] Multi-looking processing reduces noise by reducing the spatial resolution and improves the signal-to-noise ratio;

[0084] In SNAP, select the Multilooking tool (Radar -> Multilooking);

[0085] Set the parameters;

[0086] The input data is the result after radiometric correction;

[0087] Set the multilooking factors in the azimuth and range directions (e.g., 2:2);

[0088] Click "Run" to perform the multilooking process.

[0089] The purpose of multilooking is mainly to enhance the reliability of the interpretive information of polarimetric SAR data for surface features. By integrating SAR data from multiple pixels, the speckle noise effect in the data is weakened or smoothed. There are two multilooking methods for time-series Sentinel-1 dual-polarization SAR data, one is performed in the time domain and the other is performed in the frequency domain. The embodiment of the present disclosure adopts time-domain multilooking and applies an averaging window to reduce the impact of speckle noise.

[0090] Filtering process, such as Figure 5 As shown, the filtering process is used to reduce speckle noise and improve the image quality, including:

[0091] In SNAP, select the Speckle Filter tool (Radar -> Speckle Filtering -> SingleProduct Speckle Filter);

[0092] Set the parameters: select the filtering method (such as Lee, Gamma Map, etc.) and set the size of the filtering window (such as 5x5);

[0093] Click "Run" to perform the filtering process.

[0094] Whenever two or more images of the same scene are taken at different times, multi-temporal speckle filtering should be applied. The method of this embodiment utilizes the spatio-temporal variation correlation of speckles between images to reduce the noise inherent in the system, and a DeGrandi filter can be used to process the noise, speckles, etc. in time-series SAR images.

[0095] Geocoding, such as Figure 6 As shown, geocoding converts the image from the radar coordinate system to the geographic coordinate system (such as WGS84), including:

[0096] In SNAP, select the Terrain Correction tool (Radar -> Geometric -> Terrain Correction -> Range-Doppler Terrain Correction);

[0097] Set the parameters: select the output projection (such as UTM / WGS84), select the resampling method (such as bilinear interpolation);

[0098] Click "Run" to perform geocoding.

[0099] Terrain correction is used to eliminate the influence of terrain undulation on the backscattering coefficient, including:

[0100] In the geocoding step, the Range-Doppler Terrain Correction tool in SNAP already includes the terrain correction function;

[0101] Ensure that DEM data is selected (such as SRTM 30m or Copernicus DEM);

[0102] Click "Run" to perform terrain correction.

[0103] In practical applications, it is necessary to convert the slant range coordinate system of SAR data into a geographic coordinate system, and this process is the geocoding of SAR data. Terrain correction takes into account the elevation information of the terrain and corrects the SAR image to eliminate the influence caused by the terrain. In the embodiments of the present disclosure, a digital elevation model (DEM) data with a spatial resolution of 30m and a coordinate system of WGS84 is used to perform geocoding and terrain correction on the image, and the resampling window size is 5m x 5m.

[0104] In some specific implementation schemes, the present disclosure may be to construct a radar vegetation index using the backscattering coefficient, including:

[0105] Construct a radar vegetation index based on two polarization modes of the backscattering coefficient, where:

[0106] The radar vegetation index is sensitive to vegetation structure and biomass; the positive value of the dual-polarization vegetation index indicates the vegetation area, and the negative value indicates the non-vegetation area, which is applicable to areas with low vegetation coverage.

[0107] Step S2: Calculate the radar vegetation index.

[0108] In some specific implementation schemes, the present disclosure may be based on two polarization modes of the backscattering coefficient, including:

[0109] VV polarization means that both the electromagnetic waves transmitted and received by the radar are vertically polarized;

[0110] VH polarization means that the radar emits vertically polarized electromagnetic waves and receives horizontally polarized echoes.

[0111] Step S21: Extract radar backscattering data from the acquired image data and calculate various radar vegetation indices. The radar vegetation indices are constructed using the backscattering coefficient, mainly including the Radar Vegetation Index (RVI) and the Dual-Polarization Vegetation Index (DVI).

[0112] The Radar Vegetation Index (RVI)

[0113]

[0114] is sensitive to vegetation structure and biomass and is suitable for vegetation monitoring in grasslands, forests, etc.

[0115] The Dual-Polarization Vegetation Index (DVI)

[0116] DVI = σ VH - σ VV

[0117] Positive values usually indicate vegetation areas, and negative values indicate non-vegetation areas. It is suitable for areas with low vegetation coverage.

[0118] In the data products of Sentinel-1, the backscattering coefficient is an important parameter, which reflects the backscattering ability of surface objects to radar waves. VV and VH are two polarization modes of the backscattering coefficient.

[0119] The backscattering coefficient is the radar cross section per unit area, which represents the backscattering intensity of surface objects to radar waves. In radar remote sensing, after the radar emits an electromagnetic beam and reaches the surface, it is scattered by surface objects. Then the radar receives these scattered echoes and obtains the surface image and relevant information through the processing and analysis of these echoes. The larger the backscattering coefficient, the stronger the echo of the target, that is, the stronger the scattering ability of the surface object to radar waves.

[0120] VV polarization means that both the emitted and received electromagnetic waves by the radar are vertically polarized. In this polarization mode, the interaction between radar waves and surface objects mainly occurs in the vertical direction. Therefore, VV polarization is more sensitive to reflecting the vertical structure characteristics of surface objects. For example, in the remote sensing monitoring of parameters such as soil moisture and vegetation coverage, VV polarization can usually provide more accurate information.

[0121] VH polarization means that the radar emits vertically polarized electromagnetic waves and receives horizontally polarized echoes. In this polarization mode, the interaction between radar waves and surface objects includes both vertical and horizontal components. Therefore, VH polarization is somewhat sensitive to reflecting the horizontal and vertical structural characteristics of surface objects. VH polarization may perform better in certain specific application scenarios, such as distinguishing different types of vegetation and monitoring surface roughness.

[0122] In some specific implementation manners, the present disclosure may be constructing and visualizing a time series, including:

[0123] Constructing a time series and extracting time series features, and exporting a CSV file;

[0124] Using the chart tool of the Google Earth Engine platform to draw a time series chart to visualize the spatial distribution of the vegetation index in the area over a period of time.

[0125] Step S3: Constructing and visualizing a time series.

[0126] Step S31: Constructing a time series and extracting time series features. As Figures 7 to 8 shown, for each scene of the image, calculate and extract the time series features of the vegetation index in the study area, convert the time series data into a table form and export a CSV file for subsequent analysis and visualization. The extracted time series features are shown in Table 1.

[0127] Table 1

[0128]

[0129] Step S32: Visualizing the time series. Using the chart tool of the Google Earth Engine platform to draw a time series chart to visualize the spatial distribution of the vegetation index in the area from January 1, 2020 to December 30, 2020, as Figure 8 shown.

[0130] In some specific implementation manners, the present disclosure may be trend analysis, including: obtaining the analysis result of the trend direction and vegetation situation through a linear regression equation.

[0131] In some specific implementation manners, the present disclosure may be based on decomposing the time series into a trend component, a seasonal component, and a residual component to analyze trends and seasonal variations.

[0132] Step S4: Time series analysis.

[0133] Step S41: Trend analysis, which can adopt the linear regression equation y = mx + b, as shown in Table 2 and Table 3.

[0134] Table 2

[0135] Horizontal axis Vertical axis Slope Intercept <![CDATA[R 2 <!-- 7 -->]]> Time (year and date) Vegetation index value Rate of change of vegetation index Initial value Goodness of fit

[0136] Table 3

[0137] Trend direction Slope Vegetation condition Upward trend m>0 The vegetation index increases with time, and the vegetation cover or health condition improves Downward trend m<0 The vegetation index decreases with time, and the vegetation cover or health condition deteriorates Stable trend m≈0 The vegetation index has no obvious change

[0138] R 2 The value reflects the ability of the trend line to introduce data.

[0139] In this embodiment, precise calculations are performed using the scikit-learn library of Python, and the calculation formula is:

[0140]

[0141] The results obtained are m≈0.0005 and b≈0.41.

[0142] Step S42: Mutation point analysis.

[0143] Step S43: BFAST is used to detect mutation points and analyze trends and seasonal variations. BFAST decomposes the time series into three components:

[0144] Trend component (Trend): The long-term change trend of the time series;

[0145] Seasonal component (Seasonal): The periodic change of the time series;

[0146] Residual component (Residual): The part of the time series that cannot be explained by the trend and seasonality.

[0147] In this embodiment, Python is used to extract trend, seasonal, and mutation point information. The number of segments of the trend component is set to 1, and the significance level of mutation point detection is set to 0.05. Finally, mutation points in vegetation cover (such as fires, droughts, land use changes) are detected.

[0148] In some specific implementation scenarios, the present disclosure can be to complete decision tree classification, including:

[0149] Using a machine learning algorithm based on a binary tree, establish rules for object-oriented decision trees and automatic threshold decision trees, implement decision tree classification, and evaluate the accuracy of the classification results.

[0150] In some specific implementation scenarios, the present disclosure can be to complete decision tree classification, including:

[0151] Compare each pixel with the training samples and classify it into similar sample classes according to different judgment rules to complete the supervised classification of the entire image;

[0152] Combined with the ENVI remote sensing image processing software, object-oriented decision tree classification;

[0153] Use the ENVI software to automatically obtain the extended tool RuleGen based on the decision tree rules of the CART algorithm to construct decision tree rules.

[0154] Step S5: Decision tree classification CART (Classification and Regression Trees).

[0155] Detailed description of CART in classification tasks: CART (Classification and Regression Trees) is a machine learning algorithm based on binary trees and is widely used in classification tasks.

[0156] The CART classification tree constructs a binary tree by recursively dividing the data set into two subsets. Each internal node represents a feature and its corresponding threshold, which is used to divide the data set into left and right subsets. The leaf node represents the final classification result. The goal of CART is to continuously divide the data to make each subset more and more pure, so as to achieve accurate classification of the data.

[0157] The CART classification tree is a powerful classification tool that constructs a binary tree by recursively selecting the optimal features and thresholds to divide the data set. It uses the Gini index or information gain as the division criterion and prevents overfitting through pruning. The CART classification tree has the advantages of being intuitive and easy to interpret, but attention also needs to be paid to overfitting and instability problems.

[0158] This embodiment combines the ENVI remote sensing image processing software, establishes the rules of object-oriented decision trees and automatic threshold decision trees, realizes decision tree classification, and makes an accuracy evaluation of the classification results. This method uses the grassland in Xiwu Zhu Muqin Banner as the root node, and the discrimination condition is: distinguish by the change trend of the backscattering coefficient and the radar vegetation index RVI.

[0159] Step S51: Before classification, through visual interpretation and field investigation, clearly define the image object categories of the sample plots on the remote sensing image. Secondly, select a certain amount of training samples according to the classification categories, calculate and statistically analyze the information of the training sample areas through the software, and at the same time train the decision function with these generated category determination methods to make the sub-categories to be classified meet the judgment requirements. Subsequently, use the trained decision function to classify other data to be classified. Compare each pixel with the training samples and divide it into similar sample classes according to different judgment rules, and the supervised classification of the entire image can be completed. The technical roadmap is as Figure 9 shown.

[0160] Step S52: Object-oriented decision tree classification. Use ENVI software to extract the thresholds of object classification categories, and use the image statistics function to extract the thresholds of classified ground objects for the selected objects. Then, formulate decision tree rules based on the interpreters' experience in ground object recognition. Finally, classify the remote sensing image according to the established rules.

[0161] In this embodiment, the Sentinel-1 radar data is used as the classification object. It is not difficult to see that the grassland ground objects in the image have good recognition effects. The grassland vegetation can be divided into four types of ground objects: forest land, grassland, wasteland, and wetland. After selecting the four types of objects, namely wetland, forest land, sandy land, and grassland, it is necessary to confirm the separability of the samples. Generally, the separability of the samples should not be lower than 1.8. When the separability of the samples is lower than 1.8, it is necessary to consider whether to classify this sample into other categories, otherwise the decision tree rules will be affected.

[0162] Step S53: Automatic threshold decision tree classification. Use the ENVI software to automatically obtain the extended tool RuleGen based on the decision tree rules of the CART algorithm to construct the decision tree rules. First, select the samples corresponding to the object-oriented decision tree classification method to facilitate the comparison of the classification results. Then, use the RuleGen module to run the definition of the decision tree classification rules and display the decision tree, and complete the automatic threshold decision tree classification in sequence. The result is as Figure 10 shown.

[0163] Compared with the prior art, the grassland vegetation index time series analysis classification method based on SAR remote sensing images in each embodiment of the present disclosure utilizes the powerful computing resources and massive geospatial data sets of the Google Earth Engine platform to overcome the problems of long cycle and data loss caused by weather influence. Define the study area using geometric tools, set the data type and perform preprocessing. By calculating radar vegetation indices such as RVI and DVI, visualize and construct a time series, and extract features from the time series, such as mean, variance, seasonal amplitude, trend, etc. Combine with ENVI remote sensing image processing software to establish rules for object-oriented decision trees and automatic threshold decision trees, realize decision tree classification, and evaluate the accuracy of the classification results. Combining the technical advantages of SAR remote sensing and the method advantages of time series analysis, it can realize large-scale and high-frequency dynamic monitoring of grassland vegetation. This method has broad application prospects in the fields of ecological monitoring, disaster assessment, policy formulation, etc., and is one of the hot directions in current remote sensing technology and ecological research.

[0164] Based on the above inventive concept, the method for classifying grassland vegetation index time series analysis based on SAR remote sensing images in various embodiments of the present disclosure at least obtains meteorological data, Sentinel-1 SAR data, and terrain data of the study area; constructs radar vegetation indices using backscattering coefficients; constructs and visualizes time series; performs time series analysis, including at least trend analysis and breakpoint analysis; completes decision tree classification, mainly using the sensitivity of SAR images to vegetation, combines time series analysis techniques to classify ground objects in the grassland, so as to use the Google Earth Engine platform for time series analysis and improved methods for ground object classification, uses Google Earth Engine to realize time series analysis of grassland vegetation index based on SAR remote sensing images, and classifies ground objects in the grassland according to time series characteristics.

[0165] The present disclosure also provides a device for classifying grassland vegetation index time series analysis based on SAR remote sensing images, including one or more processing modules configured to perform the classification of grassland vegetation index time series analysis based on SAR remote sensing images in the foregoing, and at least configured to be able to perform the specific implementation manners of steps S1 to S5.

[0166] The present disclosure also provides a computer-readable storage medium, on which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they mainly implement the method for classifying grassland vegetation index time series analysis based on SAR remote sensing images as described above, including at least:

[0167] Obtain meteorological data, Sentinel-1 SAR data, and terrain data of the study area;

[0168] Construct radar vegetation indices using backscattering coefficients;

[0169] Construct and visualize time series;

[0170] Perform time series analysis, including at least trend analysis and breakpoint analysis;

[0171] Complete decision tree classification.

[0172] The above embodiments are only exemplary embodiments of the present disclosure and are not used to limit the present disclosure. The protection scope of the present disclosure is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present disclosure, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present disclosure.

Claims

1. A method for classifying grassland vegetation indices based on the time-series analysis of SAR remote sensing images, including: Obtaining meteorological data, Sentinel-1 SAR data, and terrain data of the study area; Constructing radar vegetation indices using backscatter coefficients; Constructing and visualizing time series; Time series analysis, including at least trend analysis and breakpoint analysis; Completing decision tree classification.

2. The method according to claim 1, wherein Obtaining meteorological data, Sentinel-1 SAR data, and terrain data of the study area, including: Taking the sampling point location as the core, selecting typical plots to record key information including the longitude, latitude, altitude, and main vegetation types of each plot; Performing processing including meteorological interpolation, data mosaicking, and cropping; Loading SAR image data; Preprocessing of time-series dual-polarization Sentinel-1 data, including radiometric calibration, multi-look processing, registration, filtering processing, geocoding, and terrain correction.

3. According to the method described in claim 2, wherein Radiometric calibration includes: using the radiometric calibration information provided by the Sentinel-1 satellite to convert the original radar signal into the radiance intensity information of the ground object; Multi-look processing includes: by integrating SAR data of multiple pixels, weakening or smoothing the speckle noise effect in the data; Filtering processing includes: using a DeGrandi filter to process the noise and speckle in the time-series SAR image; Geocoding includes: converting the slant range coordinate system of the SAR data into a geographic coordinate system.

4. The method according to claim 3, wherein Constructing radar vegetation indices using backscatter coefficients, including: Constructing radar vegetation indices based on two polarization modes of backscatter coefficients, where: The radar vegetation index is sensitive to vegetation structure and biomass; the positive value of the dual-polarization vegetation index represents the vegetation area, and the negative value represents the non-vegetation area, which is applicable to areas with low vegetation coverage.

5. The method according to claim 4, wherein Based on two polarization modes of backscatter coefficients, including: VV polarization means that both the electromagnetic waves transmitted and received by the radar are vertically polarized; VH polarization means that the radar transmits vertically polarized electromagnetic waves and receives horizontally polarized echoes.

6. The method according to claim 5, wherein, Constructing and visualizing time series, including: Constructing time series and extracting time series features, and exporting a CSV file; Using the chart tool of the Google Earth Engine platform to draw time series charts and visualize the spatial distribution of vegetation indices in this area for a period of time.

7. The method according to claim 6, wherein, Trend analysis, including: Obtaining the analysis results of the trend direction and vegetation situation through a linear regression equation.

8. The method according to claim 7, wherein Breakpoint analysis, including: Based on decomposing the time series into trend components, seasonal components, and residual components, analyzing trends and seasonal variations.

9. The method according to claim 8, wherein, Completing decision tree classification, including: Using a machine learning algorithm based on a binary tree to establish rules for object-oriented decision trees and automatic threshold decision trees, realizing decision tree classification, and evaluating the accuracy of the classification results.

10. The method according to claim 9, wherein, Completing decision tree classification, including: Comparing each pixel with the training samples and classifying it into similar sample classes according to different judgment rules to complete the supervised classification of the entire image; Combining with the ENVI remote sensing image processing software for object-oriented decision tree classification; Use the ENVI software to automatically obtain the extended tool RuleGen based on the decision tree rules of the CART algorithm to implement the construction of decision tree rules.

Images