Satellite image-based tea garden identification method and device

Abstract

The application provides a tea garden identification method and device based on satellite images, which comprises the following steps: determining a evergreen vegetation area of a to-be-identified area based on Sentinel-2 image data and Landsat image data of the to-be-identified area; and identifying a tea garden area from the evergreen vegetation area through a decision tree model; wherein the decision tree model is constructed based on the following features and the classification threshold values corresponding to the features: a tea leaf phenology feature index, a terrain feature, and a spectral index determined by a separability index, wherein the SI is used to reflect the spectral reflectance separability of the tea garden and other evergreen vegetation; wherein the tea leaf phenology feature index is determined by an enhanced vegetation index in month N and a land surface water index in month M; the SI between the enhanced vegetation index in month N and the land surface water index in month M is the largest compared with the SI between the EVI in any other month and the LSWI in any other month, and N and M are integers greater than 0 and less than 13.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a tea garden recognition method and device based on satellite images. BACKGROUND

[0002] Tea tree as an important evergreen commercial tree species, its development space is expanding, the global demand continues to grow. Provide timely and accurate tea tree spatial distribution information is essential to improve the development of tea industry, improve the management level of tea industry, and then promote tea yield increase. Traditional field survey and agricultural census method has been used to collect tea garden area data in history. But this method is time-consuming, laborious, subjective, slow updating speed, and lack of detailed spatial distribution information. With the development of land remote sensing satellite, satellite remote sensing data has become an increasingly rich source of information.

[0003] The rapid development of remote sensing technology makes it possible to monitor vegetation in detail and in a timely manner. Compared with traditional survey methods, remote sensing has the advantages of wide monitoring range, strong spatio-temporal continuity and low cost. In the past few decades, satellite images have been widely used for monitoring the planting distribution of rapeseed, sugarcane, rice, corn and wheat. However, only a few studies have successfully extracted tea garden areas using high-resolution images, but these methods have strong limitations. There are more cloudy days in the south, which easily leads to cloud cover, missing data coverage and reduced spectral quality. The method of using machine learning to identify tea trees requires a large number of training sample sets, and is easily affected by temporal and spatial spectral changes in geographical features, climate conditions and observation conditions, which limits the spatial transferability and temporal reproducibility of automatic tea tree mapping algorithms. Due to the heterogeneity of tea garden landscape, artificial management intervention and frequent cloud cover, accurate identification and mapping of tea gardens is challenging.

[0004] How to provide a tea garden recognition method with high accuracy has become a problem to be solved. SUMMARY

[0005] The present application provides a tea garden recognition method and device based on satellite images, which solves the problem of low accuracy of tea garden recognition in the prior art and improves the accuracy of tea garden recognition.

[0006] The present application provides a tea garden recognition method based on satellite images, comprising:

[0007] Based on the Sentinel-2 image data and Landsat image data of the to-be-identified region, the evergreen vegetation region of the to-be-identified region is determined;

[0008] The tea garden region is identified from the evergreen vegetation region through a decision tree model;

[0009] The decision tree model is constructed based on the following features and the classification threshold values corresponding to the features.

[0010] The tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI, wherein the SI is used to reflect the separability of the spectral reflectance of the tea garden and other evergreen vegetation.

[0011] The tea phenology feature index TPI is determined by the enhanced vegetation index EVI of the Nth month and the land surface water index LSWI of the Mth month.

[0012] Compared with the SI between the EVI of any other month and the LSWI of any other month, the SI between the EVI of the Nth month and the LSWI of the Mth month is the largest, and N and M are integers greater than 0 and less than 13.

[0013] According to the tea garden recognition method based on satellite images provided by the application, the decision tree model is constructed based on the following steps:

[0014] The separability and threshold algorithm SEaTH is used to determine the classification threshold value corresponding to the tea phenology feature index TPI, the classification threshold value corresponding to the terrain feature DEM, and the classification threshold value corresponding to the spectral index determined by the separability index SI.

[0015] Based on the classification threshold value corresponding to the tea phenology feature index TPI, the classification threshold value corresponding to the terrain feature DEM, and the classification threshold value corresponding to the spectral index determined by the separability index SI, the decision tree model is constructed.

[0016] According to the tea garden recognition method based on satellite images provided by the application, the separability and threshold algorithm SEaTH is used to determine the classification threshold value corresponding to the tea phenology feature index TPI, the classification threshold value corresponding to the terrain feature DEM, and the classification threshold value corresponding to the spectral index determined by the separability index SI.

[0017] The classification threshold value corresponding to any one feature t in the tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI, wherein the SI is used to reflect the separability of the spectral reflectance of the tea garden and other evergreen vegetation, is T, wherein:

[0018]

[0019]

[0020] m1 and m2 represent the mean value of tea garden and other evergreen vegetation on feature t respectively; δ1 and δ2 represent the standard deviation of tea garden and other evergreen vegetation on feature t respectively; n1 and n2 represent the sample number of tea garden and other evergreen vegetation respectively.

[0021] According to the tea garden recognition method based on satellite images provided by the application, the calculation formula of the tea leaf phenology feature index TPI is as follows:

[0022] TPI = EVI N - LSWI M ;

[0023] Wherein, EVI N represents the EVI in N month, and LSWI M represents the LSWI in M month.

[0024] According to the tea garden recognition method based on satellite images provided by the application, the calculation formula of the separation index SI is as follows:

[0025]

[0026] m = {Blue, Green, Red, Nir, Swir1, Swir2, CB};

[0027] n = {"8, 28, 48, 58, 68, 98, 118, 128, 138, 158, 188, 198, 208, 218, 228, 238, 248, 258, 278, 288, 298, 308, 318, 328, 338, 348, 358"}.

[0028] Wherein, m represents six original spectral indices, including blue, green, red, Nir, Swir1, Swir2 bands and cumulative band CB; n represents the year DOY corresponding to the time series image; i and j represent tea garden and other evergreen vegetation respectively; and represent the mean value of tea garden and other evergreen vegetation on DOY index (m); and σ i and σ j represent the standard deviation of tea garden and other evergreen vegetation on DOY index (m); represent the spectral heterogeneity between classes; (σ i + σ j ) represents the spectral heterogeneity within the class;

[0029] The spectral index determined by the separation index SI includes a cumulative band CB of P months and a green band Green of Q months, wherein a difference between the cumulative band CB of P months and the green band Green of Q months is the largest compared with a difference between the cumulative band CB of other months and the green band Green of other months, P and Q are integers greater than 0 and less than 13.

[0030] According to the tea garden recognition method based on satellite images provided by the application, the Sentinel-2 image data and the Landsat image data of the region to be recognized are used to determine the evergreen vegetation region of the region to be recognized, and the method comprises the following steps:

[0031] The Sentinel-2 image data and the Landsat image data of the region to be recognized in the preset time duration are unified in bands and combined into images, and the median composite image is generated based on the image data in each preset sub-time duration within the preset time duration;

[0032] Based on the median composite image, the normalized difference vegetation index NDVI, the enhanced vegetation index EVI, the land surface water index LSWI, the modified normalized difference water index mNDWI, and the time distribution and seasonal information of the vegetation greenness and water content are calculated;

[0033] Based on the normalized difference vegetation index NDVI, the enhanced vegetation index EVI, the land surface water index LSWI, the modified normalized difference water index mNDWI, and the time distribution and seasonal information of the vegetation greenness and water content, the evergreen vegetation region of the region to be recognized is determined.

[0034] The application provides a tea garden recognition device based on satellite images, which comprises:

[0035] A determination module is configured to determine the evergreen vegetation region of the region to be recognized based on the Sentinel-2 image data and the Landsat image data of the region to be recognized;

[0036] A recognition module is configured to recognize the tea garden region from the evergreen vegetation region by using a decision tree model.

[0037] A decision tree model is constructed based on the following features and the classification threshold values corresponding to the features:

[0038] A tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separation index SI, wherein the SI is used to reflect the separation of the spectral reflectance of the tea garden and other evergreen vegetation.

[0039] The tea phenology feature index TPI is determined by the enhanced vegetation index EVI of N months and the land surface water index LSWI of M months.

[0040] Among them, the SI between the N-month enhanced vegetation index EVI and the M-month land surface water index LSWI is the largest compared with the SI between the EVI of any other month and the LSWI of any other month, N and M are integers greater than 0 and less than 13.

[0041] The application also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the tea garden identification method based on satellite images according to any one of the above descriptions when executing the program.

[0042] The application also provides a non-transitory computer-readable storage medium, which stores a computer program, wherein the computer program is executable on a processor to implement the tea garden identification method based on satellite images according to any one of the above descriptions.

[0043] The application also provides a computer program product, comprising a computer program, wherein the computer program is executable on a processor to implement the tea garden identification method based on satellite images according to any one of the above descriptions.

[0044] The application provides a tea garden identification method and device based on satellite images, which constructs a tea leaf phenology characteristic index, a terrain feature DEM, and a spectral index determined by a separability index SI, uses separability and threshold algorithms to construct a semi-automatic decision tree model, combines a TPI phenology characteristic index, terrain and original spectral features to identify tea trees, and is used for extracting tea garden distribution, wherein the TPI index is determined by calculating the difference between the EVI and LSWI of the two months corresponding to the maximum separability index, the separation of tea trees and other evergreen vegetation is strengthened, and the accuracy of tea garden identification is improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0046] Figure 1 is a flowchart of the tea garden identification method based on satellite images provided by the application;

[0047] Figure 2 is a schematic diagram of the tea garden identification based on satellite images provided by the application;

[0048] Figure 3 is a schematic diagram of the vegetation index provided by the application;

[0049] Figure 4 is a curve diagram of reflectivity of CB and Green band provided by the present application;

[0050] Figure 5 is a structural schematic diagram of a tea garden identification device based on satellite images provided by the present application;

[0051] Figure 6 is a structural schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0053] Classification algorithms based on phenology generate classification rules by using time indicators representing the growth and development rhythm of vegetation, showing the potential for large-scale vegetation mapping. However, due to the diversification of artificial management and the heterogeneity of tea garden landscape, mapping tea gardens is a unique challenge. During the tea picking period, tea buds are very small and contribute little to the spectral signal, making it difficult to accurately identify tea trees with only one period of phenological indicators. Tea trees and other evergreen vegetation have similar spectral reflectance, so it is difficult to extract obvious phenological indicators from time-series vegetation index curves. In addition, tea gardens are often a mixture of evergreen trees and cross-paths, resulting in an increase in mixed pixels in low and medium resolution images, ultimately affecting the accuracy of tea garden identification.

[0054] Affected by cloudy and rainy weather, most satellite data are contaminated by clouds, resulting in a large number of missing data available, making it difficult to apply.

[0055] The method of using machine learning to identify tea trees requires a large number of training samples and is easily affected by temporal and spatial spectral changes in geographical features, climate conditions and observation conditions, which limits the spatial transferability and temporal reproducibility of automatic tea tree mapping algorithms.

[0056] The method of simply using phenological differences or vegetation indices to identify tea gardens is difficult to improve the recognition accuracy in the case of complex southern surface vegetation, similar spectra, and greater impact of satellite data on clouds and rain.

[0057] The above defects all result in a decrease in the accuracy of tea garden identification.

[0058] In order to solve these problems, the present application provides a tea garden recognition and extraction method based on Sentinel-2 and Landsat satellite images, so as to realize simple and high-precision tea garden recognition monitoring. The algorithm uses Landsat7 / 8 28 and Sentinel-2 satellite images on the Google Earth Engine (GEE) platform, and uses a semi-automatic decision tree model to recognize tea gardens, effectively improving the accuracy of tea garden recognition.

[0059] The satellite image-based tea garden recognition method and device of the present application will be described below in conjunction with the accompanying drawings.

[0060] Figure 1 The present application provides a flowchart of a satellite image-based tea garden recognition method, as shown in Figure 1 The method comprises the following steps:

[0061] Step 100, based on the Sentinel-2 image data and Landsat image data of the to-be-recognized region, the evergreen vegetation region of the to-be-recognized region is determined;

[0062] Specifically, the present application can first obtain the top-of-atmosphere (TOA) reflectance data from Landsat-7 / 8 and Sentinel-2; it can be continuous data of a preset time length, such as image data collected in the study area from August 1, 2019 to July 31, 2021, obtained by accessing the Google Earth Engine (GEE) platform.

[0063] In addition, some ground verification data can be further obtained, such as the field investigation data in late May 2021, and 50 rows and 50 columns of fishnets are generated in the study area, and a point is randomly generated in each fishnet. Each point is surrounded by a 15-meter rectangular buffer zone, thereby forming 2500 30m*30m polygons in the entire study area. In the present application, 977 sample survey data are used for tea trees, 1329 for other evergreen vegetation, and 393 for crops.

[0064] Specifically, after obtaining the Sentinel-2 image data and Landsat image data of the to-be-recognized region, in order to ensure the reliability of the comprehensive data from Landsat-7ETM+, Landsat-8OLI and Sentinel-2MSI sensors, some preprocessing operations can be performed first to obtain data for recognizing the evergreen vegetation region from the to-be-recognized region.

[0065] Specifically, after the image data is preprocessed, the evergreen vegetation region can be identified from the to-be-identified region, and it should be noted that all kinds of methods for identifying the evergreen vegetation region from the to-be-identified region are applicable to the present application, such as a method of combining the seasonal characteristics of the time series vegetation index (VI) of each land cover type with a historical pixel-based algorithm to mask water bodies, buildings and bare soil to obtain the evergreen vegetation region, or a method of using machine learning to identify the four-season evergreen region in the Sentinel-2 image data and the Landsat image data of the to-be-identified region, and the present application does not limit this.

[0066] In step 110, the tea garden region is identified from the evergreen vegetation region by a decision tree model.

[0067] The decision tree model is constructed based on the following features and the classification threshold values corresponding to the features.

[0068] The tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI, which is used to reflect the separability of the spectral reflectance of the tea garden and other evergreen vegetation.

[0069] The tea phenology feature index TPI is determined by the enhanced vegetation index EVI in month N and the land surface water index LSWI in month M.

[0070] The SI between the enhanced vegetation index EVI in month N and the land surface water index LSWI in month M is the largest compared to the SI between the EVI in any other month and the LSWI in any other month, and N and M are integers greater than 0 and less than 13.

[0071] The present application constructs a semi-automatic decision tree model using separability and threshold algorithms by constructing a tea phenology feature index, a terrain feature DEM, and a spectral index determined by a separability index SI, and combines the TPI phenology feature index, terrain and original spectral features to identify tea trees for tea garden distribution.

[0072] Optionally, the evergreen vegetation region of the to-be-identified region is determined based on the Sentinel-2 image data and the Landsat image data of the to-be-identified region, and includes:

[0073] unify bands of the Sentinel-2 image data and the Landsat image data in a preset time length and combine the images, and generate a median composite image based on the image data in each preset sub-time length within the preset time length;

[0074] based on the median composite image, calculate a normalized difference vegetation index NDVI, an enhanced vegetation index EVI, a land surface water index LSWI, a modified normalized difference water index mNDWI, and time distribution and seasonal information of vegetation greenness and water content;

[0075] based on the normalized difference vegetation index NDVI, the enhanced vegetation index EVI, the land surface water index LSWI, the modified normalized difference water index mNDWI, and the time distribution and seasonal information of vegetation greenness and water content, determine an evergreen vegetation region of the region to be identified.

[0076] Figure 2 is a schematic diagram of tea garden identification based on satellite images provided by the present application, as Figure 2 After obtaining the Sentinel-2 image data and the Landsat image data of the region to be identified, in order to ensure the reliability of the comprehensive data from the Landsat-7 ETM+, Landsat-8 OLI and Sentinel-2 MSL sensors, some preprocessing operations can be performed first; for example, the OLS regression is used to adjust the reflectivity of the Landsat-7 ETM+ and the Sentinel-2 MSL to the Landsat-8 OLI. After the bands are unified, the images of the Landsat 7 / 8 and the Sentinel-2 are combined, and a median composite image is generated in each preset sub-time length (such as every 10 days). The images with an interval of 10 days can fill in the pixels with poor observation to a great extent. For the pixels with missing good observation values in the composite image, linear interpolation is performed on the adjacent high-quality observation values to fill in the blanks.

[0077] In the present application, four vegetation indices (VIs) are calculated using the median composite image to analyze the time distribution and seasonality of vegetation greenness and water content. These VIs are the normalized difference vegetation index (NDVI, formula (1)), the enhanced vegetation index (EVI, formula (2)), the land surface water index (LSWI, formula (3)) and the modified normalized difference water index (mNDWI, formula (4)). The cumulative band (CB) is used to represent the changes in surface reflectivity and spectral characteristics during the sprouting and picking processes of tea trees, and is calculated as the sum of the six original bands. The following formulas are used to calculate the VIs and CB based on the blue, green, red, near-infrared, SWIR1 and SWIR2 bands:

[0078]

[0079]

[0080]

[0081]

[0082] CB = p Blue + p Green + p Red + p NIR + p SWIR1 + p SWIR2 (5)

[0083] where p Blue , p Green , p Red , p NIR , p SWIR1 , p SWIR2 are the reflectance in the blue, green, red, near-red, SWIR1 and SWIR2 bands, respectively.

[0084] In the present application, the land cover types can be classified into four categories: evergreen vegetation, cropland, buildings and bare soil, and water.

[0085] Figure 3 is a schematic diagram of the vegetation index provided by the present application, as Figure 3 shown, after obtaining four vegetation indices (VIs) to analyze the temporal distribution and seasonality of vegetation greenness and water content, the seasonal characteristics of the time series vegetation index (VI) of each land cover type can be combined with historical pixel-based algorithms to mask water, buildings and bare soil.

[0086] wherein the water body can be identified using time series-based mNDWI, EVI and NDVI-based algorithms.

[0087] For example, pixels with mNDWI greater than NDVI or EVI for more than 75% of observations in a year and EVI less than 0.1 are considered as surface water pixels. In addition, buildings and bare soil are defined by pixels with LSWI values less than 0 for more than 90% of observations in a year using time series-based LSWI-based algorithms.

[0088] Evergreen vegetation includes other evergreen vegetation and tea plantations. A decision tree classification algorithm can be used in the present application to identify evergreen vegetation based on time series EVI and LSWI. The algorithm defines a pixel with more than 90% of the observations of LSWI>0 and minimum EVI>0.2 in a year as evergreen vegetation. Since the phenological characteristics of cultivated land and evergreen vegetation are very different, the present application adds a phenological index to the decision tree classification algorithm to improve the accuracy of evergreen vegetation identification. The time series harmonic analysis method (HANTS) is used to fit the NDVI time series, extract the key phenological indexes of evergreen vegetation and cultivated land, including the NDVI at the beginning of the growing season (SOS NDVI) and the NDVI at the end of the growing season (EOS NDVI); and identify the evergreen vegetation area.

[0089] Optionally, the decision tree model is constructed based on the following steps:

[0090] The separability and threshold algorithm SEaTH is used to determine the classification threshold corresponding to the tea phenological feature index TPI, the classification threshold corresponding to the terrain feature DEM, and the classification threshold corresponding to the spectral index determined by the separability index SI, respectively.

[0091] Based on the classification threshold corresponding to the tea phenological feature index TPI, the classification threshold corresponding to the terrain feature DEM, and the classification threshold corresponding to the spectral index determined by the separability index SI, respectively, the decision tree model is constructed.

[0092] The present application proposes a method for improving the recognition rate of tea plantations by using a tea plantation phenological feature index (TPI), and combining terrain features (DEM) and original spectral features (CBApr and GreenOct) to optimize the tea plantation recognition and mapping algorithm. By using the Google Earth Engine platform, the present application combines Landsat 7 / 8 and Sentinel-2 images through band coordination, and uses the SEaTH algorithm to calculate the classification thresholds of the four features. The semi-automatic decision tree model is used to extract the distribution of southern tea plantations, which can fully capture the high separability features that represent tea plantations in the case of insufficient images, and can automatically determine the optimal classification threshold of the decision tree model for tea plantation mapping to ensure that the obtained algorithm is reliable and reusable.

[0093] Optionally, the separability and threshold algorithm SEaTH is used to determine the classification threshold corresponding to the tea phenological feature index TPI, the classification threshold corresponding to the terrain feature DEM, and the classification threshold corresponding to the spectral index determined by the separability index SI, respectively.

[0094] The tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separation index SI, wherein the classification threshold corresponding to any one feature t of the separation of the tea garden and other evergreen vegetation spectral reflectance is T, and the calculation formula of the separation index SI is as follows:

[0095]

[0096]

[0097] m1 and m2 respectively represent the mean value of the tea garden and other evergreen vegetation on the feature t; δ1 and δ2 respectively represent the standard deviation of the tea garden and other evergreen vegetation on the feature t; n1 and n2 respectively represent the sample number of the tea garden and other evergreen vegetation.

[0098] Optionally, the calculation formula of the tea phenology feature index TPI is as follows:

[0099] TPI = EVI N - LSWI M ;

[0100] Wherein, EVI N represents the EVI in N month, and LSWI M represents the LSWI in M month.

[0101] For example, if the SI of the difference between the EVI in October and the LSWI in February is the highest, the calculation formula of the tea phenology feature index TPI is as follows:

[0102] TPI = EVI Otc - LSWI Feb ;

[0103] Wherein, EVI Otc represents the EVI in October, and LSWI Feb represents the LSWI in February.

[0104] Optionally, the calculation formula of the separation index SI is as follows:

[0105]

[0106] m = {Blue, Green, Red, Nir, Swir1, Swir2, CB};

[0107] n = {"8, 28, 48, 58, 68, 98, 118, 128, 138, 158, 188, 198, 208, 218, 228, 238, 248, 258, 278, 288, 298, 308, 318, 328, 338, 348, 358"}.

[0108] Wherein, m represents six original spectral indices, including blue, green, red, Nir, Swir1, Swir2 and cumulative band CB; n represents the year DOY corresponding to the time series image; i and j represent tea garden and other evergreen vegetation respectively; and represents the average value of the index (m) of tea garden and other evergreen vegetation in DOY; and i and j represents the standard deviation of the index (m) of tea garden and other evergreen vegetation in DOY; represents the spectral heterogeneity between classes; (σ i +σ j ) represents the spectral heterogeneity within the class;

[0109] The spectral index determined by the separation index SI includes the cumulative band CB in P month and the green band Green in Q month, wherein the difference between the cumulative band CB in P month and the green band Green in Q month is the largest compared with the difference between the cumulative band CB in other months and the green band Green in other months, and P and Q are integers greater than 0 and less than 13.

[0110] For example, Figure 4 is a curve example of the reflectivity of CB and Green bands provided by the application, the abscissa is date, and the ordinate is the reflectivity of CB and Green bands respectively; from the shadow part of Figure 4 , it can be known that the cumulative band in April and the green band in October are the most different periods of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.) in a year, so the two spectral reflectance characteristics CBApr and GreenOct can be used to effectively and more accurately distinguish the pixels of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.). Therefore, the spectral index determined by the separation index SI can include the cumulative band CB Apr in April and the green band Green Oct in October.

[0111] Specifically, the separation index (SI) is the ratio of the spectral variability between classes to the spectral variability within the class, which can reflect the separation of the spectral reflectance of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.). The calculation formula of SI index is as follows:

[0112]

[0113] m={Blue, Green, Red, Nir, Swir1, Swir2, CB};

[0114] n = {"8, 28, 48, 58, 68, 98, 118, 128, 138, 158, 188, 198, 208, 218, 228, 238, 248, 258, 278, 288, 298, 308, 318, 328, 338, 348, 358"};

[0115] where m represents six original spectral indices, including blue, green, red, Nir, Swirl, Swirl2 bands and cumulative band (CB); n represents the year (DOY) corresponding to the time series image; i and j represent tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.) respectively; and represents the average value of the index (m) of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.) at DOY; and and represents the standard deviation of the index (m) of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.) at DOY; represents the spectral heterogeneity between classes; (σ i +σ j ) represents the spectral heterogeneity within the class.

[0116] Wherein, the difference between EVI in October and LSWI in February has the highest SI, indicating that EVIOct-LSWIFeb is a suitable index for distinguishing tea garden from other evergreen vegetation (evergreen forest, evergreen forest, etc.), and therefore, the present application constructs a tea phenological feature index (TPI).

[0117] Wherein, TPI = EVI Otc -LSWI Feb ;

[0118] Separability and threshold (SEaTH) algorithm is used to automatically determine the optimal classification threshold of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.) based on Gaussian probability distribution formula.

[0119] In the present application, the TPI phenological feature index is used together with digital elevation model (DEM) and two selected spectral indices (such as selecting cumulative band in a certain month and green band in a certain month, which are the spectral indices with the largest difference between tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.) in a year, such as cumulative band in April (CBApr) and green band in October (GreenOct)), and SEaTH algorithm is used to automatically determine the classification threshold of each feature. Using the four features and their automatically calculated classification thresholds, a semi-automatic decision tree model is constructed to identify tea garden area. SEaTH algorithm can be calculated by the following formula:

[0120]

[0121]

[0122] Where T is the threshold value of feature t among the four features; m1 and m2 are the mean values of tea garden and other evergreen vegetation on feature t; "δ1" and "δ2" are the standard deviations of tea garden and other evergreen vegetation on feature t; n1 and n2 represent the sample numbers of tea garden and other evergreen vegetation.

[0123] For example, the classification threshold of the tea phenology feature index TPI can be calculated by the following formula:

[0124]

[0125]

[0126] Where T is the classification threshold of the tea phenology feature index TPI; m1 and m2 are the mean values of tea garden and other evergreen vegetation on the tea phenology feature index TPI; "δ1" and "δ2" are the standard deviations of tea garden and other evergreen vegetation on the tea phenology feature index TPI; n1 and n2 represent the sample numbers of tea garden and other evergreen vegetation.

[0127] In the present application, first, Landsat7 / 8 and Sentinel-2 images are preprocessed, then time series vegetation indices (VI) are used to detect evergreen vegetation. Then, a tea phenology feature index (TPI) is constructed to identify tea garden areas in the area to be identified based on a semi-automatic decision tree model. The TPI index is calculated by the difference between EVI in October (EVIOct) and LSWIFeb in February (LSWIFeb), and is based on the separable index (SI) to enhance the separation of tea garden and other evergreen vegetation (evergreen forest, evergreen forest, etc.). A semi-automatic decision tree model is constructed using the separability and threshold (SEaTH) algorithm, which combines the TPI phenology feature index, terrain and original spectral features to identify tea trees for tea garden distribution.

[0128] The present application can be used to improve the recognition rate of tea gardens by using a single satellite data source or simply the crop growth index or the difference in phenology to extract the distribution of tea gardens, which is difficult to extract. The present application proposes a tea garden phenology index (TPI) to improve the recognition rate of tea gardens, and combines terrain features (DEM) and original spectral features (CBApr and GreenOct) to optimize the tea garden mapping algorithm. Using the Google Earth Engine platform, the present application combines Landsat 7 / 8 and Sentinel-2 images through band coordination, and uses the SEaTH algorithm to calculate the classification threshold of the four features. Then, the present application uses a semi-automatic decision tree model to extract the distribution of tea gardens in the identified area.

[0129] In one embodiment, a certain city is used as the identified area, and a detailed land cover map of the city in 2020 is generated, including evergreen vegetation, farmland, buildings, and wasteland, and water bodies. Then, the present application uses the separability and threshold (SEaTH) algorithm to calculate the classification threshold of the three indicators of crops and evergreen vegetation: pixels with LSWI>0 in more than 90% of the observations in a year, pixels with SOS NDVI>0.45, and pixels with EOS NDVI>0.4.

[0130] The SEaTH algorithm is used to determine the optimal classification threshold of the four feature indexes in the TPI-based tea garden mapping algorithm, i.e., the tea phenology index (TPI), the digital elevation model (DEM), the cumulative band in April (CBApr), and the green band in October (GreenOct), which are used for tea gardens and other evergreen vegetation. The calculated classification thresholds of TPI, DEM, CBApr, and GreenOct are 0.5, 157, 0.87, and 0.08, respectively. When the TPI phenology index is set to a classification threshold of 0.5, 96.1% of the tea gardens are correctly identified, and only 3.9% are misidentified as evergreen vegetation. In addition, the spectral feature indexes (CBApr and GreenOct) and the DEM threshold can also effectively separate more than 90% of the tea gardens. Based on these accurate classification thresholds and the evergreen vegetation map of a certain city in 2020, the present application establishes a semi-automatic decision tree classification model to identify the tea gardens in the city. Pixels with TPI>0.5, DEM<157, CBApr>0.87, and GreenOct>0.08 are defined as tea gardens, and the rest are determined as evergreen vegetation, and the distribution map of tea gardens in the city is extracted.

[0131] The satellite image-based tea garden recognition device provided by the present application is described below. The satellite image-based tea garden recognition device described below can be referred to in conjunction with the satellite image-based tea garden recognition method described above.

[0132] Figure 5is a structural schematic view of a tea garden recognition device based on satellite images provided by the present application, as shown in the figure, the device 500 comprises: Figure 5

[0133] A determination module 510 is configured to determine a evergreen vegetation area of a to-be-recognized area based on Sentinel-2 image data and Landsat image data of the to-be-recognized area.

[0134] An identification module 520 is configured to identify a tea garden area from the evergreen vegetation area through a decision tree model.

[0135] The decision tree model is constructed based on the following features and the classification threshold values corresponding to the features.

[0136] A tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separability index SI, wherein the SI is used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation.

[0137] The tea phenology feature index TPI is determined by an enhanced vegetation index EVI in N months and a land surface water index LSWI in M months.

[0138] Compared with the SI between the EVI in any other month and the LSWI in any other month, the SI between the EVI in N months and the LSWI in M months is the largest, and N and M are integers greater than 0 and less than 13.

[0139] It should be noted that the tea garden recognition device based on satellite images can realize the above-mentioned method embodiments or the same or similar parts of the method embodiments.

[0140] The tea garden recognition device based on satellite images provided by the present application constructs a semi-automatic decision tree model by constructing a tea phenology feature index, a terrain feature DEM, and a spectral index determined by a separability index SI, using separability and threshold algorithms, and combines the TPI phenology feature index, terrain, and original spectral features to identify tea trees, and is used to extract tea garden distribution, wherein the TPI index is determined by calculating the difference between the EVI and LSWI corresponding to the two months with the largest separability index, which enhances the separation of tea trees and evergreen vegetation and improves the accuracy of tea garden recognition.

[0141] Figure 6 An example of an entity structure schematic diagram of an electronic device is shown in the figure, Figure 6 ​As shown, the electronic device can include a processor 610, a communications interface 620, a memory 630, and a communications bus 640, wherein the processor 610, the communications interface 620, and the memory 630 complete mutual communication through the communications bus 640. The processor 610 can invoke a logic instruction in the memory 630 to execute a tea garden identification method based on satellite images, including:

[0142] Based on the Sentinel-2 image data and the Landsat image data of the to-be-identified region, a evergreen vegetation region of the to-be-identified region is determined;

[0143] A tea garden region is identified from the evergreen vegetation region through a decision tree model;

[0144] The decision tree model is constructed based on the following features and the classification threshold values corresponding to the following features:

[0145] A tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separability index SI, wherein the SI is used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation;

[0146] The tea phenology feature index TPI is determined by an enhanced vegetation index EVI in N months and a land surface water index LSWI in M months;

[0147] Compared with the SI between the EVI in any other month and the LSWI in any other month, the SI between the EVI in N months and the LSWI in M months is the largest, and N and M are integers greater than 0 and less than 13.

[0148] In addition, the logic instruction in the memory 630 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0149] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program being stored in a non-transitory computer-readable storage medium and executable by a processor to enable a computer to perform the tea garden identification method based on satellite images provided by the above-mentioned methods, comprising:

[0150] Based on the Sentinel-2 image data and the Landsat image data of the to-be-identified region, a evergreen vegetation region of the to-be-identified region is determined;

[0151] A tea garden region is identified from the evergreen vegetation region by a decision tree model;

[0152] The decision tree model is constructed based on the following features and classification thresholds corresponding to the features:

[0153] A tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separability index SI, the SI being used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation;

[0154] The tea phenology feature index TPI is determined by an enhanced vegetation index EVI in N months and a land surface water index LSWI in M months;

[0155] Compared with the SI between the EVI in any other month and the LSWI in any other month, the SI between the EVI in N months and the LSWI in M months is the largest, N and M being integers greater than 0 and less than 13.

[0156] In another aspect, the present application also provides a non-transitory computer-readable storage medium, which stores a computer program, the computer program being executable by a processor to implement the tea garden identification method based on satellite images provided by the above-mentioned methods, comprising:

[0157] Based on the Sentinel-2 image data and the Landsat image data of the to-be-identified region, a evergreen vegetation region of the to-be-identified region is determined;

[0158] A tea garden region is identified from the evergreen vegetation region by a decision tree model;

[0159] The decision tree model is constructed based on the following features and classification thresholds corresponding to the features:

[0160] A tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separability index SI, the SI being used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation;

[0161] The tea phenology index TPI is determined by the enhanced vegetation index EVI of the Nth month and the land surface water index LSWI of the Mth month.

[0162] The SI between the EVI of the Nth month and the LSWI of the Mth month is the largest compared with the SI between the EVI of any other month and the LSWI of any other month, and N and M are integers greater than 0 and less than 13.

[0163] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0164] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software plus a necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0165] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying tea gardens based on satellite imagery, characterized in that, The method comprises the following steps: determining a region of evergreen vegetation in a to-be-identified region based on Sentinel-2 image data and Landsat image data of the to-be-identified region; identifying a tea garden region from the region of evergreen vegetation by using a decision tree model; wherein the decision tree model is constructed based on the following features and classification thresholds corresponding to the features: a tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separability index SI, wherein the SI is used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation; wherein the TPI is determined by an enhanced vegetation index EVI in the Nth month and a land surface water index LSWI in the Mth month; wherein the SI between the EVI in the Nth month and the LSWI in the Mth month is the largest compared with the SI between the EVI in any other month and the LSWI in any other month, and N and M are integers greater than 0 and less than 13; the decision tree model is constructed based on the following steps: determining the classification threshold corresponding to the TPI, the classification threshold corresponding to the DEM, and the classification threshold corresponding to the spectral index determined by the SI respectively by using a separability and threshold algorithm SEaTH; constructing the decision tree model based on the classification threshold corresponding to the TPI, the classification threshold corresponding to the DEM, and the classification threshold corresponding to the spectral index determined by the SI respectively; the determination of the classification threshold corresponding to the TPI, the classification threshold corresponding to the DEM, and the classification threshold corresponding to the spectral index determined by the SI respectively by using the SEaTH comprises: determining the classification threshold corresponding to any one of the TPI, the DEM, and the spectral index determined by the SI, wherein: ； ； m1 and m2 represent the mean values of the tea garden and the other evergreen vegetation respectively on the feature t; δ1 and δ2 represent the standard deviations of the tea garden and the other evergreen vegetation respectively on the feature t; and n1 and n2 represent the sample numbers of the tea garden and the other evergreen vegetation respectively; the calculation formula of the TPI is as follows: TPI = EVI N -LSWI M ; where EVI N represents EVI of N month, LSWI M represents LSWI of M month; the calculation formula of the SI is as follows: ； ； n={"8,28,48,58,68,98,118,128,138,158,188,198,208,218,228,238,248,258,278,288,298,308,318,328,338,348,358"}； where m represents six original spectral indices including blue, green, red, Nir, Swirl, Swirl2 bands and cumulative band CB; n represents the year DOY corresponding to the time series image; i and j represent tea garden and other evergreen vegetation respectively; and represents the average value of index (m) of tea garden and other evergreen vegetation at DOY; and and represents the standard deviation of index (m) of tea garden and other evergreen vegetation at DOY; represents the spectral heterogeneity between classes; represents the spectral heterogeneity within classes; the spectral index determined by the SI comprises a cumulative band CB in the Pth month and a green band Green in the Qth month, wherein the difference between the CB in the Pth month and the Green in the Qth month is the largest compared with the difference between the CB in any other month and the Green in any other month, and P and Q are integers greater than 0 and less than 13.

2. The satellite image based tea garden identification method as claimed in claim 1, wherein, the determination of the region of evergreen vegetation in the to-be-identified region based on the Sentinel-2 image data and the Landsat image data of the to-be-identified region comprises: unify bands and combine images of the Sentinel-2 image data and the Landsat image data of the to-be-identified region in a preset time length, and generate a median composite image based on the image data in each preset sub-time length within the preset time length; based on the median composite image, calculate a normalized difference vegetation index NDVI, an enhanced vegetation index EVI, a land surface water index LSWI, a modified normalized difference water index mNDWI, and time distribution and seasonal information of vegetation greenness and water content; based on the normalized difference vegetation index NDVI, the enhanced vegetation index EVI, the land surface water index LSWI, the modified normalized difference water index mNDWI, and the time distribution and seasonal information of vegetation greenness and water content, determine an evergreen vegetation region of the to-be-identified region.

3. A satellite image-based tea plantation identifying apparatus, characterized by, comprise: a determination module configured to determine an evergreen vegetation region of a to-be-identified region based on Sentinel-2 image data and Landsat image data of the to-be-identified region; an identification module configured to identify a tea garden region from the evergreen vegetation region by a decision tree model; wherein the decision tree model is constructed based on the following features and classification thresholds corresponding to the features: a tea phenology feature index TPI, a terrain feature DEM, and a spectral index determined by a separability index SI, the SI being used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation; wherein the tea phenology feature index TPI is determined by an enhanced vegetation index EVI of N months and a land surface water index LSWI of M months; wherein the SI between the EVI of N months and the LSWI of M months is the largest compared to the SI between the EVI of any other month and the LSWI of any other month, N and M being integers greater than 0 and less than 13; the decision tree model is constructed based on the following steps: determining classification thresholds corresponding to the tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI respectively by using a separability and threshold algorithm SEaTH; constructing the decision tree model based on the classification thresholds corresponding to the tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI respectively; the determination of the classification thresholds corresponding to the tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI respectively by using the separability and threshold algorithm SEaTH comprises: determining a classification threshold T corresponding to any one feature t of the tea phenology feature index TPI, the terrain feature DEM, and the spectral index determined by the separability index SI, the SI being used to reflect the separability of spectral reflectance of the tea garden and other evergreen vegetation, wherein: ； ； m1 and m2 represent the mean values of tea garden and other evergreen vegetation on feature t, respectively; δ1 and δ2 are the standard deviations of tea garden and other evergreen vegetation on feature t, respectively; n1 and n2 represent the sample numbers of tea garden and other evergreen vegetation, respectively; The calculation formula of the tea leaf phenology feature index TPI is as follows: TPI = EVI N -LSWI M ; where EVI N represents EVI of N month, LSWI M represents LSWI of M month; The calculation formula of the separation index SI is as follows: ； ； n={"8,28,48,58,68,98,118,128,138,158,188,198,208,218,228,238,248,258,278,288,298,308,318,328,338,348,358"}； where m represents six original spectral indices, including blue, green, red, Nir, Swirl, Swirl2 bands and cumulative band CB; n represents the year DOY corresponding to the time series image; i and j represent tea garden and other evergreen vegetation respectively; and represents the average value of index (m) of tea garden and other evergreen vegetation at DOY; and and represents the standard deviation of index (m) of tea garden and other evergreen vegetation at DOY; represents the spectral heterogeneity between classes; represents the spectral heterogeneity within the class; The spectral index determined by the separation index SI includes: the cumulative band CB of P months and the green band Green of Q months, wherein the difference between the cumulative band CB of P months and the green band Green of Q months is the largest compared with the difference between the cumulative band CB of any other month and the green band Green of any other month, P and Q are integers greater than 0 and less than 13.

4. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the tea garden identification method based on satellite images according to any one of claims 1-2 when executing the program.

5. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the tea garden identification method based on satellite images according to any one of claims 1-2 when executed by the processor.

6. A computer program product comprising a computer program, characterized in that, The computer program implements the tea garden identification method based on satellite images according to any one of claims 1-2 when executed by the processor.

Images