Automatic generation method and system of peanut samples based on multi-source TWDTW

Abstract

The present invention belongs to the field of agricultural remote sensing technology and relates to a method and system for automatically generating peanut samples based on multi-source TWDTW. The method comprises: establishing standard phenological curves corresponding to various characteristics of crops in various phenological zones; splicing the multi-source characteristic standard phenological curves; obtaining the multi-source characteristic phenological curves for the sample points to be tested; calculating the cumulative distance between the multi-source characteristic phenological curves and the multi-source characteristic standard phenological curves for each type of crop; and determining the crop type of the sample point to be tested, thereby obtaining sample data for that crop type. The present invention proposes a method for automatically generating peanut samples that combines multi-source remote sensing data with growth cycle characteristics. The method establishes the multi-source characteristic standard phenological curves, calculates the cumulative distance, and utilizes machine learning methods to improve the accuracy of early peanut identification and reduce the difficulty and cost of obtaining peanut samples.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural remote sensing technology, and in particular relates to a method and system for automatically generating peanut samples based on multi-source TWDTW. Background Art

[0002] Peanuts are an important oilseed and cash crop. Accurate and timely information on their spatial distribution is crucial for agricultural management, food security, and oilseed market forecasting and decision-making. Remote sensing is an effective tool for crop mapping, and supervised classification methods such as machine learning and deep learning are the primary approaches. However, the performance of these supervised classification methods is highly dependent on the quantity and quality of training samples. While traditional field surveys provide the most direct and reliable means of sample collection, they are time-consuming and costly, especially for mapping large areas during the early growing season. Obtaining sufficient samples early in the crop growing season is a key issue in improving the accuracy and timeliness of peanut mapping.

[0003] Currently, automated sample generation technology that combines remote sensing time-series data with crop phenotypes has become an effective means of addressing sample acquisition challenges. However, most methods primarily utilize optical data features and focus on extracting data from bulk crops (wheat and corn). Variations in crop electromagnetic radiation spectral characteristics are often more pronounced in complex growing regions and growing seasons, posing a challenge to the early identification of peanuts. This is particularly true given the similar growth periods of peanuts and root crops like sweet potatoes, which further complicates identification. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a method, system and electronic device for automatically generating peanut samples based on multi-source TWDTW.

[0005] In a first aspect, the present invention provides a method for automatically generating peanut samples based on multi-source TWDTW, comprising:

[0006] Several phenological zones are established based on the area size of each type of crop in the region. Based on the field survey sample point data and pre-processed remote sensing image time series data, standard phenological curves corresponding to the various characteristics of crops in each phenological zone are established;

[0007] The standard phenological curves with different characteristics in the same area and the same time period are spliced ​​into a multi-source characteristic standard phenological curve;

[0008] Obtain the land cover data of the points to be tested, randomly generate sample points within the cultivated land, calculate the time series data of the sample points to be tested, and obtain the multi-source characteristic phenological curves of the sample points to be tested;

[0009] Calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, and determine the threshold interval corresponding to the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve;

[0010] Calculate the cumulative distance between the multi-source characteristic phenological curve of the sample point to be tested and the multi-source characteristic standard phenological curve;

[0011] The crop type of the sample point to be detected is determined according to the size of the cumulative distance and the threshold interval, and the sample data of the crop type is obtained.

[0012] In a second aspect, the present invention provides a peanut sample automatic generation system based on multi-source TWDTW, comprising a standard phenological curve establishment unit, a splicing unit, a multi-source characteristic phenological curve generation unit, a first data processing unit, a second data processing unit and an output unit;

[0013] The standard phenological curve establishment unit is used to establish several phenological zones according to the area size of each type of crop in the region, and to establish the standard phenological curve corresponding to each characteristic of the crop in each phenological zone based on the field survey sample point data and the pre-processed remote sensing image time series data;

[0014] The splicing unit is used to splice the standard phenological curves with different characteristics in the same area and the same time period into a multi-source characteristic standard phenological curve;

[0015] A multi-source characteristic phenological curve generating unit is used to obtain land cover data of the points to be detected, randomly generate sample points in the cultivated land plot, calculate the time series data of the sample points to be detected, and obtain the multi-source characteristic phenological curve of the sample points to be detected;

[0016] The first data processing unit is configured to calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, and determine a threshold interval corresponding to the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve;

[0017] The second data processing unit is used to calculate the cumulative distance between the multi-source characteristic phenological curve of the sample point to be detected and the multi-source characteristic standard phenological curve;

[0018] The output unit is used to determine the crop type of the sample point to be detected according to the size of the cumulative distance and the threshold interval, and obtain sample data of the crop type.

[0019] On the basis of the above technical solution, the present invention can also be improved as follows.

[0020] Furthermore, remote sensing image time series data includes SAR (Synthetic Aperture Radar) data, multispectral remote sensing data and land cover data.

[0021] Furthermore, the preprocessing of crop remote sensing image time series data includes: filtering the remote sensing image time series data, converting the coordinate system into the world geodetic coordinate system, and resampling the image azimuth and range resolution to 10m; using the SG smoothing filter algorithm to smooth the time series and remove thermal noise; radiation correction; and terrain correction using DEM data.

[0022] Furthermore, crop characteristics include band characteristics, texture characteristics and vegetation index characteristics.

[0023] Furthermore, the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve is calculated, including: assuming that the time series of the multi-source characteristic phenological curve of the crop is , the length is , No. The feature is ; Assume that the time series of the multi-source characteristic standard phenological curve is , the length is , No. The feature is ;but:

[0024] ;

[0025] ;

[0026] set up For the Features and The base distance of the features, For the Features and The distance matrix of features is:

[0027] ;

[0028] Assume the time weight factor is , is the gain factor, is the distance factor, the distance factor is Features and The difference in the time series length of the features, , is the set ratio of the series time span, then:

[0029] ;

[0030] ;

[0031] set up For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, and exist The cumulative distance at the node is , find the shortest cumulative distance, the previous node is 、 and The minimum value between , then:

[0032] ;

[0033] Assume the minimum cumulative distance path is L, then:

[0034] .

[0035] Furthermore, when calculating the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, the first The cumulative distance of crops is , For the The weight of the cumulative distance of each crop, the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone is , No. The cumulative distance weight of each crop is determined by the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone, then:

[0036] ;

[0037] set up For the The cumulative distance between each standard phenological curve in a phenological zone is For the The weight of the cumulative distance of crops is:

[0038] .

[0039] Furthermore, the crop type of the sample point to be detected is determined according to the size of the cumulative distance and the threshold interval, and the initial classification sample is obtained. The time series of the point to be detected is resampled to be consistent with the standard time series of the crop.

[0040] Furthermore, the Manhattan distance between each initial classification sample and the corresponding labeled crop is calculated. If the Manhattan distance is greater than the set threshold, the initial classification sample is retained, otherwise the initial classification sample is discarded. Let the Manhattan distance be , is the time series of the points to be detected at time The eigenvalue at For crop standard time series in time The eigenvalue at is the number of data in the time series, then:

[0041] .

[0042] Furthermore, before calculating the Manhattan distance between each initial sample and the corresponding labeled crop, the time series of the point to be detected is resampled to be consistent with the standard time series of the crop.

[0043] The beneficial effects of the present invention are: the present invention proposes a method for automatically generating peanut samples by combining multi-source remote sensing data and growth cycle characteristics, establishes a multi-source characteristic standard phenological curve, calculates the cumulative distance between the multi-source characteristic phenological curve of the sample point to be detected and the multi-source characteristic standard phenological curve, and uses machine learning methods to improve the accuracy of early peanut identification and reduce the difficulty and cost of obtaining peanut samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of the automatic peanut sample generation method based on multi-source TWDTW provided in Example 1 of the present invention;

[0045] Figure 2 This is a comparison chart before and after SG smoothing filtering;

[0046] Figure 3 It is a multi-source TWDTW partition diagram, attached Figure 3 (a) is the multi-source TWDTW partition diagram, Figure 3 (b) is the normalized vegetation index curve of each crop, Figure 3 (c) is the contrast curve, Figure 3 (d) in the figure is the phenological curve of the three stages of crop growth;

[0047] Figure 4 This is a schematic diagram of the automatic peanut sample generation system based on multi-source TWDTW provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0049] Example 1

[0050] As an example, as shown in the attached Figure 1 As shown, to solve the above technical problems, this embodiment provides a method for automatically generating peanut samples based on multi-source TWDTW, including:

[0051] Several phenological zones are established based on the area size of each type of crop in the region. Based on the field survey sample point data and pre-processed remote sensing image time series data, standard phenological curves corresponding to the various characteristics of crops in each phenological zone are established;

[0052] The standard phenological curves with different characteristics in the same area and the same time period are spliced ​​into a multi-source characteristic standard phenological curve;

[0053] Obtain the land cover data of the points to be tested, randomly generate sample points within the cultivated land, calculate the time series data of the sample points to be tested, and obtain the multi-source characteristic phenological curves of the sample points to be tested;

[0054] Calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, and determine the threshold interval corresponding to the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve;

[0055] Calculate the cumulative distance between the multi-source characteristic phenological curve of the sample point to be tested and the multi-source characteristic standard phenological curve;

[0056] The crop type of the sample point to be detected is determined according to the size of the cumulative distance and the threshold interval, and the sample data of the crop type is obtained.

[0057] In the actual application, the SAR data selected were from the Sentinel-1 satellite ground-range multi-look imagery within the region. This imagery includes three spatial resolutions: 10m, 25m, and 40m, four polarization band combinations, and three imaging modes. Only images with the Interferometric Wide Swath (IW) scan mode and the Vertical-Horizontal (VH) and Vertical-Vertical (VV) polarization modes were selected. The multispectral remote sensing data used were Sentinel-2MSI L2A data (Sentinel-2 Multi-Spectral Instrument Level-2A). The required bands for the experiment were 10: visible (B2-B4), red edge (B5-B7), near-infrared (B8 and B8A), and shortwave infrared (B11 and B12). The land cover data uses the 10 m resolution global land cover map ground crop sample data as crop field sampling data, which includes latitude and longitude information and crop type labels.

[0058] Optionally, remote sensing image time series data includes SAR (Synthetic Aperture Radar) data, multispectral remote sensing data, and land cover data.

[0059] Optionally, preprocessing of remote sensing image time series data of crops includes: filtering the remote sensing image time series data, converting the coordinate system to the world geodetic coordinate system, and resampling the image azimuth and range resolution to 10m; using the SG smoothing filter algorithm to smooth the time series and remove thermal noise; radiation correction; and terrain correction using DEM data.

[0060] In actual application, after obtaining Sentinel-1 satellite remote sensing images in GEE (Google Earth Engine), in order to reduce or eliminate the error influence of speckle noise and retain the image edge details, the Lee-sigma algorithm is used to filter the Sentinel-1 satellite remote sensing images. At the same time, in order to facilitate the registration of heterogeneous images of Sentinel-1 satellite remote sensing images and multispectral remote sensing data, the coordinate system of Sentinel-1 satellite remote sensing images needs to be converted to the world geodetic coordinate system, and the image azimuth and range resolutions need to be resampled to 10m. Due to the influence of solar angle, observation angle, sensor sensitivity, bidirectional reflection of ground objects and aerosols, Sentinel-1 satellite remote sensing images and multispectral remote sensing data often contain a lot of noise. Therefore, time series smoothing through SG smoothing filter (Savitzky-Golay Smoothing Filter) can more easily reflect the changing trend reflected by the data itself, such as Figure 2 As shown, the horizontal axis is time (unit: day), the vertical axis is the normalized vegetation index, D is the original remote sensing image time series data, L1 is the linear interpolation curve, and L2 is the data after SG smoothing filter.

[0061] Optionally, the crop characteristics include band characteristics, texture characteristics, and vegetation index characteristics.

[0062] The basic principle of the TWDTW algorithm (Time Weighted Danymic Time Warping) is to find the best alignment between time series, calculate the cumulative distance, and then evaluate the similarity.

[0063] Optionally, the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve is calculated, including: assuming that the time series of the multi-source characteristic phenological curve of the crop is , the length is , No. The feature is ; Assume that the time series of the multi-source characteristic standard phenological curve is , the length is , No. The feature is ;but:

[0064] ;

[0065] ;

[0066] set up For the Features and The base distance of the features, For the Features and The distance matrix of features is:

[0067] ;

[0068] Assume the time weight factor is , is the gain factor, is the distance factor, the distance factor is Features and The difference in the time series length of the features, , is the set ratio of the series time span, then:

[0069] ;

[0070] ;

[0071] set up For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, and exist The cumulative distance at the node is , find the shortest cumulative distance, the previous node is 、 and The minimum value between , then:

[0072] ;

[0073] Assume the minimum cumulative distance path is L, then:

[0074] .

[0075] Optionally, when calculating the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, set The cumulative distance of crops is , For the The weight of the cumulative distance of each crop, the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone is , No. The cumulative distance weight of each crop is determined by the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone, then:

[0076] ;

[0077] set up For the The cumulative distance between each standard phenological curve in a phenological zone is For the The weight of the cumulative distance of crops is:

[0078] .

[0079] set up is the cumulative distance of corn and peanuts, is the cumulative distance of corn and sweet potato, is the cumulative distance between peanuts and sweet potatoes, is the total cumulative distance of the region, then:

[0080] .

[0081] When using the TWDTW algorithm to calculate the cumulative distances between standard phenological curves of different crops, there are differences in the time when the cumulative distances of standard phenological curves with different characteristics are the largest.

[0082] As attached Figure 3 As shown, L1 represents peanuts, L2 represents sweet potatoes, L3 represents corn, and Figure 3 (a) is a multi-source TWDTW partition diagram. The horizontal axis is the crop growth time (unit: day), with March 1, 2024 as the starting date, and the vertical axis is VV+VH (the sum of vertical-horizontal polarization and vertical-vertical polarization). Figure 3 (b) is the normalized vegetation index curve of each crop, the horizontal axis is time, unit: day, the vertical axis is the normalized vegetation index; Figure 3 (c) is the contrast curve, the horizontal axis is time, unit: day, the vertical axis is the contrast calculated using the near-infrared band; Figure 3 (d) in the figure divides the crop growth period into three stages, using different phenological curves for each stage. Differences in the standard phenological curves for SAR imagery are primarily concentrated in the early stages of crop sowing (0-80 days), differences in the Normalized Difference Vegetation Index (NDVI) for optical imagery are primarily concentrated in the crop growth and development period (80-130 days), and differences in texture features calculated using the near-infrared band of multispectral remote sensing data are primarily concentrated in the late stages of crop growth (after 130 days).

[0083] In order to make full use of the existing remote sensing image data to improve the accuracy of sample generation, this paper proposes to use the multi-source TWDTW algorithm to splice the standard phenological curves with different characteristics into a more differentiated multi-source characteristic standard phenological curve based on the characteristic sensitivity differences of crops in different growth stages. Figure 3 When the sample is generated, the phenological curve to be matched is also generated according to the different characteristics of different periods.

[0084] Optionally, the crop type of the sample point to be detected is determined according to the size of the cumulative distance and the threshold interval, and the initial classification sample is obtained, and the time series of the point to be detected is resampled to be consistent with the standard time series of the crop.

[0085] Optionally, calculate the Manhattan distance between each initial classification sample and the corresponding labeled crop. If the Manhattan distance is greater than the set threshold, the initial classification sample is retained; otherwise, the initial classification sample is discarded. Let the Manhattan distance be , is the time series of the points to be detected at time The eigenvalue at For crop standard time series in time The eigenvalue at is the number of data in the time series, then:

[0086] .

[0087] Optionally, before calculating the Manhattan distance between each initial sample and the corresponding labeled crop, the time series of the point to be detected is resampled to be consistent with the standard time series of the crop.

[0088] To verify the reliability of the present invention, field survey points and the method of the present invention were used to generate samples for the target area, and the sample generation results were overlaid and analyzed with the peanut plot map of the first year. The generated sample points basically all fell on the peanut plots, indicating that the method of the present invention generated samples with high accuracy. Further sample generation experiments were conducted using the existing second-year field sampling point data. If the area lacked field sampling data, the field sampling points in the area in previous years were used instead. The sample point generation results also proved that the method of the present invention generated samples with high accuracy using limited field sampling points to generate sample data.

[0089] This paper proposes a method for automatically generating peanut samples by combining multi-source remote sensing data and growth cycle characteristics. Through SAR data, multispectral remote sensing data and land cover data, a multi-source characteristic standard phenological curve is established. The cumulative distance between the multi-source characteristic phenological curve of the sample point to be tested and the multi-source characteristic standard phenological curve is calculated. Machine learning methods are used to improve the accuracy of early peanut identification and reduce the difficulty and cost of obtaining peanut samples.

[0090] Example 2

[0091] Based on the same principle as the method shown in Example 1 of the present invention, as shown in the attached Figure 4 As shown, an embodiment of the present invention further provides a peanut sample automatic generation system based on multi-source TWDTW, including a standard phenological curve establishment unit, a splicing unit, a multi-source characteristic phenological curve generation unit, a first data processing unit, a second data processing unit and an output unit;

[0092] The standard phenological curve establishment unit is used to establish several phenological zones according to the area size of each type of crop in the region, and to establish the standard phenological curve corresponding to each characteristic of the crop in each phenological zone based on the field survey sample point data and the pre-processed remote sensing image time series data;

[0093] The splicing unit is used to splice the standard phenological curves with different characteristics in the same area and the same time period into a multi-source characteristic standard phenological curve;

[0094] A multi-source characteristic phenological curve generating unit is used to obtain land cover data of the points to be detected, randomly generate sample points in the cultivated land plot, calculate the time series data of the sample points to be detected, and obtain the multi-source characteristic phenological curve of the sample points to be detected;

[0095] The first data processing unit is configured to calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, and determine a threshold interval corresponding to the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve;

[0096] The second data processing unit is used to calculate the cumulative distance between the multi-source characteristic phenological curve of the sample point to be detected and the multi-source characteristic standard phenological curve;

[0097] The output unit is used to determine the crop type of the sample point to be detected according to the size of the cumulative distance and the threshold interval, and obtain sample data of the crop type.

[0098] Optionally, remote sensing image time series data includes SAR (Synthetic Aperture Radar) data, multispectral remote sensing data, and land cover data.

[0099] Optionally, preprocessing of remote sensing image time series data of crops includes: filtering the remote sensing image time series data, converting the coordinate system to the world geodetic coordinate system, and resampling the image azimuth and range resolution to 10m; using the SG smoothing filter algorithm to smooth the time series and remove thermal noise; radiation correction; and terrain correction using DEM data.

[0100] Optionally, the crop characteristics include band characteristics, texture characteristics, and vegetation index characteristics.

[0101] Optionally, the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve is calculated, including: assuming that the time series of the multi-source characteristic phenological curve of the crop is , the length is , No. The feature is ; Assume that the time series of the multi-source characteristic standard phenological curve is , the length is , No. The feature is ;but:

[0102] ;

[0103] ;

[0104] set up For the Features and The base distance of the features, For the Features and The distance matrix of features is:

[0105] ;

[0106] Assume the time weight factor is , is the gain factor, is the distance factor, the distance factor is Features and The difference in the time series length of the features, , is the set ratio of the series time span, then:

[0107] ;

[0108] ;

[0109] set up For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, and exist The cumulative distance at the node is , find the shortest cumulative distance, the previous node is 、 and The minimum value between , then:

[0110] ;

[0111] Assume the minimum cumulative distance path is L, then:

[0112] .

[0113] Optionally, when calculating the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, set The cumulative distance of crops is , For the The weight of the cumulative distance of each crop, the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone is , No. The cumulative distance weight of each crop is determined by the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone, then:

[0114] ;

[0115] set up For the The cumulative distance between each standard phenological curve in a phenological zone is For the The weight of the cumulative distance of crops is:

[0116] .

[0117] Optionally, the crop type of the sample point to be detected is determined according to the size of the cumulative distance and the threshold interval, and the initial classification sample is obtained, and the time series of the point to be detected is resampled to be consistent with the standard time series of the crop.

[0118] Optionally, calculate the Manhattan distance between each initial classification sample and the corresponding labeled crop. If the Manhattan distance is greater than the set threshold, the initial classification sample is retained; otherwise, the initial classification sample is discarded. Let the Manhattan distance be , is the time series of the points to be detected at time The eigenvalue at For crop standard time series in time The eigenvalue at is the number of data in the time series, then:

[0119] .

[0120] Optionally, before calculating the Manhattan distance between each initial sample and the corresponding labeled crop, the time series of the point to be detected is resampled to be consistent with the standard time series of the crop.

[0121] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. The automatic generation method of peanut samples based on multi-source TWDTW is characterized by: include: Several phenological zones were established based on the planting structure of various types of crops in the region. Standard phenological curves corresponding to the various characteristics of crops in each phenological zone were established based on field survey sample point data and pre-processed remote sensing image time series data. Remote sensing image time series data includes SAR data, multispectral remote sensing data, and land cover data. Crop characteristics include band characteristics, texture characteristics, and vegetation index characteristics. The standard phenological curves with different growth period characteristics in the same area and time period are spliced ​​into a multi-source characteristic standard phenological curve; the crop growth period is divided into three stages, and different characteristic phenological curves are used for each stage; the differences in the standard phenological curves based on SAR image characteristics are mainly concentrated in the early stage of crop sowing, the differences in the normalized vegetation index characteristics of optical images are concentrated in the crop growth and development period, and the differences in texture characteristics calculated using the near-infrared band of multispectral remote sensing data are mainly concentrated in the late stage of crop growth; SAR image characteristics are selected in the early stage of crop sowing, the normalized vegetation index characteristics of optical images are selected in the growth and development period, and the texture characteristics calculated using the near-infrared band of multispectral remote sensing data are selected in the late stage of crop growth, and the whole multi-source characteristic standard phenological curve is spliced ​​together; Obtain the land cover data of the points to be tested, randomly generate sample points within the cultivated land, calculate the time series data of the sample points to be tested, and obtain the multi-source characteristic phenological curves of the sample points to be tested; Calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, and determine the threshold interval corresponding to the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve; Calculate the cumulative distance between the multi-source characteristic phenological curve of the sample point to be tested and the multi-source characteristic standard phenological curve; The crop type of the sample point to be detected is determined according to the size of the cumulative distance and the threshold interval, and the initial classification sample is obtained. The time series of the sample to be detected is resampled to be consistent with the standard time series of the crop, and the Manhattan distance between each initial classification sample and the corresponding labeled crop is calculated. If the Manhattan distance is greater than the set threshold, the initial classification sample is retained, otherwise it is discarded. Let the Manhattan distance be , is the time series of the points to be detected at time The eigenvalue at For crop standard time series in time The eigenvalue at is the number of data in the time series, then the Manhattan distance between the initial sample and the corresponding labeled crop is: 。 2. The method for automatically generating peanut samples based on multi-source TWDTW according to claim 1, characterized in that: The preprocessing of crop remote sensing image time series data includes: filtering the remote sensing image time series data, converting the coordinate system into the world geodetic coordinate system, and resampling the image azimuth and range resolution to 10m; using the SG smoothing filter algorithm to smooth the time series and remove thermal noise; radiation correction; and terrain correction using DEM data.

3. The method for automatically generating peanut samples based on multi-source TWDTW according to claim 1, characterized in that: Calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, including: assuming that the time series of the multi-source characteristic phenological curve of the crop is , the length is , No. The feature is ; Assume that the time series of the multi-source characteristic standard phenological curve is , the length is , No. The feature is ;but: ； ； set up For the Features and The base distance of the features, For the Features and The distance matrix of features is: ； Assume the time weight factor is , is the gain factor, is the distance factor, the distance factor is Features and The difference in the time series length of the features, , is the set ratio of the series time span, then: ； ； set up For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, For the Features and The node corresponding to each feature, and exist The cumulative distance at the node is , find the shortest cumulative distance, the previous node is 、 and The minimum value between , then: ； Assume the minimum cumulative distance path is L, then: 。 4. The method for automatically generating peanut samples based on multi-source TWDTW according to claim 1, characterized in that: When calculating the cumulative distance between the multi-source characteristic phenological curve and the multi-source characteristic standard phenological curve of each type of crop, The cumulative distance of crops is , For the The weight of the cumulative distance of each crop, the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone is , No. The cumulative distance weight of each crop is determined by the sum of the total distances of the standard phenological curves of each crop in the corresponding phenological zone, then: ； set up For the The cumulative distance between each standard phenological curve in a phenological zone is For the The weight of the cumulative distance of crops is: 。 5. The automatic generation system of peanut samples based on multi-source TWDTW is characterized by: It includes a standard phenological curve establishment unit, a splicing unit, a multi-source characteristic phenological curve generation unit, a first data processing unit, a second data processing unit and an output unit; The standard phenological curve establishment unit is used to establish several phenological zones based on the various types of crop planting structures in the region. Based on the field survey sample point data and pre-processed remote sensing image time series data, the standard phenological curve corresponding to the various characteristics of crops in each phenological zone is established. The remote sensing image time series data includes SAR data, multispectral remote sensing data, and land cover data. The crop characteristics include band characteristics, texture characteristics, and vegetation index characteristics. The stitching unit is used to stitch together standard phenological curves with different growth period characteristics in the same region and time period into a multi-source characteristic standard phenological curve. The crop growth period is divided into three stages, and each stage uses a phenological curve with different characteristics. The differences in the standard phenological curves based on SAR image characteristics are mainly concentrated in the early stage of crop sowing, the differences in the normalized vegetation index characteristics of optical images are concentrated in the crop growth and development period, and the differences in texture characteristics calculated using the near-infrared band of multispectral remote sensing data are mainly concentrated in the late stage of crop growth. SAR image characteristics are selected in the early stage of crop sowing, the normalized vegetation index characteristics of optical images are selected in the growth and development period, and the texture characteristics calculated using the near-infrared band of multispectral remote sensing data are selected in the late stage of crop growth, and the whole process is stitched together to form a complete multi-source characteristic standard phenological curve. A multi-source characteristic phenological curve generating unit is used to obtain land cover data of the points to be detected, randomly generate sample points in the cultivated land plot, calculate the time series data of the sample points to be detected, and obtain the multi-source characteristic phenological curve of the sample points to be detected; The first data processing unit is configured to calculate the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve, and determine the threshold interval corresponding to the cumulative distance between the multi-source characteristic phenological curve of each type of crop and the multi-source characteristic standard phenological curve; The second data processing unit is used to calculate the cumulative distance between the multi-source characteristic phenological curve of the sample point to be detected and the multi-source characteristic standard phenological curve; The output unit is used to determine the crop type of the sample point to be detected based on the size of the cumulative distance and the threshold interval, obtain the initial classification sample, resample the time series of the point to be detected to be consistent with the standard time series of the crop, calculate the Manhattan distance between each initial classification sample and the corresponding labeled crop, and retain the initial classification sample if the Manhattan distance is greater than the set threshold, otherwise discard the initial classification sample; let the Manhattan distance be , is the time series of the points to be detected at time The eigenvalue at For crop standard time series in time The eigenvalue at is the number of data in the time series, then the Manhattan distance between the initial sample and the corresponding labeled crop is: 。

Images