A crop extraction method for time-series SAR images based on CTLR and DTW K-means

Abstract

A method for crop extraction from time-series SAR images based on CTLR and DTW K-means, comprising: 1) performing polarization decomposition of the scattering matrix S of dual-polarization SAR time-series data based on the circular polarization transmission and linear polarization reception mode to obtain scattering components; 2) constructing a reduced polarization radar vegetation index using the scattering components, and constructing a standard time-series curve of the radar vegetation index of typical ground objects in the study area; 3) using the standard time-series curve of the radar vegetation index of typical ground objects in the study area as a reference, using the DTW algorithm to compare the similarity between the time-series curve of the pixel to be classified and the standard time-series curve; based on the DTW similarity, performing K-mean clustering iteration with a given K value and a random initial center, obtaining K clusters and corresponding cluster centers, and performing ground object classification. The method of the present invention overcomes the errors caused by image outliers when using single-time dual-polarization SAR images for crop classification, and can obtain full-caliber, high-precision remote sensing monitoring results of regional crop planting areas.

Description

Technical Field

[0001] The present invention relates to the field of crop remote sensing technology, and more particularly to a crop extraction method from time-series SAR images based on CTLR and DTW K-means. Background Art

[0002] Crop area monitoring is significantly challenged by the long-term and large-scale loss of optical data due to meteorological conditions such as clouds, rain, and fog, as well as the fragmentation of crop fields. Synthetic Aperture Radar (SAR) not only offers all-day, all-weather observation capabilities, unaffected by meteorological conditions, but also features side-view slant-range projection imaging, which is sensitive to crop and soil structure and characteristics. It has been applied in agricultural resource surveys, land resource utilization, and crop area monitoring, making it an optimal alternative to optical remote sensing data for crop remote sensing monitoring. In particular, C-band radar microwaves can reach the crop canopy and are highly sensitive to the three-dimensional structure of crop plants. They are multiply scattered between crop plants, such as flowers, fruits, and leaves, and contain information about crop plants at different growth stages. Furthermore, due to the fragmentation of crop fields, other land features such as forests and water bodies also exhibit seasonal variations, often displaying the same features as crops in single or multiple remote sensing images, making it difficult to extract crop area.

[0003] The microwave radiation emitted by crops themselves, as well as the interaction characteristics between crops and radar microwaves, are relatively complex. With the rapid development of radar remote sensing technology, which has generated a large amount of data, extracting target crop characteristics from radar remote sensing data has become a key issue in the application of microwave remote sensing to crops. Polarization decomposition technology can effectively obtain radar polarization characteristics, thereby more accurately extracting ground feature distribution information. However, polarization decomposition is mainly applied to the analysis of fully polarized SAR data, while some commonly used radar data, such as Sentinel-1 data, only have dual-polarization bands. In recent years, the emerging simplified polarization decomposition technology has provided new technical support for polarization decomposition and polarization feature analysis of dual-polarization SAR data. With the development of reduced polarimetric theory and methods, research on its applications and potential analysis are ongoing. Currently, three reduced polarimetric SAR operating modes have been proposed: the π / 4 mode, which transmits 45° linearly polarized waves and receives H and V linearly polarized waves; the dual circular polarization (DCP) mode, which transmits left-hand or right-hand circularly polarized waves and receives both left and right circularly polarized waves; and the hybrid polarimetric (HP) mode, which transmits left-hand or right-hand circularly polarized waves and receives H and V linearly polarized waves, also known as the circular transmit and linear receive (CTLR) mode. Compared to traditional linear dual-polarization SAR, reduced polarimetric SAR can store the phase of the echo signal, allowing for more flexible signal combination methods and thus obtaining richer scattering information. In many applications, it has achieved results comparable to those obtained with fully polarimetric SAR data. Furthermore, the radar vegetation index can reflect crop growth status and distribution and can be applied to the analysis of time-series radar data. Commonly used radar vegetation indices include: RVI (Radio Vibration Index) based on the ratio of the cross-polarization backscatter coefficient to the co-polarization backscatter coefficient; RVI based on the ratio of the coherence matrix eigenvalues; and RVI based on the ratio of volume scattering, dihedral scattering, and surface scattering using Freeman polarization decomposition. However, these three radar vegetation indices require polarization decomposition of fully polarized radar data, making them difficult to apply to dual-polarization radar data.

[0004] Crops exhibit specific, regular plant characteristics that change over time. Using time-series remote sensing imagery to establish relationships between crop plants at different phenological stages and extract crop planting areas has long been an important research direction and hotspot in agricultural remote sensing. The key to using time-series remote sensing imagery for land feature classification lies in similarity analysis of the time-series data. This involves using specific criteria to quantitatively evaluate the degree of similarity between the time series of the pixel under investigation and the reference time series of the target land classification. Representative similarity metrics include Euclidean distance (ED) and dynamic time warping distance (DTW). Due to factors such as sensor pixel variation and imaging noise, pixel values ​​at certain time points in time-series remote sensing data may be missing or abnormal. Therefore, using Euclidean distance to measure the similarity of sequences of unequal lengths can result in significant deviations. The DTW algorithm is a typical optimization method based on dynamic programming principles, primarily used to detect similarity between sequences. The DTW algorithm employs a time warping function that satisfies certain conditions to describe the temporal correspondence between the input sequence and the reference sequence, and then finds the warping function that minimizes the cumulative distance between the two sequences when they match. Compared with other similarity metrics, DTW overcomes the scale shift problem to some extent, solves the matching problem of time series with unequal time intervals, and is robust to outliers, thus achieving better matching results for similar features. DTW was first applied in speech recognition and video retrieval. In recent years, the algorithm, combined with the optical remote sensing vegetation index, has been applied to the classification of time-series optical remote sensing images and has been widely used for the classification and extraction of vegetation or land cover types. When the remote sensing time series comprises a large amount of data, using the DTW algorithm to traverse the time-series curves of all pixels to be classified and compare them one by one with a standard time-series curve, results in a large amount of inefficient computation and a high computational load. K-means is the most widely used dynamic iterative unsupervised classification method for remote sensing image classification. The K-means algorithm offers fast computational speed and excellent clustering results. Incorporating DTW into the K-means algorithm instead of the Euclidean distance, it considers both temporal and spatial information for remote sensing classification and target object extraction. This not only overcomes the errors caused by image outliers when using a single moment in time for object classification, but also reduces the workload of traversing and comparing the time-series curves of all pixels to be classified. Summary of the Invention

[0005] In response to the problems in the background technology, in order to obtain full-caliber, high-precision remote sensing monitoring results of regional crop planting area, the present invention constructs the CTLR reduced polarimetric decomposition radar vegetation index; combines the DTW and K-means classification methods to extract SAR remote sensing images of crop planting areas.

[0006] The present invention proposes a crop extraction method for time-series SAR images based on CTLR and DTW K-means, comprising: 1) polarization decomposition: performing polarization decomposition on the scattering matrix S of dual-polarization SAR time-series data based on the circular polarization transmission and linear polarization reception mode to obtain scattering components; 2) constructing a standard time-series curve: using the scattering components to construct a reduced polarimetric radar vegetation index, and combining ground sample points to construct a standard time-series curve of the radar vegetation index of typical ground objects in the study area; 3) performing time-series classification: referring to the standard time-series curve of the radar vegetation index of typical ground objects in the study area, using the DTW algorithm to compare the similarity between the time-series curve of the pixel to be classified and the standard time-series curve; based on the DTW similarity, performing K-means clustering iteration with a given K value and a random initial center to obtain K clusters and corresponding cluster centers, and then performing ground object classification.

[0007] Furthermore, the method of the present invention further comprises: in step 1), obtaining the even scattering R Dbl , volume scattering R Rnd and surface scattering R Odd Component; In step 2), using even scattering R Dbl , volume scattering R Rnd and surface scattering R Odd The simplified polarimetric radar vegetation index is constructed based on the components, and the standard time series curve of the radar vegetation index of typical ground objects in the study area is constructed by combining ground samples.

[0008] Furthermore, the method of the present invention also includes: before the polarization decomposition step, it also includes: preprocessing the dual-polarization SAR time series data and extracting the scattering matrix, wherein the preprocessing includes: orbit correction, radiation calibration, band synthesis and terrain correction.

[0009] Furthermore, the method of the present invention further comprises: in step 1), obtaining the even scattering R by the following formula Dbl , volume scattering R Rnd and surface scattering R Odd Serving size: R Rnd =(1-m)g0, Here, m represents polarizability and χ represents circularity.

[0010] Furthermore, the method of the present invention further comprises: in step 2), constructing the reduced polarimetric radar vegetation index (RVI) by the following formula: CTLR :

[0011] Furthermore, step 2) of the method of the present invention includes: editing, adding attributes, and projecting the ground sample points, and overlaying them on the time-series SAR remote sensing image to obtain the time-series remote sensing pixel values ​​corresponding to the ground sample point space; performing confidence analysis on the pixel values ​​at each moment, and taking the mean of the data within the confidence interval set at each moment as the standard time series curve value at that moment; connecting the values ​​at each moment to construct a standard time series curve of the radar vegetation index of typical ground objects in the study area.

[0012] Further, step 3) of the method of the present invention comprises:

[0013] 31) Use the DTW algorithm to compare the similarity between the time series curve of the pixel to be classified and the standard time series curve;

[0014] 32) Based on DTW similarity, a K-means clustering iteration is performed with a given K value and a random initial center to obtain K clusters and corresponding cluster centers and perform object classification.

[0015] Further, step 32) of the present invention includes:

[0016] 321) Given the number of clusters K, randomly select K objects as the initial cluster centers;

[0017] 322) Using DTW similarity distance as the distance metric, calculate the distance between each sample sequence and each center, and divide the sequence into the center closest to it to form K clusters;

[0018] Furthermore, step 32) of the present invention further includes:

[0019] 323) Recalculate the mean of each cluster as the new cluster center;

[0020] 324) Repeat 321)-333) until the movement amplitude of each cluster center is less than the set threshold to obtain the final time series data classification result.

[0021] Furthermore, step 3) of the present invention further includes: merging other types of land features to obtain the final crop planting distribution result.

[0022] The beneficial effects of the present invention are as follows: the method of the present invention can be applied to dual-polarization SAR remote sensing data, contains the mechanism of radar microwaves and crops, overcomes the error caused by image outliers when using single-time dual-polarization SAR images for crop classification, and can obtain full-aperture, high-precision regional crop planting area remote sensing monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to make the present invention more easily understood, the present invention will be described in more detail with reference to the specific embodiments shown in the accompanying drawings. These drawings only depict typical embodiments of the present invention and should not be considered as limiting the scope of protection of the present invention.

[0024] Figure 1 The present invention provides a technical roadmap for an embodiment of the method.

[0025] Figure 2 An overview of a study area used to validate the method of the present invention.

[0026] Figure 3 for Figure 2 Figure 2 shows the Sentine-1 dual-polarization SAR image of the study area.

[0027] Figure 4 The mean of the time series curves of typical features is shown.

[0028] Figure 5 The time series curves and mean values ​​for rapeseed are shown.

[0029] Figure 6 The remote sensing classification results of the study area are shown.

[0030] Figure 7-9 The ground sample map and the six sample plots for regional rapeseed extraction quadrat verification accuracy are shown. DETAILED DESCRIPTION

[0031] The following describes the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can better understand the present invention and implement it. However, the enumerated embodiments are not intended to limit the present invention. Unless there is a conflict, the following embodiments and the technical features in the embodiments can be combined with each other, wherein the same components are represented by the same figure marks.

[0032] In order to obtain full-aperture, high-precision remote sensing monitoring results of regional crop planting area, the method of the present invention first constructs the radar vegetation index based on the CTLR reduced polarimetric decomposition and analyzes the microwave scattering time series characteristics of typical landforms in the study area. Then, based on the polarimetric decomposition, the CTLR reduced polarimetric decomposition radar vegetation index is constructed using the even scattering, volume scattering and surface scattering components. Finally, combined with DTW and K-means classification, remote sensing extraction of regional crop planting distribution is performed based on dual-polarization SAR data.

[0033] Figure 1 The technical roadmap of the method of the present invention is shown, and the specific steps are as follows:

[0034] S1. Preprocess the dual-polarization SAR time series data and extract the scattering matrix S. The preprocessing includes orbit correction, radiation calibration, band synthesis, terrain correction, etc.

[0035] S2, perform polarization decomposition of the pre-processed dual-polarization SAR time series data scattering matrix S based on the circular polarization transmission line polarization receiving mode to obtain the even scattering R Dbl , volume scattering R Rnd and surface scattering R Odd Quantity.

[0036] S3, using even scattering R Dbl , volume scattering R Rnd and surface scattering R Odd The simplified polarimetric radar vegetation index is constructed based on the components, and then combined with ground sample points to construct the standard time series curve of the radar vegetation index of typical ground objects in the study area.

[0037] S4. Using the standard time-series curve of the radar vegetation index for typical features in the study area, the DTW algorithm was used to compare the similarity between the time-series curve of the pixel to be classified and the standard time-series curve. Based on the DTW similarity, a K-means clustering algorithm was performed with a given K value and random initial centers to obtain K clusters and their corresponding cluster centers for feature classification. Other features were then merged to obtain the final crop planting distribution results.

[0038] The method of the present invention is described in more detail below.

[0039] In step S2, even scattering R is obtained Dbl , volume scattering R Rnd and surface scattering R Odd The process of componentization is as follows:

[0040] From the reduced polarization Stokes vector or covariance matrix, characteristic parameters with clear physical meaning can be derived, including the polarization degree m, the phase difference δ between the two orthogonal polarization components, and the circularity χ. The polarization degree m characterizes the randomness of the scattering; the phase difference δ distinguishes between even-order scattering and odd-order scattering; and the circularity χ reflects the proportion of surface scattering and even-order scattering in a fully polarized wave.

[0041] The scattering matrix S of VV and VH dual-polarization SAR data is:

[0042]

[0043] Scattering Vector From S we can get:

[0044]

[0045] Where T represents matrix transpose, the subscript of the electromagnetic field E represents the polarization mode, the subscript RH represents right-hand circular polarization transmission and horizontal polarization reception, and the subscript RV represents right-hand circular polarization transmission and vertical polarization reception.

[0046] Based on the scattering vector The compact polarization covariance matrix C2 can be obtained as

[0047]

[0048] Where * represents complex conjugation and <.> represents spatial statistical averaging.

[0049] The Stokes vector expression in the CTLR reduced polarization mode uses g = [g0 g1 g2 g3] T express:

[0050]

[0051] Where, It means taking the real part of a complex number; represents the imaginary part of the complex number; g0 represents the total power of the electromagnetic wave, g1 represents the power value of the horizontal or vertical linear polarization component, g2 represents the power value of the linear polarization component when the inclination angle is 45° or 135°, and g3 represents the power value of the right-hand circular polarization component.

[0052] The polarization degree m, relative phase δ, and circularity χ are obtained from the Stokes vector, as shown in Equations (5) to (7):

[0053]

[0054]

[0055]

[0056] The CTLR polarization decomposition expressions of m-χ decomposition are shown in Equations (8) to (10):

[0057]

[0058] R Rnd =(1-m)g0 (9)

[0059]

[0060] Where R Dbl , R Rnd , R Odd They represent even scattering, volume scattering and surface scattering components respectively, which are different scattering mechanism components of the scattering target.

[0061] In step S3, the process of constructing the reduced polarimetric radar vegetation index is as follows:

[0062] The radar incident wave enters the vegetation and is scattered multiple times. Its scattered echo is a random scattered wave. The effect of vegetation on the radar is mainly volume scattering. Therefore, the larger the proportion of volume scattering in the total scattering, the more likely it is vegetation. In order to better distinguish the characteristics of crops and other typical land features in the study area, the present invention constructs the CTLR Reduced Polarization Decomposition Radar Vegetation Index. In one embodiment, the CTLR Reduced Polarization Decomposition Radar Vegetation Index (RVI) CTLR As shown in formula (11).

[0063]

[0064] Where R Dbl represents the even-order scattered component, R Rnd represents the volume scattering component, R Odd Represents the surface scattering component.

[0065] RVI CTLR The theoretical value range is [0,1]. When the radar wave irradiation area is water or bare ground, the volume scattering component P v Theoretically, it tends to 0, RVI CTLR The value of also tends to 0; when the radar observation field includes forest land or crops, the radar microwaves penetrate the vegetation or crop canopy to generate single scattering with the ground and the radar microwaves are incident on the ground and reflected by the trunk (stem) to generate dihedral angle reflection echo energy are reduced, and the surface scattering P s and even scattering P d The components are all reduced, and the volume scattering component P v Alternatively, different scattering components can be used to construct the CTLR reduced polarization decomposition radar index.

[0066] The ground sample points were then edited, attributes were added, and projections were transformed. The images were then overlaid onto the time-series SAR remote sensing imagery to obtain the time-series remote sensing pixel values ​​corresponding to the ground sample points. A confidence analysis was performed on the pixel values ​​at each moment, and the mean of the data within the 90% confidence interval was taken as the standard time series curve value at that moment. The values ​​at each moment were connected to construct a standard time series curve for the radar vegetation index of typical land features in the study area.

[0067] The following describes in detail step S4: performing DTW K-means clustering.

[0068] DTW is essentially a template matching algorithm. When there are a lot of data making up the remote sensing time series, the DTW algorithm is used to traverse the time series curves of all pixels to be classified and compare them with the standard time series curves one by one. There are a lot of invalid calculations and the amount of calculation is large. K-means is the most widely used dynamic iterative solution unsupervised classification method in the current remote sensing image classification. The K-means algorithm has a fast calculation speed and good clustering effect. The present invention introduces DTW into the K-means algorithm to replace the Euclidean distance, taking into account the spatiotemporal information for remote sensing classification and target crop extraction. It can not only overcome the errors caused by image outliers when using single-time dual-polarization SAR for crop classification, but also reduce the workload of traversing and comparing the time series curves of all pixels to be classified.

[0069] The DTW K-means classification method is mainly divided into two steps: 1) using the DTW algorithm to compare the similarity between the time series curve of the pixel to be classified and the standard time series curve; 2) based on the DTW similarity, a K value is given and a random initial center is used to perform K-means clustering iteration to obtain K clusters and corresponding cluster centers and perform ground object classification.

[0070] The principle of DTW algorithm is to give two sequences T = {T1, T2, T3, ..., T n} and R={R1,R2,R3,…,R n}, whose lengths are m and n respectively. Define the distance matrix D m×n , the matrix elements is the element q in T i and the element r in R j The Euclidean distance.

[0071]

[0072] In the matrix D m×n In the algorithm, a set of adjacent matrix elements (W) is defined as a curved path, the starting and ending elements of the path are the two end elements of the diagonal line of the distance matrix, and the continuity and monotonicity constraints are satisfied.

[0073]

[0074] Where max(m,n)≤K≤m+n-1. DTW uses dynamic programming to find the path with the smallest cumulative distance, which is the optimal temporal distance.

[0075] The principle of the K-means classification algorithm based on DTW similarity is to randomly select k objects as the initial cluster centers given the number of clusters k. Using DTW similarity distance as the distance metric, the distance between each sample sequence and each center is calculated, and the sequence is divided into the center closest to it, forming k clusters C = {C1, C2, C3, ..., C n}. Then recalculate the mean of each cluster as the new cluster center. Repeat the above steps until the movement of each cluster center is less than the set threshold to obtain the final time series data classification result.

[0076] The method of the present invention has been verified.

[0077] The study area is selected as follows. The rapeseed in southern Hunan in the middle and lower reaches of the Yangtze River in China was taken as the research object to verify the feasibility and applicability of the proposed rapeseed planting area extraction method based on the CTLR polarization decomposition vegetation index and DTW K-means classification. The total area of ​​the study area is 12,771 square kilometers. According to the 2020 Hunan Statistical Yearbook (http: / / tjj.hunan.gov.cn / hntj / index.html), the rapeseed sowing area in the study area is 140,800 hectares, accounting for about a quarter of the total rapeseed sowing area in southern Hunan. It is the main production area of ​​rapeseed in southern Hunan. The study area is distributed along the Xiangjiang River. The terrain is hilly and located in the subtropical monsoon climate zone. The soil type is mainly red soil, and the main crop planting system is winter rapeseed-single-season rice double-cropping system. Overview of the study area is as follows Figure 2 Winter rapeseed in the study area is sown from October each year to harvested in mid-May of the following year: the seedling stage is from late November to December, and the bud stage, flowering stage, silique stage, and maturity stage of winter rapeseed are in early January, early March, early April, and early May of the following year, respectively.

[0078] The remote sensing data used are as follows: The experimental remote sensing data uses Sentine-1 SLC (Single Look Complex) data covering the study area. The data is in Interferometric Wide Swath (IW) mode and has VV and VH dual polarization. This paper uses a total of 13 Sentine-1 remote sensing images to construct time series data. The imaging times are December 9, 2020, December 21, 2020, January 2, 2021, January 14, 2021, January 26, 2021, February 7, 2021, February 19, 2021, March 3, 2021, March 15, 2021, March 27, 2021, April 8, 2021, April 20, 2021 and May 2, 2021. The relative orbit number is 11. The rapeseed growth period and Sentine-1 image list of the study area are as follows: Figure 3 The present invention uses SNAP software to perform orbit correction, radiometric calibration, band synthesis, and terrain correction on the downloaded Sentinel-1 time series data, and resamples its spatial resolution to 20m×20m.

[0079] The ground sampling data was collected on March 3, 2021, during peak flowering, when rapeseed plant characteristics are most pronounced. A total of 94 rapeseed sampling points and six ground quadrats were collected throughout the study area to verify the accuracy of regional rapeseed extraction and to examine the stability of the CTLR-based Reduced Polarimetric Decomposition Radar Vegetation Index and the feasibility of extracting rapeseed planted areas using DTW K-means classification. The ground sampling points were edited, attributes were added, and projections were converted. The data was then overlaid on the experimental Sentinel-1 remote sensing imagery time series data.

[0080] The verification results are as follows.

[0081] 1. RVI CTLR Results of time series data analysis

[0082] The Sentnel-1 time series data from December 2020 to May 2021 in the study area were preprocessed, including orbit correction, radiation calibration, band synthesis and terrain correction, and the scattering matrix S was extracted. The scattering matrix S of the preprocessed Sentinel-1 time series data was decomposed based on CTLR reduced polarization to obtain the even scattering R Dbl , volume scattering R Rnd and surface scattering R Odd Component. The simplified polarimetric radar vegetation index was constructed using even scattering, volume scattering and surface scattering components. Combined with ground samples, the standard time series curve of the radar vegetation index of typical land features in the study area was constructed. The mean time series curves of five typical land features, namely water bodies, bare land, rapeseed, woodland and buildings, are shown in the figure below. Figure 4-5 shown.

[0083] from Figure 4 The average values ​​of the typical landform time series curves (100 points each) show that among the five typical landforms in the study area, the RVI of water bodies, bare land, forest land and buildings is CTLR The value changes less with time. RVI of water bodies and bare land CTLR The value is small, and the water body RVI CTLR The value is lower than the bare land RVI CTLR This is because both water bodies and bare land appear mainly as surface scattering in radar images, and the surface scattering component P s Large, volume scattering component P v Smaller; RVI of the building CTLR The value is in the middle. This is because the building mainly appears as even scattering and volume scattering in the radar image. The even scattering component P d and volume scattering component P v The values ​​are large, acting on RVI respectively CTLR The values ​​in the numerator and denominator make the RVI of the building CTLR The value is in the middle; RVI of forest landCTLR The value is relatively high because the forest land mainly appears as volume scattering in radar images, and the volume scattering component P v Higher. RVI of rapeseed CTLR The value changes greatly in time series, and the RVI of rapeseed CTLR值 As rapeseed grows and develops, the rapeseed plant is small in size when it is just sown, and the volume scattering component P shown on the radar image is v As the rape blossoms and the pods grow, the volume of the rape increases, and the volume scattering component P v Especially in the early and late stages of flowering, when rapeseed flowers or siliques grow, the plant volume increases significantly, RVI CTLR The value also increases significantly. In the late stage of rapeseed silique, the photosynthetic organs of rapeseed gradually transform into rapeseed siliques, the leaves begin to decline, the plant volume becomes smaller, and the RVI CTLR In addition, Figure 5 The time series curve and mean of the rapeseed radar vegetation index show that due to the differences in rapeseed varieties, growth environments and field management measures in the study area, there are also certain differences in the growth period of rapeseed, especially the differences in flowering period and leaf decline period. CTLR Rapeseed and other typical landforms in the study area can be distinguished in time series.

[0084] 2. Extraction results of rapeseed planting areas

[0085] The present invention is based on the typical ground feature RVI of the structure CTLR Time series curve, remote sensing image classification and rapeseed planting area extraction for the entire study area. In the selection of time series combination, the accuracy of remote sensing image classification and rapeseed extraction (the overall accuracy of remote sensing image classification is greater than 60%, and the overall accuracy of rapeseed extraction is greater than 70%) and remote sensing classification efficiency (small amount of time series data, early data acquisition time) are taken into account. The present invention uses 7 scenes of Sentinel-1 remote sensing data on January 2, 2021, February 7, 2021, February 19, 2021, March 3, 2021, March 15, 2021 and March 27, 2021 and combines them for remote sensing image classification and rapeseed extraction applications for the entire study area. The remote sensing image classification results are as follows: Figure 6 shown.

[0086] The accuracy of rapeseed extraction results in the study area was verified using 94 rapeseed sample points and 6 sample plots measured on March 3, 2021. Of the 94 rapeseed sample points measured on the ground, 69 sample points were accurately classified as rapeseed, with an overall accuracy of 73.4%. The accuracy of rapeseed extraction results was verified using 6 sample plots. The accuracy of ground sample plots and regional rapeseed extraction sample plots is shown in Figure 2. Figure 7-9 and as shown in Table 1.

[0087] Table 1 Verification accuracy of regional rapeseed extraction quadrat

[0088]

[0089] Depend on Figure 7-9 The plots of the ground samples shown show that most rapeseed plots were accurately classified as rapeseed, while some were misclassified as rapeseed and other ground features. The accuracy of the rapeseed extraction samples from the region in Table 1 was verified. Looking at the six individual plots, the overall accuracy ranged from 42.15% to 81.88%, and the rapeseed F-1 coefficient ranged from 58.33% to 90.04%. Quadrangle 3, which had the lowest overall accuracy and rapeseed F-1 coefficient, was the smallest of the six plots. Since Sentinel-1 SAR imagery has a spatial resolution of only 20 meters, its ability to resolve small plots is relatively poor. Looking at the six plots as a whole, the overall accuracy and rapeseed F-1 coefficient were 68.20% and 80.41%, respectively. The above research results show that the rapeseed planting area extraction method based on the CTLR simplified polarimetric decomposition radar vegetation index and DTW K-means classification has high accuracy in extracting rapeseed planting areas in the study area. It also shows the stability of the CTLR simplified polarimetric decomposition radar vegetation index and the feasibility of DTW K-means classification in time series extraction of rapeseed planting areas.

[0090] The embodiments described above are merely preferred embodiments of the present invention. The phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments" used in this specification may refer to one or more of the same or different embodiments of the present disclosure. Any common changes and substitutions made by those skilled in the art within the scope of the present invention are intended to be encompassed within the scope of protection of the present invention.

Claims

1. A method for crop extraction from time-series SAR images based on CTLR and DTWK-means, characterized by: include: 1) Polarization decomposition: scattering matrix of dual-polarization SAR time series data S Perform polarization decomposition based on circular polarization transmission and linear polarization receiving mode to obtain the scattered components: even scattering R Dbl , volume scattering R Rnd and surface scattering R Odd Quantity; 2) Constructing a standard time series curve: Using the scattering component to construct the simplified polarimetric radar vegetation index, combined with ground sample points, a standard time series curve of the radar vegetation index of typical land features in the study area is constructed, including: editing the ground sample points, adding attributes, projecting, and overlaying them on the time series SAR remote sensing image to obtain the time series remote sensing pixel values ​​corresponding to the ground sample point space; performing confidence analysis on the pixel values ​​at each moment, taking the mean of the data within the 90% confidence interval at each moment as the standard time series curve value at that moment; connecting the values ​​at each moment to construct a standard time series curve of the radar vegetation index of typical land features in the study area, wherein the simplified polarimetric radar vegetation index is constructed by the following formula RVI CTLR : ； 3) Perform time series classification: 31) Refer to the standard time series curve of the radar vegetation index of typical objects in the study area, and use the DTW algorithm to compare the similarity between the time series curve of the pixel to be classified and the standard time series curve; 32) Based on DTW similarity, give a K value and a random initial center to perform K-mean clustering iteration to obtain K clusters and corresponding cluster centers and perform object classification, including: 321) Given the number of clusters K, randomly select K objects as initial cluster centers; 322) Using DTW similarity distance as the distance metric, calculate the distance between each sample sequence and each center, and divide the sequence into the center closest to it to form K clusters.

2. The method for crop extraction from time-series SAR images according to claim 1, wherein: Before the polarization decomposition step, the method further includes: preprocessing the dual-polarization SAR time series data and extracting a scattering matrix, wherein the preprocessing includes: orbit correction, radiation calibration, band synthesis and terrain correction.

3. The method for crop extraction from time-series SAR images according to claim 1, wherein: In step 1), the even scattering is obtained by the following formula R Dbl , volume scattering R Rnd and surface scattering R Odd Serving size: ， ， ， in, m represents the polarization degree, χ Indicates roundness, g 0 represents the total power of electromagnetic waves.

4. The method for crop extraction from time-series SAR images according to claim 1, wherein: Step 32) further includes: 323) Recalculate the mean of each cluster as the new cluster center; 324) Repeat 321)-333) until the movement amplitude of each cluster center is less than the set threshold to obtain the final time series data classification result.

5. The method for crop extraction from time-series SAR images according to claim 1, wherein: Step 3) also includes: merging other types of land features to obtain the final crop planting distribution result.

Images