A rapid rice identification method based on phenological feature extraction from time-series radar data

Abstract

The present invention discloses a rice rapid identification method based on phenological feature extraction of time series radar data, wherein the time series radar data is constructed based on the GEE platform; the method comprises the following steps: preprocessing the radar data; extracting phenological features of three stages of rice transplanting period, growth period and maturity period; performing X-Means clustering on the extracted phenological features; establishing a rice pixel discrimination standard to determine the rice cluster category. The present invention enhances reliability, improves extraction accuracy, improves processing speed, and has wide applicability.

Description

Technical Field

[0001] The present invention relates to the technical field of crop identification through radar data, and in particular to a method for rapid rice identification based on phenological feature extraction of time-series radar data. Background Art

[0002] With the continuous development of radar sensors and the successive launch of related satellites, the difficulty and cost of obtaining radar image data have continued to decline, and it has been widely used in various fields such as marine oil spill monitoring, forest biomass estimation, crop identification, and geological disaster risk warning. In the field of agricultural remote sensing, cloud pollution in optical images is the main reason for the lack of available image data during the crop growth period, and the characteristics of radar data that penetrate clouds and fog can just make up for this deficiency, and available image data can still be obtained in cloudy and rainy weather. However, the preprocessing of radar data is more complicated and requires certain professional knowledge. In addition, the characteristics of more speckle noise also restrict the application of radar images in crop identification.

[0003] Rice is one of the three major staple crops in the world and has a long history of cultivation in East Asia. It also plays an important role in the diet. Accurately identifying rice and obtaining its planting information are important guarantees for ensuring food security and rationally adjusting the planting structure.

[0004] At present, due to the advantages of optical images such as large amount of information, low threshold for interpretation, and mature application technology, the main data source used in rice extraction is optical remote sensing images. However, optical images are greatly affected by weather factors, especially in the monsoon climate zone where the main rice-growing areas are rainy and hot at the same time. The rice growing period coincides with the precipitation period, which increases the difficulty of optical image acquisition and rice extraction. In order to increase the available image data, avoid missing the optimal window period for identification, and extract the rice planting area more quickly and efficiently, it is necessary to select radar image data that is less affected by weather and more efficient processing algorithms.

[0005] The GEE platform integrates a large amount of remote sensing data, meteorological data and geographic data, and has the characteristics of supporting parallel cloud computing, providing a technical platform for large-scale remote sensing data processing and data mining. The GEE platform can be used to perform radar data preprocessing and rice extraction online, omitting the original data download process and reducing the amount of local processing calculations. Summary of the invention

[0006] To solve the above technical problems, the purpose of the present invention is to provide a method for rapid rice identification based on phenological feature extraction of time series radar data. This method can effectively solve the problem of insufficient available data during the rice growth period, and at the same time, construct time series data of radar images by stacking time phases and extract rice-related phenological features, and combine unsupervised clustering technology to quickly extract rice planting areas, which can effectively improve the efficiency and accuracy of rice identification.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A method for rapid rice identification based on phenological feature extraction of time series radar data, comprising: the time series radar data is constructed based on a GEE platform; the method comprises the following steps:

[0009] Step A preprocesses radar data;

[0010] Step B extracts the phenological characteristics of rice during the three stages of transplanting, growing and maturing;

[0011] Step C performs X-Means clustering on the extracted phenological features;

[0012] Step D establishes rice pixel discrimination criteria to determine the rice cluster category.

[0013] Further, step A comprises:

[0014] A1 obtains the SDWI index by mathematically calculating the backscatter coefficients of the VV and VH polarization channels;

[0015] A2 performs filtering, smoothing and reconstruction on the SDWI time series data.

[0016] Further, step B comprises:

[0017] B1 obtains the irrigation signal characteristics of the rice field during the transplanting period by extracting the maximum value of the SDWI sequence during the transplanting period;

[0018] B2 obtains the vegetation signal characteristics of rice during the growing period by extracting the median of the SDWI sequence during the growing period;

[0019] B3 obtains the vegetation senescence signal characteristics of rice in the mature stage by extracting the median of the SDWI sequence in the mature stage and the difference between the medians of the SDWI in the mature stage and the growth stage.

[0020] Further, step C comprises:

[0021] C1 determines the output range of the number of clusters K;

[0022] C2 performs bisection clustering on each cluster.

[0023] Furthermore, in step D, the classification of rice clusters is based on the SDWI of the phenological characteristic image. flood ,SDWI grow and SDWI harvest Determine the value;

[0024] Step D specifically includes:

[0025] D1 sets the discrimination criteria of typical rice pixels according to the rice phenological characteristics, generates typical rice images that meet the local plant phenology based on this criterion, and establishes random sampling points in the typical rice images;

[0026] D2 uses sampling points to sample the clustering results, extracts the cluster information corresponding to the sampling points, and performs statistics on the cluster situations corresponding to the sampling points to establish a mapping relationship between clusters and rice. The obtained cluster categories represent the cluster categories of rice.

[0027] D3 uses the obtained rice cluster category as a mask value, masks the clustering result, and obtains the rice recognition result.

[0028] Furthermore, the calculation formula of the SDWI index in A1 is:

[0029] SDWI=log(10*VV*VH)-8 (1)

[0030] Where VV represents the VV polarization channel data of Sentinel-1 image, and VH represents the VH polarization channel data of Sentinel-1 image;

[0031] The calculation formula for filtering and smoothing reconstruction of SDWI time series data in A2 is:

[0032]

[0033] Where x(t) is the tth predicted value, k is the degree of the polynomial, and a is the coefficient of the t+1th predicted point in the window.

[0034] Furthermore, the calculation formula for obtaining the irrigation signal characteristics of the rice field in B1 is:

[0035] SDWI flood =max(SDWI f1 ,SDWI f2 ,SDWI f3 ,···,SDWI fn ) (3)

[0036] Where SWDI flood is the maximum value of SDWI during transplanting period, SWDI f1 , SWDI f2 , SWDI f3and SWDI fn They are the first, second, third, and nth observations of the time series SDWI during the transplanting period;

[0037] The calculation formula for extracting the vegetation signal characteristics of rice during the growth period in B2 is:

[0038] SDWI grow =median(SDWI g1 ,SDWI g2 ,SDWI g3 ,···,SDWI gn ) (4)

[0039] Where SDWI grow is the median SDWI value during the growth period, SDWI g1 ,SDWI g2 ,SDWI g3 and SDWI gn They are the first, second, third, and nth observations of the time series SDWI during the growth period;

[0040] The vegetation senescence signal characteristics of rice extracted in the mature stage of B3 are:

[0041] SDWI difference =SDWI harvest -SDWI grow (5)

[0042] SDWI harvest =median(SDWI h1 ,SDWI h2 ,SDWI h3 ,···,SDWI hn ) (6)

[0043] Where SDWI difference SDWI is the difference between the median SDWI of the mature stage and the median SDWI of the growing stage. harvest is the median SDWI value at maturity, SDWI grow is the median SDWI value during the growth period, SDWI h1 ,SDWI h2 ,SDWI h3 and SDWI hn They are the first, second, third and nth observation values ​​of the time series SDWI in the maturity period.

[0044] Furthermore, in C1, the minimum value k of the given output min As the initial k value, the maximum value k max As a convergence condition;

[0045] The C2 includes: calculating the BIC score of each cluster into one category and two categories, and deciding whether to perform binary clustering according to the BIC score, wherein the BIC calculation formula is:

[0046]

[0047] Where m is the number of samples, p is the number of model parameters, is the maximum likelihood function value.

[0048] Furthermore, if k < k max , then proceed to step C2, otherwise return the result.

[0049] Furthermore, in D1, the discrimination standard of typical rice pixels is SDWI flood >0∩SDWI grow <0∩SDWI harvest <0;

[0050] In D2, the rice cluster is Cluster paddy , and its calculation formula is:

[0051] Cluster paddy =f class (Point ClassicPaddy ) (8)

[0052] Where: Point ClassicPaddy is the sampling point of a typical rice image, f class () is to obtain the cluster category of the sampling points;

[0053] In D3, the calculation formula for finally identifying rice information is:

[0054] Paddy=Mask(Cluster Paddy ) (9)

[0055] Where: Paddy is the final identified rice information, Cluster Paddy represents the cluster of rice, and Mask() masks the image according to the rice cluster.

[0056] Compared with the prior art, one or more embodiments of the present invention may have the following advantages:

[0057] (1) Enhanced reliability: Optical images are greatly affected by weather factors. Weather conditions such as rain and fog can cause serious cloud pollution in optical images, making it difficult to obtain usable data. Radar data is available all day and all weather, ensuring data reliability. The ability of radar data to penetrate clouds and fog ensures that images can be obtained during the critical period of rice growth, as well as the temporal continuity of images.

[0058] (2) Improving extraction accuracy: The phenological characteristics of crops are the main basis for identifying the crops using remote sensing technology, and images of key phenological stages are crucial in the identification process. Optical satellites are often affected by weather and cannot obtain usable images within a specific time. Radar satellites can obtain image data of the entire growth period including key periods without being affected by weather. At the same time, the backscatter variation characteristics of radar images in different growth phenological stages of rice are significantly different, which has significant advantages for rice crop extraction.

[0059] (3) Improved processing speed: The GEE platform integrates the corresponding Sentinel-1 dataset and performs image preprocessing, eliminating the need to download images to the local computer for processing. The entire rice identification process can be completed on the GEE platform, greatly improving processing speed.

[0060] (4) Wide applicability: Rice extraction is mainly based on the phenological characteristics of rice at different stages of extraction. The characteristics are relatively robust and have wide applicability. At the same time, the recognition process is highly automated and can be completed without human intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a method for extracting rice features and identifying rice based on time series radar data;

[0062] Figure 2 This is a schematic diagram of typical rice pixel extraction;

[0063] Figure 3 This is the rice recognition result diagram. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with embodiments and drawings.

[0065] like Figure 1 As shown, a method for rapid rice identification based on phenological feature extraction of time series radar data is provided, wherein the time series radar data is constructed based on the GEE platform; the method comprises the following steps:

[0066] Step 10 pre-processes the radar data;

[0067] Step 20 extracts the phenological characteristics of the three stages of rice transplanting, growth and maturity;

[0068] Step 30: performing X-Means clustering on the extracted phenological features;

[0069] Step 40 is to establish a rice pixel discrimination standard to determine the rice cluster category.

[0070] In step 10 above:

[0071] Ground Range Detected (GRD) is a type of image product of Sentinel-1. It is generally obtained by multi-view processing and ground range conversion from single-view complex products. The Sentinel-1GRD product integrated by GEE adds thermal noise removal, radiation calibration and geocoding processing on this basis, and can be directly used for subsequent analysis. The preprocessing in this step is mainly SDWI (Sentinel-1 Dual-Polarized Water Index, SDWI) index calculation and Savitzky-Golay filter smoothing. SDWI fully combines the information of VV and VH polarization channels, and can highlight the differences between water bodies and other landforms and the difference in water content of different vegetation better than using single channel data. Savitzky-Golay filtering can effectively remove unnecessary noise while retaining the shape and characteristics of the original data.

[0072] 1) SDWI index calculation: SDWI is obtained by mathematical operation of the backscatter coefficients of the VV and VH polarization channels. The calculation formula is as follows:

[0073] SDWI=log(10*VV*VH)-8 (1)

[0074] Where VV represents the VV polarization channel data of the Sentinel-1 image, and VH represents the VH polarization channel data of the Sentinel-1 image.

[0075] 2) Savitzky-Golay filtering and smoothing: Due to the characteristics of radar imaging, its images often have many noise spots. In order to improve the data quality, it is necessary to filter and smooth the SDWI time series data and reconstruct it. The calculation formula is as follows:

[0076]

[0077] Where x(t) is the tth predicted value, k is the degree of the polynomial, and a is the coefficient of the t+1th predicted point in the window.

[0078] In step 20 above:

[0079] According to the phenological differences of rice, the growth period is divided into three main phenological stages: transplanting period, growth period and maturity period, and phenological characteristics are extracted for different phenological stages.

[0080] 1) Extract the irrigation signal characteristics of the rice field during the transplanting period. After the rice field is irrigated, it presents similar water body characteristics, and the SDWI index increases. The transplanting characteristics are obtained by extracting the maximum value of the SDWI sequence during the transplanting time. The calculation formula is as follows:

[0081] SDWI flood =max(SDWI f1 ,SDWI f2 ,SDWI f3 ,···,SDWI fn ) (3)

[0082] Where SDWI flood is the maximum SDWI value during the transplanting period, SDWI f1 ,SDWI f2 ,SDWI f3 and SDWI fn They are the 1st, 2nd, 3rd and nth observation values ​​of the time series SDWI during the transplanting period.

[0083] 2) Extract the vegetation signal characteristics of rice during the growth period. During the growth period, the rice plants gradually develop and form a complete canopy. At this time, the SDWI decreases. The growth period characteristics are obtained by extracting the median of the SDWI during the growth period. The calculation formula is as follows:

[0084] SDWI grow =median(SDWI g1 ,SDWI g2 ,SDWI g3 ,···,SDWI gn ) (4)

[0085] Where SDWI grow is the median SDWI value during the growth period, SDWI g1 ,SDWI g2 ,SDWI g3 and SDWI gn They are the first, second, third and nth observation values ​​of the time series SDWI during the growth period.

[0086] 3) Extract the vegetation senescence signal characteristics of rice at maturity, which is obtained by extracting the median SDWI at maturity and the difference between the median SDWI at maturity and the median SDWI at growth stage. The calculation formula is as follows:

[0087] SDWI difference =SDWI harvest -SDWI grow (5)

[0088] SDWI harvest =median(SDWI h1 ,SDWI h2 ,SDWI h3 ,···,SDWI hn )(6)

[0089] Where SDWI difference SDWI is the difference between the median SDWI of the mature stage and the median SDWI of the growing stage. harvest is the median SDWI value at maturity, SDWI grow is the median SDWI value during the growth period, SDWI h1 ,SDWI h2 ,SDWI h3 and SDWI hn They are the first, second, third and nth observation values ​​of the time series SDWI in the maturity period.

[0090] In the above step 30:

[0091] X-Means clustering is an unsupervised classification algorithm improved from K-Means clustering. It can mine the intrinsic characteristics of data and divide data into several categories (categories after clustering are called clusters, and the number of clusters is k) through the intrinsic relationship of data without inputting samples, so that the difference within the same cluster is minimized and the difference between different clusters is maximized. Compared with other clustering algorithms, X-Means clustering is not easy to converge locally and automatically determine the optimal k value. The main steps of X-Means clustering of extracted phenological characteristics are:

[0092] 1) Determine the value output range: k value is the minimum value k of the given output min As the initial k value, the maximum value k max as a convergence condition.

[0093] 2) Perform binary clustering on each cluster: Calculate the BIC Score (Bayesian Information Criterion) of each cluster divided into category 1 and category 2, and decide whether to perform binary clustering based on the BIC Score. The BIC calculation formula is as follows:

[0094]

[0095] Where m is the number of samples, p is the number of model parameters, is the maximum likelihood function value.

[0096] 3) If k < k max , then proceed to 2), otherwise return the result.

[0097] In the above step 40:

[0098] The different clusters obtained by clustering do not have semantic information and cannot identify the cluster representing rice. Therefore, it is necessary to further identify the cluster representing rice. The rice clusters are mainly classified by statistically analyzing the phenological characteristics of the clusters. flood ,SDWI growand SDWI harvest The main steps of rice identification are as follows (e.g. Figure 2 shown):

[0099] 1) Typical rice image generation: According to the rice phenological characteristics, the discrimination criteria of typical rice pixels are set, specifically SDWI flood >0∩SDWI grow <0∩SDWI harvest <0, based on this criterion, a typical rice image that meets the local plant phenology is generated, and random sampling points are established in the typical rice image.

[0100] 2) Cluster sampling and statistics: Use sampling points to sample the clustering results, extract the cluster information corresponding to the sampling points, and count the cluster situations corresponding to the sampling points to establish a mapping relationship between clusters and rice (the obtained cluster categories represent the cluster categories of rice);

[0101] Rice cluster is Cluster paddy , and its calculation formula is:

[0102] Cluster paddy =f class (Point ClassicPaddy ) (8)

[0103] Where: Point ClassicPaddy is the sampling point of a typical rice image, f class () is to obtain the cluster category of the sampling points;

[0104] 3) Rice cluster mask: The obtained rice cluster category is used as the mask value, the clustering result is masked, and the rice recognition result is obtained (such as Figure 3 shown);

[0105] In D3, the calculation formula for finally identifying rice information is:

[0106] Paddy=Mask(Cluster Paddy ) (9)

[0107] Where: Paddy is the final identified rice information, Cluster Paddy represents the cluster of rice, and Mask() masks the image according to the rice cluster.

[0108] Although the embodiments disclosed in the present invention are as above, the contents described are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for rapid rice identification based on phenological feature extraction of time series radar data, characterized in that: The time series radar data is constructed based on the GEE platform; the method comprises the following steps: Step A preprocesses radar data; Step B extracts the phenological characteristics of rice during the three stages of transplanting, growing and maturing; Step C performs X-Means clustering on the extracted phenological features; Step D: establishing rice pixel discrimination criteria to determine the rice cluster category; In step D, the classification of rice clusters is based on the SDWI of the phenological characteristic image. flood ,SDWI grow and SDWI harvest The value is used for judgment; among them, SWDI flood is the maximum SDWI value during the transplanting period, SDWI grow is the median SDWI during the growth period, SWDI harvest is the median SDWI value at maturity; Step D specifically includes: D1 Generation of typical rice images: According to the rice phenological characteristics, the discrimination criteria of typical rice pixels are set, and based on this criteria, typical rice images that meet the local plant phenology are generated, and random sampling points are established in the typical rice images; D2 uses sampling points to sample the clustering results, extracts the cluster information corresponding to the sampling points, and performs statistics on the cluster situations corresponding to the sampling points to establish a mapping relationship between clusters and rice. The obtained cluster categories represent the cluster categories of rice. D3 uses the obtained rice cluster category as a mask value, masks the clustering result, and obtains the rice recognition result.

2. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 1 is characterized in that: The step A comprises: A1 obtains the SDWI index by mathematically calculating the backscatter coefficients of the VV and VH polarization channels; A2 performs filtering, smoothing and reconstruction on the SDWI time series data.

3. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 1 is characterized in that: The step B comprises: B1 obtains the irrigation signal characteristics of the rice field during the transplanting period by extracting the maximum value of the SDWI sequence during the transplanting period; B2 obtains the vegetation signal characteristics of rice during the growing period by extracting the median of the SDWI sequence during the growing period; B3 obtains the vegetation senescence signal characteristics of rice in the mature stage by extracting the median of the SDWI sequence in the mature stage and the difference between the medians of the SDWI in the mature stage and the growth stage.

4. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 1 is characterized in that: The step C comprises: C1 determines the output range of the number of clusters K; C2 performs bisection clustering on each cluster.

5. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 2 is characterized in that: The calculation formula of the SDWI indicator in A1 is: SDWI=log(10*VV*VH)-8 (1) Where VV represents the VV polarization channel data of Sentinel-1 image, and VH represents the VH polarization channel data of Sentinel-1 image; The calculation formula for filtering and smoothing reconstruction of SDWI time series data in A2 is: Where x(t) is the t-th predicted value, k is the degree of the polynomial, and a k is the coefficient of the t+1th prediction point in the window.

6. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 3 is characterized in that: The calculation formula for obtaining the irrigation signal characteristics of the rice field in B1 is: <h2 style=";text-align:left;direction:ltr">SDWI<h2 style=";text-align:left;direction:ltr"> flood <h2 style=";text-align:left;direction:ltr"> =max(SDWI<h2 style=";text-align:left;direction:ltr"> f1 <h2 style=";text-align:left;direction:ltr"> SDWI<h2 style=";text-align:left;direction:ltr"> f2 <h2 style=";text-align:left;direction:ltr"> SDWI<h2 style=";text-align:left;direction:ltr"> f3 <h2 style=";text-align:left;direction:ltr"> SDWI<h2 style=";text-align:left;direction:ltr"> fn <h2 style=";text-align:left;direction:ltr"> ) (3) Where SWDI flood is the maximum value of SDWI during transplanting period, SWDI f1 , SWDI f2 , SWDI f3 and SWDI fn They are the first, second, third, and nth observations of the time series SDWI during the transplanting period; The calculation formula for extracting the vegetation signal characteristics of rice during the growth period in B2 is: SDWI grow =median(SDWI g1 , SDWI g2 , SDWI g3 ,···, SDWI gn ) (4) Where SDWI grow is the median SDWI value during the growth period, SDWI g1 ,SDWI g2 ,SDWI g3 and SDWI gn They are the first, second, third, and nth observations of the time series SDWI during the growth period; The vegetation senescence signal characteristics of rice extracted in the mature stage of B3 are: SDWI difference = SDWI harvest - SDWI grow (5) SDWI harvest =median(SDWI h1 , SDWI h2 , SDWI h3 ,···, SDWI hn ) (6) Where SWDI difference SWDI is the difference between the median SDWI of the mature stage and the median SDWI of the growing stage. harvest is the median value of SDWI at maturity, SWDI grow is the median SDWI during the growth period, SWDI h1 , SWDI h2 , SWDI h3 and SWDI hn They are the first, second, third and nth observation values ​​of the time series SDWI in the maturity period.

7. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 4 is characterized in that: The C1, with a given output minimum value k min As the initial k value, the maximum value k max As a convergence condition; The C2 includes: calculating the BIC score of each cluster into one category and two categories, and deciding whether to perform binary clustering according to the BIC score, wherein the BIC calculation formula is: Where m is the number of samples, p is the number of model parameters, is the maximum likelihood function value.

8. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 7, characterized in that: If k < k max , then proceed to step C2, otherwise return the result.

9. The method for rapid rice identification based on phenological feature extraction of time series radar data according to claim 1, characterized in that: In D1, the criterion for distinguishing typical rice pixels is SDWI flood >0∩SDWI grow <0∩SDWI harvest <0; In D2, the rice cluster is Cluster paddy , and its calculation formula is: Cluster paddy =f class (Point ClassicPaddy ) (8) Where: Point ClassicPaddy is the sampling point of a typical rice image, f class () is to obtain the cluster category of the sampling points; In D3, the calculation formula for finally identifying rice information is: Paddy=Mask(Cluster Paddy ) (9) Where: Paddy is the final identified rice information, Cluster Paddy represents the cluster of rice, and Mask() masks the image according to the rice cluster.

Images