Remote sensing identification method, device, medium and system for rice planting area distribution

Abstract

The invention provides a remote sensing recognition method, device, medium and system for rice planting area distribution, and belongs to the field of agricultural remote sensing, the remote sensing recognition method adopts a hierarchical classification algorithm for extraction, does not depend on actual measurement sample points, and obtains high-precision agricultural irrigation area rice spatial distribution rasterized data after automatic training; specifically, an agricultural area and a non-agricultural area are classified through a supervised learning algorithm classification regression tree (CART), a support vector machine (SVM) and a random forest (RF), and then a decision tree algorithm is adopted to further classify and discriminate a rice planting area based on obtained agricultural area classification raster data. The method does not depend on actual measurement data, classification time and economic cost are greatly reduced, and negative effects of the quality of a sample training set of a traditional machine algorithm on a classification result are effectively avoided; meanwhile, on the premise of ensuring high classification efficiency, the method meets the rice drawing precision required by agricultural remote sensing application, and is beneficial to popularization and application.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of agricultural remote sensing, and particularly relates to a remote sensing identification method, device, medium and system for distribution of rice planting areas. BACKGROUND

[0002] With the rapid development of science and technology in recent years, scientific and modern agricultural management has gradually become the focus of attention, and reliable crop planting distribution information is a prerequisite for fine and modern management of irrigation areas and a key basis for saving and intensive use of regional water and soil resources. Therefore, accurately and efficiently extracting crop planting distribution information has important research significance and value.

[0003] At present, there are various methods for obtaining crop planting distribution and structure information. Among them, the method based on sampling investigation and statistics is relatively traditional, but it is time-consuming and laborious, and has the problems of poor timeliness and reliability. In recent years, remote sensing means for extracting crop planting distribution information has gradually become the mainstream. However, the reliability of existing data products is subject to sample point distribution, and there may be certain errors in areas without sample points (such as agricultural irrigation areas), so it cannot meet the needs of irrigation area scale research and management. Therefore, high-precision crop planting distribution information in irrigation areas is often not obtained through existing regional scale data products, but needs to be extracted through remote sensing inversion means.

[0004] In the aspect of extracting crop planting distribution information based on remote sensing technology, using a single image and based on its spectral and spatial characteristics for extraction is a relatively common means at present. This method only needs to use a single period or a few periods of remote sensing images, and has the advantages of high efficiency and strong operability under the premise that the image features are sensitive to the classifier and the key identification period image is reliable. However, this method is easily limited by the available images and their quality, and once the image of the key identification period is disturbed by clouds and rain, the reliability of the crop planting distribution information extraction result may decrease significantly. Therefore, this shortcoming significantly restricts the application of this means in cloudy and rainy areas (such as southern rice irrigation areas).

[0005] In practical applications, existing SVM, random forest, decision tree and other classifiers all have their limitations and shortcomings, so a new remote sensing identification method needs to be proposed to meet the identification needs of rice agricultural irrigation areas and improve the efficiency and accuracy of classification. SUMMARY

[0006] In view of the technical problems in the background art, the present application provides a remote sensing identification method, device, medium and system for distribution of rice planting areas, which aims to solve the problem that the existing remote sensing identification method for distribution of rice planting areas cannot efficiently and accurately identify different ground objects and different climate conditions.

[0007] In a first aspect, the embodiments of the present application provide a remote sensing identification method for distribution of rice planting areas, comprising the following steps:

[0008] S1, constructing a classification feature set about the original satellite image based on satellite remote sensing images and SRTM DEM data, and constructing a ground sample set based on Planet images, GlobeLand 30 data and field investigation data. The classification feature set includes texture feature data, classification identification bands and terrain feature data about the original satellite image.

[0009] S2, based on the ground sample set, using classification regression tree CART, support vector machine SVM and random forest RF supervised learning algorithm to train and learn the classification feature set and to convert data, to obtain farmland area raster data and non-farmland area raster data.

[0010] S3, using a decision tree algorithm to identify the rice planting area under the conditions of no cloud and cloud for the farmland area raster data, and integrating to obtain rice planting area spatial distribution data.

[0011] S4, integrating the non-farmland area raster data and the rice planting area spatial distribution data to obtain a classification map with agricultural planting area and / or rice planting area identification.

[0012] In some embodiments, the S1 step specifically comprises the following steps:

[0013] S11, using satellite remote sensing images to obtain original satellite images in different periods and different weather conditions, selecting original satellite images in the double-cropping rice growing period, and obtaining clear satellite images after preprocessing.

[0014] S12, selecting different bands that can be distinguished in the clear satellite images as classification identification bands, using a step probability statistical method to construct a gray tone space correlation matrix about the clear satellite images, and based on a 3x3 window and 64 gray scale quantization levels to calculate the angular two matrix, contrast, correlation and variance of the classification identification bands, to obtain texture feature data.

[0015] S13, using the GEE remote sensing big data platform to obtain SRTM DEM data with a spatial resolution of 30m, extracting elevation based on the SRTM DEM data and calculating to obtain slope information, i.e. terrain feature data.

[0016] In some embodiments, in the S11 step, the GEE remote sensing big data cloud platform is used to resample each band of a plurality of original satellite images in different periods and different weather conditions to 10m spatial resolution, to select a median value of the original satellite images of Sentinel-2 in the double-cropping rice growing period to synthesize a clear satellite image, and to obtain texture feature data based on the clear satellite image.

[0017] In some embodiments, the classification identification waveband in the S12 step includes a visible light waveband in which ground objects can be distinguished, and infrared, near-infrared and short-wave infrared wavebands in which plant growth conditions can be distinguished.

[0018] In some embodiments, the S2 step specifically includes the following steps: based on the ground sample set, using a classification regression tree CART, a support vector machine SVM and a random forest RF supervised learning algorithm to train and learn the classification feature set, to obtain land use classification data about the farmland area and the non-farmland area, to convert the land use classification data into raster data and reassign values, to obtain farmland area raster data and non-farmland area raster data; wherein the land use classification data of the non-farmland area includes water bodies, grasslands, building sites and unused land.

[0019] In some embodiments, the S3 step specifically includes the following steps:

[0020] S31, converting the farmland area raster data into vector data to obtain shp vector data containing only the farmland area.

[0021] S32, under cloudless conditions, calculating farmland distribution data in a cloudless area based on the shp vector data containing only the farmland area.

[0022] S33, under cloudy conditions, obtaining farmland distribution data in a cloudy area based on the shp vector data containing only the farmland area, based on the rice phenology index and the normalized vegetation index data set.

[0023] S34, integrating the farmland distribution data in the cloudless area and the farmland distribution data in the cloudy area to obtain rice planting area spatial distribution data.

[0024] Further, in more specific embodiments, the S32 step specifically includes:

[0025] S321, under cloudless conditions, cropping the original satellite images under multiple periods of clear sky cloudless conditions during the rice growing period based on the shp vector data containing only the farmland area to obtain original satellite images containing only the farmland area, and calculating a land water index LSWI data set and a dual-band enhanced vegetation index EVI2 data set.

[0026] S322, determining the rice heading date based on the maximum value in the dual-band enhanced vegetation index EVI2 data set and determining the tillering date accordingly, calculating the EVI2 amplitude C v of the tillering date, the minimum value LSWI min of the LSWI, the maximum value LSWI max of the LSWI, and the LSWI amplitude C w , and then calculating the ratio RCLE of the LSWI amplitude and the EVI2 amplitude.

[0027] S323, based on LSWI min The judgment rule of θ1 and RCLE < θ2 identifies the rice distribution and obtains the rice distribution data in the cloudless area, wherein the values of θ1 and θ2 are obtained by trial and error based on the ground sample set.

[0028] In this embodiment, the calculation formulae of the land water index LSWI, the dual-band enhanced vegetation index EVI2 and the ratio of the LSWI amplitude to the EVI2 amplitude RCLE are as follows:

[0029]

[0030] Further, in a more specific embodiment, the step S33 specifically comprises:

[0031] S331, under the cloud condition, the original satellite image under the multi-period clear sky and cloudless condition in the rice growth period is cropped based on the shp vector data containing only the farmland area to obtain the original satellite image containing only the farmland area, a normalized vegetation index NDVI data set is calculated, and the Savitzky-Golay filter is used to smooth the normalized vegetation index NDVI data set to obtain an NDVI data smooth curve.

[0032] S332, based on the NDVI data smooth curve, each pixel in the farmland area is judged and the phenological index is extracted to obtain a phenological index grid data, and the phenological index grid data contains three phenological indexes of the growth start period SOS, the growth end period EOS and the growth period length LOS.

[0033] S333, based on the judgment rule of SOS ∈ [α1, α2], SOS ∈ [β1, β2] and LOS > γ, the rice is identified, and the rice distribution data in the cloud area is obtained, wherein the values of α1, α2, β1, β2 and γ are obtained by trial and error based on the ground sample set.

[0034] In this embodiment, the calculation formula of the normalized vegetation index NDVI is as follows:

[0035]

[0036] In formulae (1) to (4), ρ NIR represents the reflectivity of the near-infrared band, ρ Red represents the reflectivity of the red light band.

[0037] In this embodiment, the specific steps of judging and extracting the phenological index of each pixel in the farmland area are as follows: taking the normalized vegetation index NDVI threshold NDVI th as the judgment condition, the first time higher and lower than NDVI thThe dates of SOS and EOS are determined respectively, and LOS is calculated; wherein the length of growth period LOS and the normalized vegetation index NDVI threshold NDVI th The calculation formula of LOS is as follows:

[0038] LOS = EOS - SOS (5)

[0039] NDVI th = (NDVI max - NDVI min ) x r + NDVI min (6)

[0040] In formula (6), NDVI th is the NDVI threshold; NDVI max is the maximum value of NDVI; NDVI min is the minimum value of NDVI; and r is the calculation parameter of NDVI th .

[0041] Further preferably, in the step S323, the set value of θ1 is 0.10, and the set value of θ2 is 0.42.

[0042] Further preferably, in the step S333, the set values of α1, α2, β1, β2 and γ are 110, 140, 180, 210 and 70 respectively under the condition of early rice, and the set values of α1, α2, β1, β2 and γ are 195, 225, 275, 305 and 70 respectively under the condition of late rice.

[0043] In some embodiments, the step S4 specifically comprises the following steps: integrating the non-agricultural land area raster data and the spatial distribution data of rice planting area, and adopting different numerical values and / or colors to identify different classifications in the agricultural planting area and the rice planting area respectively, to obtain a classification map with identification of the agricultural planting area and / or the rice planting area.

[0044] In a second aspect, the embodiments of the present application provide a device for remote sensing identification of rice planting area distribution, which is used to implement the method for remote sensing identification of rice planting area distribution as described in the first aspect. The device comprises: a data acquisition module, which is used to acquire original satellite images and input SRTM DEM data, Planet images, GlobeLand 30 and field investigation data to form a classification feature set and a ground sample set; a data training module, which is used to receive the classification feature set and the ground sample set from the data acquisition module and perform training learning and data conversion to obtain farmland area raster data and non-farmland area raster data; a decision operation module, which is used to receive the farmland area raster data from the data training module, and perform rice planting area identification under cloud-free conditions and cloud conditions by using a decision tree algorithm on the farmland area raster data, and obtain rice planting area spatial distribution data after integration; and an image output module, which is used to receive the non-farmland area raster data from the data training module and the rice planting area spatial distribution data from the decision operation module, and output a classification map with agricultural planting area and / or rice planting area identification after integration.

[0045] In a third aspect, the embodiments of the present application provide a computer readable storage medium, which stores a computer program. When the program is run by the device in the second aspect, it is used to implement the method for remote sensing identification of rice planting area distribution as described in the first aspect.

[0046] In a fourth aspect, the embodiments of the present application provide a system for remote sensing identification of rice planting area distribution, which is used to implement the method for remote sensing identification of rice planting area distribution as described in the first aspect.

[0047] Compared with the prior art, the beneficial effects of the present application include:

[0048] 1. The present application not only uses machine learning algorithm to realize the identification of farmland area and non-farmland area by using satellite remote sensing image spectrum and texture features and slope information, but also establishes classification rules by using LSWI, EVI2, NDVI and phenological features under cloud and cloud-free conditions, and realizes further accurate identification of rice planting area in farmland area based on decision tree algorithm, so as to realize high-precision land use and remote sensing identification of rice planting area in agricultural irrigation area.

[0049] 2. The present application solves the problems of limited number of field samples, high quality requirement of measured samples and difficulty in obtaining samples, and ensures high precision and stability of the identification of rice planting area without measured samples.

[0050] 3. The application solves the problem of limited satellite image quality and quantity under cloudy conditions, which leads to difficulties in identifying rice planting areas. The application effectively integrates the distribution data of rice planting areas under cloudy and non-cloudy conditions, achieving efficient identification of rice planting areas under complex climate conditions.

[0051] 4. Compared with traditional algorithms, the method of the application significantly reduces time and money costs, has strong operability, and can be used to realize high-precision land use identification with ordinary hardware devices when applied to large-area high-precision remote sensing images, making it possible for existing devices to automatically acquire and identify the distribution state of rice planting areas in large agricultural areas, which is conducive to popularization and application. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the application, the drawings used in the application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0053] Figure 1 A flowchart of an embodiment of the remote sensing identification method of the distribution of rice planting areas provided by the application;

[0054] Figure 2 A structural diagram of an embodiment of the remote sensing identification device of the distribution of rice planting areas provided by the application;

[0055] Figure 3 A flowchart of the remote sensing identification method of the distribution of rice planting areas in Example 1 of the application;

[0056] Figure 4 An original satellite remote sensing image in Example 1 of the application;

[0057] Figure 5 A schematic diagram of the texture value calculation process of the original satellite remote sensing image in Example 1 of the application;

[0058] Figure 6 A schematic diagram of the slope calculation result in Example 1 of the application;

[0059] Figure 7 A schematic diagram of the calculation process of the supervised learning algorithm classification in Example 1 of the application;

[0060] Figure 8 A schematic diagram of the relationship between EVI2 and LWSI in a specific phenological phase selected in Example 1 of the application;

[0061] Figure 9 A schematic diagram of the extracted phenological parameters in Example 1 of the application;

[0062] Figure 10 Figure 1 is a land use and rice planting area classification map finally obtained in Example 1 of the present application;

[0063] Figure 11 Figure 1 is a land use and rice planting area classification map finally obtained in Example 1 of the present application;

[0064] Figure 12 Figure 1 is a land use and rice planting area classification map finally obtained in Example 1 of the present application;

[0065] In the figure: 101-data acquisition module, 102-data training module, 103-decision operation module, 104-image output module. DETAILED DESCRIPTION

[0066] The embodiments of the technical solutions of the present application will be described in detail below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, and cannot limit the protection scope of the present application.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "include" and "have" and any variations thereof in the specification and claims of the present application and the above description of drawings are intended to cover non-exclusive inclusion.

[0068] In this paper, the "embodiment" means that the specific features, structures or characteristics described in conjunction with the embodiment can be included in at least one embodiment of the present application. The phrase appears in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment to other embodiments. The skilled person in the art explicitly and implicitly understands that the embodiments described herein can be combined with other embodiments.

[0069] In the description of the embodiments of the present application, the term "and / or" is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents a "or" relationship between the front and rear associated objects.

[0070] In the description of the embodiments of the present application, the term "several" refers to more than two (including two), and "several" refers to more than two (including two).

[0071] In the prior art, a single image is often used and classification is performed based on spectral and spatial features. This approach is limited by the available image and its quality. If the image during the key identification period is disturbed by clouds and rain, the reliability of the crop planting distribution information extraction result can be significantly reduced, making it difficult to make accurate and stable identification under complex environmental conditions, especially in cloudy and rainy areas (such as southern rice irrigation areas).

[0072] To solve the above problems, the present application provides a remote sensing identification method, device, medium and system for rice planting area distribution. The method uses satellite remote sensing image spectral and texture features, and slope information to identify farmland and non-farmland areas using machine learning algorithms. Based on this, for cloudy and cloudless conditions, LSWI, EVI2, NDVI and phenological features are used to establish classification rules, and a decision tree algorithm is used to further identify farmland rice planting areas, ultimately achieving high-precision land use and rice planting area remote sensing identification in agricultural irrigation areas.

[0073] As shown in Figure 1 the first aspect, the present application provides a remote sensing identification method for rice planting area distribution, comprising the following steps:

[0074] S1, based on satellite remote sensing images and SRTM DEM data, a classification feature set for the original satellite image is constructed, and a ground sample set is constructed based on Planet images, GlobeLand 30 data and field investigation data. The classification feature set includes texture feature data, classification identification bands and terrain feature data for the original satellite image. In this step, the classification feature set contains three different feature data: texture features, spectral features and terrain features. Compared with the limitation of single classification feature parameter in the prior art, more abundant data samples and classification dimensions can be obtained, thereby improving the stability of the subsequent classification results.

[0075] In some embodiments, the S1 step specifically comprises the following steps:

[0076] S11, satellite remote sensing images are used to obtain original satellite images at different times and under different weather conditions, and original satellite images during the growth period of double-crop rice are selected to obtain clear satellite images after preprocessing. In this step, the GEE remote sensing big data cloud platform can be used to resample each band of several original satellite images at different times and under different weather conditions to 10m spatial resolution, and the median value of the original satellite images of Sentinel-2 during the growth period of double-crop rice is selected to synthesize clear satellite images. In other embodiments, other growth stages of original satellite images can also be selected to synthesize clear satellite images according to actual conditions, which are not limited herein.

[0077] S12, select distinguishable different bands in the clear satellite image as classification and recognition bands, construct a gray tone space correlation matrix about the clear satellite image using a statistical method, and calculate an angular second moment, a contrast, a correlation, and a variance of the classification and recognition bands based on a 3x3 window and 64 gray scale quantization levels to obtain texture feature data. In this step, the classification and recognition bands are used as spectral feature data, including visible light bands of distinguishable ground objects and infrared bands, near-infrared bands, and short-wave infrared bands of distinguishable plant growth conditions, and other distinguishable different bands can be set as classification and recognition bands according to actual conditions, which are not limited herein.

[0078] S13, obtain SRTM DEM data with a spatial resolution of 30m using the GEE remote sensing big data platform, extract elevation based on the SRTM DEM data, and calculate slope information to obtain terrain feature data.

[0079] S2, based on the ground sample set, train and learn the classification feature set using a classification and regression tree CART, a support vector machine SVM, and a random forest RF supervised learning algorithm, and convert data to obtain farmland area raster data and non-farmland area raster data.

[0080] In some embodiments, the S2 step specifically includes the following steps: based on the ground sample set, train and learn the classification feature set using a classification and regression tree CART, a support vector machine SVM, and a random forest RF supervised learning algorithm to obtain land use classification data about farmland areas and non-farmland areas, convert the land use classification data to raster data and reassign values to obtain farmland area raster data and non-farmland area raster data; specifically, in the actual training and learning process, the three methods of classification and regression tree CART, support vector machine SVM, and random forest RF supervised learning algorithm can be used in sequence to train and learn the test area, and the ground sample set of the test area is combined to filter out the training method that best matches the test area in terms of accuracy, and the land use classification data about farmland areas and non-farmland areas is obtained by training using this training method. Since the recognition accuracy of the same method for different test areas differs, and the recognition accuracy of different methods for the same test area also differs, the application uses the above-mentioned training and filtering method to obtain a method that best matches the test area in terms of test accuracy, thereby significantly improving the subsequent recognition accuracy. The land use classification data of the non-farmland area includes water bodies, grasslands, building sites, and unused land; the reassignment method can assign a value of 1 to the farmland area and a value of NoData to the non-farmland area, or other assignment methods can be selected according to actual needs, which are not limited herein.

[0081] S3, using a decision tree algorithm to identify the rice planting area from the farmland area raster data under cloudless conditions and cloudy conditions, and integrating the data to obtain the spatial distribution data of the rice planting area.

[0082] In some embodiments, the S3 step specifically comprises the following steps:

[0083] S31, converting the farmland area raster data into vector data to obtain shp vector data containing only the farmland area.

[0084] S32, under cloudless conditions, calculating the rice distribution data in the cloudless area based on the shp vector data containing only the farmland area. Further, the S32 step specifically comprises:

[0085] S321, under cloudless conditions, cutting the original satellite images under the multi-period sunny and cloudless conditions during the rice growing period based on the shp vector data containing only the farmland area to obtain original satellite images containing only the farmland area, and calculating the land water index LSWI dataset and the double-band enhanced vegetation index EVI2 dataset. The calculation formulas of the land water index LSWI and the double-band enhanced vegetation index EVI2 are as follows:

[0086]

[0087] In formulas (1) and (2), ρ NIR represents the reflectivity of the near-infrared band, and ρ Red represents the reflectivity of the red light band.

[0088] S322, determining the rice heading date based on the maximum value in the double-band enhanced vegetation index EVI2 dataset and determining the tillering date accordingly, calculating the EVI2 amplitude C v of the tillering date, the minimum value LSWI min of the land water index LSWI, the maximum value LSWI max of the land water index LSWI, and the LSWI amplitude C w , and then calculating the ratio RCLE of the LSWI amplitude and the EVI2 amplitude. The calculation formula of the ratio RCLE of the LSWI amplitude and the EVI2 amplitude is as follows:

[0089] RCLE = C W / C V (3)

[0090] S323, determining the rice planting area based on the LSWI minThe judgment rule of θ1 and RCLE < θ2 identifies the rice distribution and obtains the rice distribution data in the cloud-free area, wherein the values of θ1 and θ2 are obtained by trial and error based on the ground sample set. In this embodiment, the set value of θ1 is preferably 0.10, and the set value of θ2 is preferably 0.42; in other embodiments, the above parameters can be adaptively adjusted according to the actual situation, which is not limited herein.

[0091] S33, under the condition of cloud, based on the shp vector data containing only the farmland area, the rice distribution data in the cloud area is obtained based on the rice phenology index and the normalized vegetation index data set. Further, the step S33 specifically comprises:

[0092] S331, under the condition of cloud, the original satellite image under the condition of clear sky and cloud-free in the rice growth period is cropped based on the shp vector data containing only the farmland area to obtain the original satellite image containing only the farmland area, and the normalized vegetation index NDVI data set is calculated, and the Savitzky-Golay filter is used to smooth the normalized vegetation index NDVI data set to obtain the NDVI data smooth curve. The calculation formula of the normalized vegetation index NDVI is as follows:

[0093]

[0094] S332, based on the NDVI data smooth curve, each pixel in the farmland area is judged and the phenology index is extracted to obtain the phenology index grid data, and the phenology index grid data contains three phenology indexes of the growth start period SOS, the growth end period EOS and the growth period length LOS. In this step, the specific steps of judging and extracting the phenology index of each pixel in the farmland area are as follows: taking the normalized vegetation index NDVI threshold NDVI th as the judgment condition, the dates of first higher and lower than NDVI th are determined as SOS and EOS respectively and the LOS is calculated; wherein the calculation formula of the growth period length LOS and the normalized vegetation index NDVI threshold NDVI th is as follows:

[0095] LOS=EOS-SOS (5)

[0096] NDVI th =(NDVI max -NDVI min )×r+NDVI min (6)

[0097] (6) In the formula: NDVI th is the NDVI threshold, NDVI max is the maximum value of NDVI, and NDVI minNDVImin, r is NDVI th the calculated parameter setting value.

[0098] S333, rice is identified based on the judgment rule of SOS e [a1, a2], SOS e [b1, b2] and LOS > g, and rice distribution data in the cloud area is obtained, wherein the values of a1, a2, b1, b2 and g are obtained based on the ground sample set by trial and error method. In this embodiment, the preferred setting values of a1, a2, b1, b2 and g under early rice condition are 110, 140, 180, 210 and 70 respectively, and the preferred setting values of a1, a2, b1, b2 and g under late rice condition are 195, 225, 275, 305 and 70 respectively; in other embodiments, the above parameters can be adaptively adjusted according to actual conditions, which are not limited here.

[0099] S34, the rice distribution data in the cloud area and the rice distribution data in the cloud area are integrated to obtain the spatial distribution data of the rice planting area. In the prior art, a single image is used for recognition based on its spectral and spatial characteristics, which is not only limited by the available image device and its imaging quality, but also prone to the problem that the image in the recognition period is disturbed by clouds and rain, which significantly reduces the reliability of the crop planting distribution information extraction result, and it is difficult to adapt to the accurate recognition of complex climate environment; in the S3 step of the present application, the decision tree algorithm is used to identify the rice planting area in the farmland area grid data under the conditions of no cloud and cloud, and the spatial distribution data of the rice planting area is obtained after integration, so that the obtained spatial distribution data of the rice planting area can more comprehensively reflect the distribution state of the rice planting area under different cloud conditions. Compared with the prior art, the recognition applicability and accuracy of complex environmental conditions are better.

[0100] S4, integrating the non-farmland area grid data and the spatial distribution data of the rice planting area to obtain a classification map with agricultural planting area and / or rice planting area identification.

[0101] In some embodiments, the S4 step specifically includes the following steps: integrating the non-farmland area grid data and the spatial distribution data of the rice planting area, using different numerical values and / or colors to identify different classifications in the agricultural planting area and the rice planting area respectively, to obtain a classification map with agricultural planting area and / or rice planting area identification. The numerical values and / or colors corresponding to different types of regions represented by the classification map can be adaptively selected according to actual needs, and the dimensions of different types represented by the classification map can also be adaptively adjusted according to actual needs, which are not limited here.

[0102] As Figure 2As shown in the second aspect, the embodiments of the present application provide a remote sensing identification device for rice planting area distribution. The device is used to implement the remote sensing identification method for rice planting area distribution as described in the first aspect. The device comprises: a data acquisition module 101 configured to acquire original satellite images and input SRTM DEM data, Planet images, GlobeLand 30 and field research data to form a classification feature set and a ground sample set; a data training module 102 configured to receive the classification feature set and the ground sample set from the data acquisition module 101 and perform training learning and data conversion to obtain farmland area raster data and non-farmland area raster data; a decision operation module 103 configured to receive the farmland area raster data from the data training module 102, and perform rice planting area identification on the farmland area raster data under cloud-free conditions and cloud conditions by using a decision tree algorithm, and integrate to obtain rice planting area spatial distribution data; and an image output module 104 configured to receive the non-farmland area raster data from the data training module 102 and the rice planting area spatial distribution data from the decision operation module 103, and integrate to output a classification map with agricultural planting area and / or rice planting area identification.

[0103] In the embodiments, the original satellite images are first acquired by using the data acquisition module 101 and the SRTM DEM data, Planet images, GlobeLand 30 and field research data are input to form a classification feature set and a ground sample set. Then, the classification feature set and the ground sample set are transmitted to the data training module 102 for training learning and data conversion to obtain farmland area raster data and non-farmland area raster data. Then, the farmland area raster data is transmitted to the decision operation module 103 for rice planting area identification on the farmland area raster data under cloud-free conditions and cloud conditions by using a decision tree algorithm, and the rice planting area spatial distribution data is obtained after integration. Finally, the non-farmland area raster data from the data training module 102 and the rice planting area spatial distribution data from the decision operation module 103 are integrated at the image output module 104 to output a classification map with agricultural planting area and / or rice planting area identification.

[0104] In the third aspect, the embodiments of the present application provide a computer readable storage medium having a computer program stored thereon. When the program is run by the device in the second aspect, the program is used to implement the remote sensing identification method for rice planting area distribution as described in the first aspect. The storage medium can be any various types of memory electronic devices or storage electronic devices.

[0105] The term "storage medium" is intended to include an installation medium, e.g., CD- ROM, floppy disks or tapes, computer system memory or random access memory such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc., non-volatile memory such as flash memory, magnetic media, registers or other similar types of storage. The term "storage medium" can also include two or more storage mediums that can reside in different locations, e.g., in different computer systems that are connected over a network. The storage medium can store a program or programs that can be executed by one or more processors. The term "computer program medium" can include a storage medium.

[0106] In a fourth aspect, the embodiments of the present application provide a remote sensing identification system for rice planting area distribution, which can include the device described in the second aspect or hardware or equipment capable of realizing the functions of all modules of the device described in the second aspect. The remote sensing identification system for rice planting area distribution is used to execute the remote sensing identification method for rice planting area distribution described in the first aspect.

[0107] Some specific examples are listed below. It should be noted that the examples described below are exemplary and are only used to explain the present application and cannot be understood as a limitation of the present application. If the specific technology or condition is not specified in the examples, it is carried out according to the technology or condition described in the literature in the field or according to the product instruction. If the reagent or instrument used is not specified by the manufacturer, it is a conventional product that can be obtained by purchase. The following examples and comparative examples are selected from the same Ganfu plain irrigation area for early rice and late rice planting distribution extraction experiment, and the test time is 2020.

[0108] Example 1

[0109] Please refer to Figure 3 The specific steps of the remote sensing identification method for rice planting area distribution in the present embodiment are as follows:

[0110] 1) Resample each band of a plurality of original satellite images of different periods and different weather conditions to 10m spatial resolution using the GEE remote sensing big data cloud platform, as shown in Figure 4 Select the median value of the original satellite images of Sentinel-2 in the double-cropping rice growing period to synthesize a clear satellite image (i.e., preprocessing).

[0111] 2) Select the visible light band that can distinguish the ground objects in the clear satellite image and the infrared band, near-infrared band and short-wave infrared band that can distinguish the plant growth conditions as classification and identification bands, construct a gray tone space correlation matrix about the clear satellite image using the step probability statistical method, and calculate the angular second matrix, contrast, correlation and variance of the classification and identification bands based on a 3x3 window and 64 gray scale quantization levels, as shown inFigure 5 As shown in FIG. 5, the texture feature data is obtained.

[0112] 3) SRTM DEM data with a spatial resolution of 30m is obtained by using the GEE remote sensing big data platform, and slope information is calculated based on the SRTM (Shuttle Radar Topography Mission) raster digital elevation model data, as shown in FIG. 6. Figure 6 As shown in FIG. 6, the terrain feature data is obtained. The texture feature data, the classification recognition band and the terrain feature data are integrated to obtain a classification feature set.

[0113] 4) A ground sample set is obtained by selecting a land use sample set from the Planet image and the GlobeLand 30 land use product data set, and combining the field sampling sample set obtained by field investigation; based on the ground sample set, the classification feature set is trained and learned by using the classification regression tree CART, the support vector machine SVM and the random forest RF supervised learning algorithm, the accuracy of the training and learning result is verified, the random forest RF is determined as the best method of the embodiment, and the land use classification data about the farmland area and the non-farmland area is obtained by using the method, as shown in FIG. 7. Figure 7

[0114] 8) The land use classification data is converted into raster data and revalued, the farmland area is valued as 1, and the non-farmland area is valued as NoData, to obtain farmland area raster data and non-farmland area raster data; wherein the land use classification data of the non-farmland area includes water body, grassland, building land and unused land.

[0115] 9) The farmland area raster data is converted into vector data to obtain shp vector data containing only the farmland area.

[0116] 10) Under the condition of no cloud, the original satellite image under the condition of multiple periods of clear sky and no cloud in the rice growing period is cropped based on the shp vector data containing only the farmland area to obtain the original satellite image containing only the farmland area, and the land water body index LSWI data set and the dual-band enhanced vegetation index EVI2 data set are calculated. The calculation formula of the land water body index LSWI and the dual-band enhanced vegetation index EVI2 is as follows:

[0117]

[0118] In formulas (1) and (2), LSWI is the land water body index, ρ NIR represents the reflectivity of the near-infrared band, ρ Red represents the reflectivity of the red light band, and EVI2 is the dual-band enhanced vegetation index.

[0119] ​The rice heading date and tillering date (40 days before the heading date) were determined based on the maximum value of the dual-band enhanced vegetation index (EVI2) dataset. The EVI2 amplitude C of the phenological stage of the tillering date was calculated. v , the minimum value of LSWI min LSWI maximum value LSWI max and LSWI amplitude C w ,like Figure 8 As shown, the ratio of the LSWI amplitude to the EVI2 amplitude RCLE is then calculated. The calculation formula for the ratio of the LSWI amplitude to the EVI2 amplitude RCLE is as follows:

[0120] RCLE=C W / C V (3)

[0121] Based on LSWI min The judgment rule of >0.10 and RCLE<0.42 was used to identify rice distribution, and the rice distribution data in cloud-free areas were obtained.

[0122] 11) Under cloudy conditions, original satellite images from multiple periods of clear, cloudless skies during the rice growing season were cropped based on the shp vector data containing only farmland areas. This yielded a normalized difference vegetation index (NDVI) dataset, which was then smoothed using a Savitzky-Golay filter to produce a smoothed NDVI curve. The NDVI is calculated using the following formula:

[0123]

[0124] Based on the smoothed NDVI data curve (i.e., the smoothed NDVI rice growth process curve), each pixel in the farmland area is judged and phenological indicators are extracted to obtain phenological indicator raster data. The phenological indicator raster data includes three phenological indicators: the start of growth period SOS, the end of growth period EOS, and the length of the growth period LOS. In this step, the specific steps for judging each pixel in the farmland area and extracting phenological indicators are as follows: using the normalized vegetation index NDVI threshold NDVI th To judge the conditions, the first time above and below NDVI th The dates of the growing season are determined as SOS and EOS respectively and LOS is calculated; among them, the length of the growing season LOS is related to the normalized vegetation index NDVI threshold NDVI th The calculation formula is as follows:

[0125] LOS=EOS-SOS (5)

[0126] (5) where LOS is the length of growing season, EOS is the end of growing season, and SOS is the start of growing season.

[0127] NDVI th = (NDVI max - NDVI min ) x r + NDVI min (6)

[0128] (6) where NDVI th is the NDVI threshold, NDVI max is the maximum NDVI, NDVI min is the minimum NDVI, r is 0.3, and NDVI th The calculation results are shown in Table 1. Figure 9

[0129] S333, based on the judgment rule of SOS ∈ [α1, α2], SOS ∈ [β1, β2] and LOS > γ, the rice is identified, and the rice distribution data in the cloudy area is obtained, where the set values of α1, α2, β1, β2 and γ under early rice are 110, 140, 180, 210 and 70 respectively, and the set values of α1, α2, β1, β2 and γ under late rice are 195, 225, 275, 305 and 70 respectively.

[0130] 12) The rice distribution data in the cloud-free area and the rice distribution data in the cloudy area are integrated to obtain the spatial distribution data of the rice planting area.

[0131] 13) The non-agricultural land area grid data and the spatial distribution data of the rice planting area are integrated, different numerical values and colors are used to identify different classifications in the agricultural planting area and the rice planting area, and a classification map with identification of the agricultural planting area and the rice planting area is obtained, as shown in Table 2. Figure 10

[0132] Comparative Example 1

[0133] Comparative Example 1 uses the existing hierarchical classification method to identify the distribution of early rice growing season. The first layer uses machine learning algorithm as the classifier to classify land use and extract cultivated land information. On this basis, the second layer uses decision tree classification algorithm to identify rice. The specific steps are as follows:

[0134] 1) Obtain long sequence remote sensing images and elevation data required for constructing classification feature set through GEE cloud platform online, and perform data processing and feature parameter extraction.

[0135] ​​2) Based on the classification feature set, the early rice distribution information is extracted in layers on the GEE platform: first, a machine learning algorithm is used as a classifier to realize land use classification, and the classification accuracy is evaluated; second, based on the extracted farmland area distribution information, a decision tree classification algorithm is used to identify the rice planting area, and the accuracy is verified, and finally the rice planting area identification classification map of the Ganfu Plain irrigation area is obtained.

[0136] The test results are shown in Figure 11 , wherein (a) five regions (I-V) are randomly selected for classification result comparison, (b) I-1-V-1 are the 10m spatial resolution views of the five regions in the early rice growing period in the Sentinel-2 image in (a), (c) I-2-V-2 are the early rice results of the double-cropping rice 10m resolution planting distribution data in the five regions in (a), and (d) I-3-V-3 are the early rice 10m resolution planting distribution data extracted by the layered classification method in the five regions in (a).

[0137] Comparative Example 2

[0138] Comparative Example 2 uses the existing layered classification method to identify the distribution of late rice in the growing period. The first layer uses a machine learning algorithm as a classifier to classify land use and extract farmland information. Based on this, the second layer uses a decision tree classification algorithm to identify rice. The specific steps are as follows:

[0139] 1) Through the GEE cloud platform, online access to long sequence remote sensing images and elevation data required for building classification feature sets, and data processing and feature parameter extraction.

[0140] 2) Based on the classification feature set, the late rice distribution information is extracted in layers on the GEE platform: first, a machine learning algorithm is used as a classifier to realize land use classification, and the classification accuracy is evaluated; second, based on the extracted farmland area distribution information, a decision tree classification algorithm is used to identify the rice planting area, and the accuracy is verified, and finally the rice planting area identification classification map of the Ganfu Plain irrigation area is obtained.

[0141] The test results are shown in Figure 12 , wherein (a) five regions (I-V) are randomly selected for classification result comparison, the five regions selected are the same as those in Comparative Example 1, (b) I-1-V-1 are the 10m spatial resolution views of the five regions in the late rice growing period in the Sentinel-2 image in (a), (c) I-2-V-2 are the late rice results of the double-cropping rice 10m resolution planting distribution data in the five regions in (a), and (d) I-3-V-3 are the late rice 10m resolution planting distribution data extracted by the layered classification method in the five regions in (a).

[0142] Result analysis

[0143] Combination Figures 10 to 12 , the extraction results of early rice and late rice of Example 1 and Comparative Examples 1-2 are basically consistent near the irrigation test center station of Jiangxi Province with less double-cropping rice planting area (III), proving that for the selected area with less distribution of rice planting area, the recognition accuracy of the method of the application is close to that of the prior art method. For the northeast (I), the middle (II) and the southeast (V) of the Ganfu Plain Irrigation Area, which is a selected area with high distribution of rice planting area and complex recognition, it can be seen from the comparison of the drawings that the extraction results of Example 1 are significantly better than those of the single-cropping rice planting distribution data set of Comparative Example 1 or Comparative Example 2. In addition, in the southwest (IV) where late rice is mainly planted, the late rice result extracted by Example 1 is better than the existing data set. It can be seen that the distribution information of the rice planting area covered in the classification map obtained by Example 1 is significantly more than that of Comparative Example 1 or Comparative Example 2, thereby proving that the method of the application shows better recognition accuracy and stability compared with the existing recognition technology, and can also adapt to recognition work under complex cloud / no cloud environment.

[0144] In summary, compared with the existing recognition technology, the method described in the application not only ensures the recognition accuracy and stability of the rice planting area without the need for actual measurement samples, but also retains more and richer recognition information under complex environmental conditions, achieving efficient recognition of the rice planting area.

[0145] It should be noted that the application is not limited to the above-mentioned embodiments. The above-mentioned embodiments are only examples, and embodiments having the same technical idea and playing the same role and effect within the scope of the technical solutions of the application are all included in the technical scope of the application. In addition, within the scope of the main idea of the application, various modifications of the embodiments that can be thought of by those skilled in the art, as well as other ways constructed by combining part of the constituent elements in the embodiments, are also included in the scope of the application.

Claims

1. A remote sensing identification method for rice planting area distribution, characterized in that: The steps include: S1, constructs a classification feature set for original satellite images based on satellite remote sensing images and SRTM DEM data, and constructs a ground sample set based on Planet images, GlobeLand 30 data, and field survey data; S2, based on the ground sample set, using classification and regression tree CART, support vector machine SVM and random forest RF supervised learning algorithms to train and learn the classification feature set and perform data conversion to obtain farmland area raster data and non-farmland area raster data; S3, using a decision tree algorithm to identify rice-growing areas on the farmland raster data under cloudless conditions and cloud conditions, and integrating the data to obtain spatial distribution data of the rice-growing areas; S4, integrating the non-agricultural area raster data with the rice-growing area spatial distribution data to obtain a classification map with identifications of agricultural planting areas and / or rice-growing areas; The classification feature set includes texture feature data, classification identification bands and terrain feature data of the original satellite image.

2. The remote sensing identification method for rice planting area distribution according to claim 1, characterized in that: The S1 step specifically includes the following steps: S11, using satellite remote sensing images to obtain original satellite images of different periods and meteorological conditions, selecting original satellite images during the double-season rice growing period, and obtaining clear satellite images after preprocessing; S12, selecting different distinguishable bands in the clear satellite image as classification and identification bands, constructing a gray tone spatial correlation matrix for the clear satellite image using an order probability statistical method, and calculating the angular matrix, contrast, correlation, and variance of the classification and identification bands based on a 3×3 window and 64 grayscale quantization levels to obtain the texture feature data; S13, using the GEE remote sensing big data platform to obtain SRTM DEM data with a spatial resolution of 30m, extracting elevation based on the SRTM DEM data and calculating slope information, that is, obtaining the terrain feature data.

3. The remote sensing identification method for rice planting area distribution according to claim 2, characterized in that: In step S11, each band of the original satellite images of several different periods and different meteorological conditions is resampled to a spatial resolution of 10m using the GEE remote sensing big data cloud platform, the median of the original satellite images of Sentinel-2 during the double-season rice growth period is selected to synthesize the clear satellite image, and the texture feature data is obtained based on the clear satellite image; In the step S12, the classification and identification bands include a visible light band that can distinguish ground objects and an infrared band, a near-infrared band and a short-wave infrared band that can distinguish plant growth conditions.

4. The remote sensing identification method for rice planting area distribution according to claim 1, characterized in that: The S2 step specifically includes the following steps: Based on the ground sample set, the classification feature set is trained and learned using classification regression tree (CART), support vector machine (SVM), and random forest (RF) supervised learning algorithms to obtain land use classification data on farmland areas and non-farmland areas, and the land use classification data is converted into raster data and reassigned to obtain the farmland area raster data and non-farmland area raster data; The land use classification data of the non-agricultural area includes water bodies, forests and grasslands, construction land and unused land.

5. The remote sensing identification method for rice planting area distribution according to claim 1, characterized in that: The S3 step specifically includes the following steps: S31, converting the farmland raster data into vector data to obtain shp vector data containing only the farmland area; S32, under cloudless conditions, calculating rice distribution data in the cloudless area based on the shp vector data containing only the farmland area; S33, under cloudy conditions, obtaining rice distribution data in cloudy areas based on the shp vector data containing only the farmland area and the rice phenology index and normalized difference vegetation index dataset; S34, integrating the rice distribution data in the cloudless area and the rice distribution data in the clouded area to obtain the spatial distribution data of the rice-growing area; The S32 step specifically includes: S321, under cloudless conditions, cropping original satellite images under multiple periods of clear and cloudless sky during the rice growing season based on the shp vector data containing only the farmland area to obtain original satellite images containing only the farmland area, and respectively calculating a Land Water Index (LSWI) dataset and a Dual-Band Enhanced Vegetation Index (EVI2) dataset; S322, determining the rice heading date and subsequently the tillering date based on the maximum value in the dual-band enhanced vegetation index EVI2 data set, and calculating the EVI2 amplitude C of the phenological stage of the tillering date. v , the minimum value of LSWI min LSWI maximum value LSWI max and LSWI amplitude C w , then the ratio RCLE of the LSWI amplitude to the EVI2 amplitude is calculated; S323, based on LSWI min The judgment rule of >θ1 and RCLE<θ2 identifies the rice distribution and obtains the rice distribution data in the cloud-free area, wherein the values ​​of θ1 and θ2 are obtained by trial and error based on the ground sample set; The calculation formulas for the land water index LSWI, the dual-band enhanced vegetation index EVI2, and the ratio of the LSWI amplitude to the EVI2 amplitude RCLE are as follows: RCLE=C W / C V (3) The S33 step specifically includes: S331, under cloudy conditions, cropping original satellite images under multiple periods of clear and cloudless conditions during the rice growing season based on the shp vector data containing only the farmland area to obtain original satellite images containing only the farmland area, calculating a normalized difference vegetation index (NDVI) dataset, and smoothing the normalized difference vegetation index (NDVI) dataset using a Savitzky-Golay filter to obtain a smoothed NDVI data curve; S332, based on the NDVI data smoothing curve, judge each pixel in the farmland area and extract phenological indicators to obtain phenological indicator raster data, wherein the phenological indicator raster data includes three phenological indicators: the start of growth period (SOS), the end of growth period (EOS), and the length of the growing period (LOS); S333, identifying rice based on the judgment rules of SOS∈[α1, α2], SOS∈[β1, β2], and LOS>γ, and obtaining rice distribution data in the cloud area, wherein the values ​​of α1, α2, β1, β2, and γ are all obtained by trial and error based on a ground sample set; The calculation formula of normalized vegetation index NDVI is as follows: In formulas (1) to (4), ρ NIR represents the reflectivity in the near-infrared band, ρ Red Indicates the reflectivity of the red light band; The specific steps for judging each pixel in the farmland area and extracting phenological indicators are as follows: Normalized Difference Vegetation Index (NDVI) threshold th To judge the conditions, the first time above and below NDVI th The dates are determined as SOS and EOS respectively and LOS is calculated; Among them, the length of the growing season LOS and the normalized difference vegetation index NDVI threshold NDVI th The calculation formula is as follows: LOS=EOS-SOS (5) NDVI th =(NDVI max -NDVI min )×r+NDVI min (6) (6) Where: NDVI th is the NDVI threshold; NDVI max is the maximum value of NDVI; min is the minimum value of NDVI; r is the NDVI th The calculation parameter setting value.

6. The remote sensing identification method for rice planting area distribution according to claim 5, characterized in that: In the step S323, the setting value of θ1 is 0.10, and the setting value of θ2 is 0.42; In step S333, the set values ​​of α1, α2, β1, β2 and γ under the early rice condition are 110, 140, 180, 210 and 70 respectively, and the set values ​​of α1, α2, β1, β2 and γ under the late rice condition are 195, 225, 275, 305 and 70 respectively.

7. The remote sensing identification method for rice planting area distribution according to claim 1, characterized in that: The S4 step specifically includes the following steps: The non-agricultural area raster data is integrated with the spatial distribution data of the rice-growing area, and different numerical values ​​and / or colors are used to identify different classifications in the agricultural planting area and the rice-growing area, so as to obtain a classification map with identification of the agricultural planting area and / or the rice-growing area.

8. A remote sensing identification device for rice planting area distribution, characterized in that: The device comprises: The data acquisition module is used to collect original satellite images and input SRTM DEM data, Planet images, GlobeLand30 and field survey data to form classification feature sets and ground sample sets; A data training module is used to receive the classification feature set and the ground sample set from the data acquisition module and perform training learning and data conversion to obtain raster data of the farmland area and raster data of the non-farmland area; a decision operation module, configured to receive the farmland area raster data from the data training module, identify rice planting areas on the farmland area raster data under cloudless conditions and cloud conditions using a decision tree algorithm, and obtain spatial distribution data of rice planting areas after integration; An image output module is configured to receive the non-agricultural area raster data from the data training module and the rice planting area spatial distribution data from the decision operation module, integrate them and output a classification map with identification of agricultural planting areas and / or rice planting areas; The remote sensing identification device for rice planting area distribution is used to implement the remote sensing identification method for rice planting area distribution as claimed in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The medium stores a computer program, which is used to implement the remote sensing identification method for rice planting area distribution as claimed in any one of claims 1 to 7 when executed.

10. A remote sensing identification system for rice planting area distribution, characterized in that: The remote sensing identification system for rice-growing area distribution is used to execute the remote sensing identification method for rice-growing area distribution as claimed in any one of claims 1 to 7.

Images