Crop planting area extraction method and device

Abstract

The present invention provides a method and device for extracting crop planting area. The specific steps of the method include: obtaining remote sensing image data of a target area, preprocessing, vectorizing and rasterizing the data to obtain field raster data; obtaining sample data of several crops in the field of the target area to generate training samples; obtaining Sentinel‑1A image data of the target area, preprocessing and spatiotemporal filtering the data to obtain backscattering time series data, and constructing a crop time series backscattering curve; constructing typical phenological characteristics of crops; extracting a crop type distribution map using a classifier; masking the crop type distribution map using the field raster data to obtain a field crop type distribution map; estimating crop planting area. The crop planting area extraction method and device described in the present invention combines Sentinel‑1A time series data with a higher revisit period with field raster data, which can effectively solve the problem of low crop classification accuracy in cloudy and rainy areas due to weather influences.

Description

Technical Field

[0001] The invention relates to the technical field of agricultural remote sensing information extraction, and in particular to a method and device for extracting crop planting area. Background Art

[0002] Crop planting information and spatial distribution are of great significance to agricultural production and meteorological disaster management departments in formulating agricultural policies and post-disaster relief. Compared with traditional agricultural statistical data and sampling survey methods, satellite remote sensing technology can obtain large-scale, multi-temporal image data, with the characteristics of macro, timeliness and economy, providing strong technical support for the rapid and accurate acquisition of crop planting information.

[0003] The current crop information extraction technology based on remote sensing images is mainly optical remote sensing technology, and its extraction principle is: different types of crops show different spectral characteristics in each band of optical remote sensing images; through this spectral characteristic difference, different crop types can be distinguished. However, since the key phenological period of crops in southern my country is often affected by cloudy, rainy and foggy weather, it is difficult to obtain effective optical data for crop classification, which leads to certain limitations in the practical application of optical remote sensing technology. Summary of the invention

[0004] Based on this, the purpose of the present invention is to provide a crop planting area extraction method and device that is independent of weather conditions and solves the problem that optical remote sensing technology is affected by clouds and revisit cycles, resulting in low crop classification accuracy in cloudy and rainy areas.

[0005] A method for extracting crop planting area comprises the following steps:

[0006] Acquire remote sensing image data of the target area, pre-process the acquired remote sensing image data, and then perform vectorization and rasterization to obtain field raster data of the target area;

[0007] Obtain sample data of several crops in the target area, mark the types of the crops, and generate training samples of the crops in the target area;

[0008] Acquire Sentinel-1A image data of the target area, pre-process the acquired Sentinel-1A image data, perform spatiotemporal filtering to obtain crop backscatter time series data, and construct a crop time series backscatter curve according to the crop backscatter time series data;

[0009] The spatiotemporal filtering performed on the preprocessed Sentinel-1A image data includes spatial domain filtering and temporal domain filtering;

[0010] The spatial domain filtering is to perform spatial field mean filtering on the Sentinel-1A image data of the full-year time series; the field mean filtering is to extract all observable values ​​of the training samples in the Sentinel-1A image data, calculate the average value of all observable values, and obtain the backscatter coefficient value of a single time series of each type of crop;

[0011] The time domain filtering is to obtain the backscatter coefficient value of a single time series of each type of crop by the spatial domain filtering, and then take the median of the backscatter coefficient value of each type of crop at the same time to obtain the backscatter time series data under the time and space filtering;

[0012] According to the variation law of the crop time series backscattering curve, the crop time series backscattering curve is divided into different stages to obtain typical phenological characteristics of the crop;

[0013] Inputting the backscattering time series data and the typical phenological characteristics of crops into a preset crop classifier, and training the typical phenological characteristics of crops of the training samples to obtain a crop classification model; using the crop classification model to classify the backscattering time series data of the preprocessed Sentinel-1A image data to obtain a crop type distribution map of the target area;

[0014] Using the field grid data to perform mask processing on the crop type distribution map, to obtain a field crop type distribution map of a target area;

[0015] The size of the crop planting area is estimated based on the spatial resolution of the Sentinel-1A image and the number of crop pixels in the crop type distribution map of the field.

[0016] The crop planting area extraction method described in the present invention combines Sentinel-1A time series data with a higher revisit period with field raster data, which can effectively solve the problem that optical remote sensing technology is affected by weather and has low crop classification accuracy in cloudy and rainy areas.

[0017] Furthermore, after acquiring the remote sensing image data of the target area, the remote sensing image data of the target area is spliced ​​and cropped to obtain a complete remote sensing image of the target area.

[0018] Furthermore, after acquiring the Sentinel-1A image data of the target area, the method further includes splicing and cropping the Sentinel-1A image data of the target area to obtain complete Sentinel-1A image data of the target area.

[0019] Furthermore, in the step of acquiring remote sensing image data of the target area, the remote sensing image data is GF-2 image data; and preprocessing of the GF-2 image data includes radiation calibration, atmospheric correction and orthorectification.

[0020] Furthermore, the preprocessing of the Sentinel-1A image data includes thermal noise removal, orbit precision correction, radiation calibration, terrain correction and normalization processing.

[0021] Furthermore, the spatiotemporal filtering performed on the preprocessed Sentinel-1A image data includes spatial domain filtering and temporal domain filtering.

[0022] Furthermore, the spatial domain filtering is to perform spatial field mean filtering on the Sentinel-1A image data of the full-year time series; the field mean filtering is to extract all observable values ​​of the training samples in the Sentinel-1A image data, calculate the average value of all the observable values, and obtain the backscatter coefficient value of a single time series for each type of crop.

[0023] Furthermore, the time domain filtering is to obtain the backscatter coefficient value of a single time series of each type of crop by the spatial domain filtering, and then take the median of the backscatter coefficient value of each type of crop at the same time to obtain the backscatter time series data under the spatiotemporal filtering.

[0024] Furthermore, after obtaining the sample data of several crops in the target area fields, it also includes generating verification samples of the crops in the target area fields; after estimating the crop planting area, two classification accuracy evaluation systems, namely error matrix evaluation and area accuracy evaluation, are constructed based on the data of the verification samples.

[0025] The present invention also provides a crop planting area extraction device, which is characterized by comprising:

[0026] A remote sensing image data acquisition module is used to acquire remote sensing image data, pre-process the remote sensing image data, and then perform vectorization and rasterization to obtain field raster data of the target area;

[0027] A crop sample acquisition module is used to acquire sample data of several crops in a field in a target area, mark the types of the crops, and generate training samples of the crops in the field in the target area;

[0028] A Sentinel-1A image data acquisition module, used to acquire Sentinel-1A image data, perform spatiotemporal filtering on the Sentinel-1A image data after preprocessing, and construct a crop time series backscattering curve based on the crop backscattering time series data;

[0029] Wherein, the spatiotemporal filtering performed on the preprocessed Sentinel-1A image data includes spatial domain filtering and temporal domain filtering;

[0030] The spatial domain filtering is to perform spatial field mean filtering on the Sentinel-1A image data of the full-year time series; the field mean filtering is to extract all observable values ​​of the training samples in the Sentinel-1A image data, calculate the average value of all observable values, and obtain the backscatter coefficient value of a single time series of each type of crop;

[0031] The time domain filtering is to obtain the backscatter coefficient value of a single time series of each type of crop by the spatial domain filtering, and then take the median of the backscatter coefficient value of each type of crop at the same time to obtain the backscatter time series data under the time and space filtering;

[0032] A crop typical phenological characteristics generating module is used to divide the crop time series backscattering curve into different stages according to the variation law of the crop time series backscattering curve to obtain the crop typical phenological characteristics;

[0033] A crop type extraction module inputs the backscatter time series data and the typical phenological characteristics of crops into a preset crop classifier, and trains the typical phenological characteristics of crops of the training samples to obtain a crop classification model; and uses the crop classification model to classify the backscatter time series data of the preprocessed Sentinel-1A image data to obtain a crop type distribution map of the target area;

[0034] A mask processing module, used for performing mask processing on the crop type distribution map using the field raster data to obtain a field crop type distribution map;

[0035] The crop planting area calculation module is used to estimate the size of the crop planting area according to the spatial resolution of the Sentinel-1A image and the number of crop pixels in the field crop type distribution map.

[0036] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flow chart of the crop planting area extraction method in an embodiment of the present invention;

[0038] Figure 2 A field distribution map of a target area surveyed on the spot in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the distribution of crop type sample points in the target area in an embodiment of the present invention;

[0040] Figure 4 It is a backscattering curve diagram of the training sample points under co-polarization in the target area in an embodiment of the present invention;

[0041] Figure 5 is a backscattering curve diagram of the training sample points under cross-polarization in the target area in an embodiment of the present invention;

[0042] Figure 6 A distribution map of crop types in a target area in an embodiment of the present invention;

[0043] Figure 7 Schematic diagram of the structure of the crop planting area extraction device in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below through examples so that those skilled in the art can implement the invention with reference to the description.

[0045] like Figure 1 As shown, the crop planting area extraction method of this embodiment includes the following specific steps:

[0046] S1. Acquire remote sensing image data of the target area, pre-process the acquired remote sensing image data, and then perform vectorization and rasterization to obtain field raster data of the target area.

[0047] The present invention selects a crop-planting area as a target area, and the remote sensing image data of the target area can be obtained from the website of the China Resources Satellite Application Center.

[0048] In one embodiment, the acquired remote sensing image data of the target area is the GF-2 image data of the target area, and the GF-2 image data can be acquired from the GF-2 satellite data on the website of the China Resources Satellite Application Center.

[0049] In one embodiment, the method further includes performing stitching and cropping operations on the acquired GF-2 image data to obtain complete GF-2 image data.

[0050] Then, the complete GF-2 image data is preprocessed, and the preprocessing includes radiation calibration, atmospheric correction and orthorectification. The preprocessed GF-2 image data is then vectorized and rasterized to obtain the field raster data of the target area.

[0051] Specifically, the pre-processing radiation calibration of the complete GF-2 image is to convert the remote sensing data received by the sensor into radiation brightness or zenith reflectivity, and perform radiation calibration calculation on the GF-2 image data according to the sensor type of the acquired data and the corresponding calibration coefficient. The atmospheric correction is intended to eliminate the influence of factors such as atmosphere and illumination on the surface reflectivity, eliminate the radiation error caused by the atmospheric influence, and obtain the reflectivity and emissivity of the object, surface temperature and other real physical model parameters. The orthorectification is to select some ground control points on the GF-2 image, and use the digital elevation model (DEM) data within the image range to simultaneously perform tilt correction and projection error correction on the image, resample the image into an orthophoto, and correct the image point displacement caused by terrain undulation and sensor error.

[0052] The field grid data of the target area is obtained by vectorizing the pre-processed GF-2 image data to obtain field vector data, and then rasterizing the field vector data.

[0053] S2. Obtain sample data of several crops in the target area, mark the types of the crops, and generate training samples of the crops in the target area.

[0054] Specifically, the fields in the target area are taken as sampling points, and the crop types of the fields are marked, crop samples of the fields in the target area are obtained, and training samples of the crops in the fields in the target area are generated.

[0055] In one embodiment, it also includes generating verification samples of crops in the target area field, and dividing the crop samples in the target area field into training samples and verification samples according to the results of multiple division experiments.

[0056] S3. Acquire Sentinel-1A image data of the target area, pre-process the acquired Sentinel-1A image data, perform spatiotemporal filtering to obtain crop backscatter time series data, and construct a crop time series backscatter curve based on the crop backscatter time series data.

[0057] Specifically, the Sentinel-1A image data of the target area is multi-temporal Sentinel-1A image data of the whole year, and the multi-temporal Sentinel-1A image data of the target area can be obtained from the website of the Copernicus Data Center of the European Space Agency.

[0058] In one embodiment, the method further includes stitching and cropping the acquired multi-phase Sentinel-1A image data to obtain complete Sentinel-1A image data.

[0059] Then, the complete Sentinel-1A image data is preprocessed, and the preprocessing of the multi-phase Sentinel-1A image data includes thermal noise removal, orbit precision correction, radiation calibration, terrain correction and normalization processing.

[0060] Specifically, the preprocessing thermal noise removal of multi-phase Sentinel-1A image data is to improve the accuracy of radar backscatter signals. The orbit precision correction can update more accurate satellite orbit positions. The radiation calibration is to convert the received backscatter signals into backscatter coefficients. The terrain correction is to eliminate the influence of terrain undulations around the measuring point on the observation results. The normalization process is to normalize the backscatter coefficients.

[0061] Specifically, the spatiotemporal filtering includes spatial filtering and temporal filtering. The spatial filtering is to perform spatial mean filtering on the preprocessed Sentinel-1A image data, and the temporal filtering is to perform median filtering in the temporal domain on the Sentinel-1A image data after the spatial filtering. The field mean filtering is to extract all observable values ​​of the training samples in each field in the Sentinel-1A image data, and then calculate the average value of the observable values ​​of each field to obtain the backscatter coefficient value of each type of crop in a single time series; the median filtering in the temporal domain is to take the median of the backscatter coefficient value of each type of crop at the same time on the basis of the Sentinel-1A image data subjected to spatial filtering, and obtain the backscatter time series data under spatiotemporal filtering.

[0062] In one embodiment, all observable values ​​of training samples in each field extracted from the Sentinel-1A image data are observation values ​​under co-polarization (VV) and cross-polarization (VH) modes; based on the backscattering time series data, crop time series backscattering curves under co-polarization (VV) and cross-polarization (VH) modes are constructed.

[0063] S4. According to the variation law of the crop time series backscattering curve, the crop time series backscattering curve is divided into different stages to obtain typical phenological characteristics of the crop.

[0064] In one embodiment, according to the phenological characteristics of the crop time series backscatter curve, the backscatter curve is divided into three stages. The first stage corresponds to 1 to a observations, the second stage corresponds to (a+1) to b observations, and the third stage corresponds to (b+1) to c observations; and the panchromatic band data in the GF-2 image is used to construct the typical phenological characteristics of crops (Integral, Range, Pan), and the specific formulas of each characteristic index are as follows:

[0065] ρmin1 =minimum(ρ1…ρ a )

[0066] ρ max2 =maximum(ρ a+1 …ρ b )

[0067] ρ max3 =maximum(ρ b+1 …ρ c )

[0068]

[0069] Range = maximum (ρ 1…c )-minimum(ρ 1…c )

[0070] Pan=average cropfield (V PAN )

[0071] Ρ represents the backscatter coefficient value, i represents the i-th observation, ρ min1 represents the minimum value of the backscatter coefficient of each type of crop in the first stage, ρ max2 Indicates the maximum backscatter coefficient value of each type of crop in the second stage, ρ max3 represents the maximum backscatter coefficient value of each type of crop in the third stage, Integral represents the integral of the minimum backscatter coefficient value of the second and third stages relative to the first stage, Range represents the maximum difference of the backscatter coefficient value of each type of crop observed c times throughout the year, Pan represents the average value of the panchromatic band reflectance value corresponding to the field grid data (cropfield) described in step S1, V PAN Indicates the reflectance value of the panchromatic band in the GF-2 image.

[0072] S5. Input the backscatter time series data and typical crop phenological characteristics into a preset crop classifier, and train the typical crop phenological characteristics of the training samples to obtain a crop classification model; use the crop classification model to classify the backscatter time series data of the preprocessed Sentinel-1A image data to obtain a crop type distribution map of the target area.

[0073] Specifically, the backscattering time series data is the backscattering time series data under spatiotemporal filtering in S3; the typical phenological characteristics of the crop are the Integral, Range, and Pan values ​​calculated by the formula in S4.

[0074] The crop type distribution map is a crop type distribution map of the entire target area, including classification results of other non-field areas.

[0075] S6. Using the field grid data, masking the crop type distribution map is performed to obtain a field crop type distribution map of a target area.

[0076] Specifically, the mask processing is to use the field grid data described in S1 to remove the part identified as the non-field area in the crop type distribution map, and the remaining part is the field area, which can eliminate the interference of the non-field area and obtain the field crop type distribution map.

[0077] S7. Estimate the size of the crop planting area based on the spatial resolution of the Sentinel-1A image and the number of crop pixels in the crop type distribution map of the field.

[0078] Specifically, the estimated crop planting area is the product of the spatial resolution of the Sentinel-1A image and the number of crop pixels. The spatial resolution of the Sentinel-1A image is the resolution of the Sentinel-1A image data obtained by S3, and the number of crop pixels is the statistical number of pixels of each type of crop in the field crop type distribution map obtained by S6.

[0079] In one embodiment, the crop planting area extraction method further includes the following steps: based on the field survey data of the verification sample described in S2, two classification accuracy evaluation systems are constructed, namely: an error matrix evaluation system and an area accuracy evaluation system.

[0080] The error matrix evaluation system calculates the user accuracy (UA), producer accuracy (PA), overall accuracy (OA) and Kappa coefficient of the classification results through the confusion matrix accuracy evaluation method. The specific calculation formula is as follows:

[0081]

[0082] Where n is the total number of columns in the confusion matrix, i.e. the total number of crop categories, P ii is the number of correctly classified crop samples in the ith row and ith column of the confusion matrix, P i+ and P +i are the total number of crop samples in row i and column i respectively, and N is the total number of validation samples.

[0083] The index of the area accuracy evaluation system (P S ) can detect the difference in classification accuracy between different classification features. The study evaluates the total area accuracy of crop extraction results based on different polarization features (VH / VV). The specific calculation formula is as follows:

[0084]

[0085] In the formula, A i is the total crop planting area of ​​the extracted target area; A0 is the true value of the crop planting area in the statistical yearbook data.

[0086] In one embodiment of the present invention, taking the planting of sugarcane and banana in Nansha District, Guangzhou City, Guangdong Province as an example, in step S1, the acquired remote sensing image data of the target area is the GF-2 image data of Nansha District, Guangzhou City, Guangdong Province between November 17, 2019 and December 27, 2019. After splicing, cropping, preprocessing, vectorization and rasterization of the GF-2 image data, the field raster data of the area is obtained, and a total of 14,533 fields in Nansha District are extracted, such as Figure 2 As shown in the figure, it is mainly used as a filtering window in the spatiotemporal denoising process of the Sentinel-1A image in the subsequent target area, and is used to construct the crop time series backscattering curve and typical phenological characteristics in subsequent classification studies.

[0087] The average annual temperature in the region is 22.6°C, the annual sunshine duration is 1651.7 hours, and the average annual precipitation is 1673.1 mm. It is a suitable area for growing sugarcane and bananas. The 5422 hectares of sugarcane and 6051 hectares of bananas in Guangzhou are mainly concentrated in the coastal area, namely the Nansha District, the study area of ​​this embodiment. The total agricultural output value of the district in 2019 was 4.892 billion yuan, accounting for 52.1% of the agricultural, forestry, animal husbandry and fishery output value of the district, and it is a typical urban agricultural area in Lingnan. The planting of sugarcane and bananas studied in this embodiment is one year, and the entire growth cycle is relatively long. The specific growth period is shown in Table 1.

[0088] Table 1 Growth period of main crops in Nansha District

[0089]

[0090] In step S2, based on the remote sensing image data of Nansha District in 2019, the fields where the land type has not changed in the same period of 2021 are taken as sampling points, and a total of 442 typical crop samples are obtained, such as Figure 3 As shown in Table 2, the sample ratio is divided into training samples and validation samples at a ratio of 6:4.

[0091] Table 2 Number of training samples and validation samples

[0092]

[0093] In step S3, the acquired Sentinel-1A image data of the target area is the Sentinel-1A image data of Nansha District, Guangzhou City, Guangdong Province in 2019. After splicing, cropping and preprocessing the Sentinel-1A image data, spatiotemporal filtering is performed to obtain the backscattering time series data of the crops in the area. A crop time series backscattering curve is constructed based on the backscattering time series data, such as Figure 4 and 5 shown.

[0094] In one embodiment of the present invention, in step S4, by observing the backscatter coefficient curves of different crops, it is found that the curves of sugarcane and banana have more obvious characteristics, and the backscatter time curves of crops in Nansha area are divided into three stages:

[0095] The first stage was from early January to the end of April 2019 (corresponding to observations 1-10), when the backscatter coefficient values ​​of various crops were relatively low;

[0096] The second stage was from early May to mid-June 2019 (corresponding to observations 11–15), when the backscatter coefficient values ​​of sugarcane and banana increased rapidly;

[0097] The third stage was from late June to the end of 2019 (corresponding to observations 16-30), during which the backscatter coefficient values ​​of sugarcane gradually decreased, while those of bananas remained at a high value.

[0098] Based on the phenological characteristics of sugarcane and banana and the variation characteristics of their corresponding time series backscatter coefficient values, the present invention constructs several phenological characteristic variables (Integral, Range, Pan), and the specific formulas of each characteristic index are as follows:

[0099] ρ min1 =minimum(ρ1…ρ 10 )

[0100] ρ max2 =maximum(ρ 11 …ρ 15 )

[0101] ρ max3 =maximum(ρ 16 …ρ 30 )

[0102]

[0103] Range = maximum (ρ 1…30 )-minimum(ρ 1…30 )

[0104] Pan=average cropfield (V PAN )

[0105] Where ρ represents the backscatter coefficient value, i represents the i-th observation among the 30 observations throughout the year, and ρ min1 represents the minimum value of the backscatter coefficient of each type of crop in the first stage, ρ max2 Indicates the maximum backscatter coefficient value of each type of crop in the second stage, ρ max3 represents the maximum backscatter coefficient value of each type of crop in the third stage, Integral represents the integral of the minimum backscatter coefficient value of the second and third stages relative to the first stage, Range represents the maximum difference of the backscatter coefficient value of each crop within the range of 30 observations throughout the year, Pan represents the average value of the panchromatic band reflectance value corresponding to the field raster data obtained in Nansha District, V PAN Indicates the reflectance value of the panchromatic band in the GF-2 image.

[0106] In step S5, the crop classifier is an XGBoost classifier, and its parameter settings are shown in Table 3.

[0107] Table 3 Key parameters used in XGBoost classifier

[0108]

[0109]

[0110] The construction of the two classification accuracy assessment systems is based on 176 verification sample data and historical statistical data sampled on the spot in Nansha District as the true value of the accuracy evaluation. The crop classification accuracy and area estimation accuracy of sugarcane and banana in Nansha District are shown in Tables 4 and 5. The overall classification accuracy of this method is 94.31%, and the Kappa coefficient is 0.91. In addition, the extracted sugarcane and banana distribution results are compared with the crop planting area statistics in 2019. The extraction result of sugarcane is 4030.59 hectares, and the area accuracy reaches 82.04%; the extraction result of banana is 2673.63 hectares, and the area accuracy reaches 71.01%.

[0111] Table 4 Evaluation table of different crop recognition accuracy based on verification sample points

[0112]

[0113] Table 5 Evaluation table of the accuracy of different crop area estimation based on statistical data

[0114]

[0115] like Figure 6From the crop distribution shown, it can be seen that in the extracted results, the planting areas of sugarcane and banana are mainly concentrated in Dagang, Lanhe, and Hengli Towns, which is consistent with the results of the field survey.

[0116] In one embodiment of the present invention, a crop planting area extraction device is also provided. Figure 7 As shown, the crop planting area extraction device comprises:

[0117] The remote sensing image data acquisition module 10 is used to acquire remote sensing image data, and after pre-processing the remote sensing image data, perform vectorization and rasterization to obtain field raster data of the target area;

[0118] The crop sample acquisition module 11 is used to acquire sample data of several crops in the target area, mark the types of the crops, and generate training samples of the crops in the target area;

[0119] A Sentinel-1A image data acquisition module 12 is used to acquire Sentinel-1A image data, perform spatiotemporal filtering on the Sentinel-1A image data after preprocessing to obtain crop backscatter time series data, and construct a crop time series backscatter curve according to the crop backscatter time series data;

[0120] The crop typical phenological characteristics generating module 13 is used to divide the crop time series backscattering curve into different stages according to the variation law of the crop time series backscattering curve to obtain the crop typical phenological characteristics;

[0121] The crop type extraction module 14 is used to input the backscattering time series data and the typical phenological characteristics of crops into a preset crop classifier, and train the typical phenological characteristics of crops of the training samples to obtain a crop classification model; use the crop classification model to classify the backscattering time series data of the preprocessed Sentinel-1A image data to obtain a crop type distribution map of the target area;

[0122] A mask processing module 15 is used to perform mask processing on the crop type distribution map using the field grid data to obtain a field crop type distribution map;

[0123] The crop planting area calculation module 16 is used to estimate the size of the crop planting area according to the spatial resolution of the Sentinel-1A image and the number of crop pixels in the field crop type distribution map;

[0124] In one embodiment of the present invention, the crop planting area extraction device further includes a classification accuracy evaluation module 17 for constructing two classification accuracy evaluation systems, namely error matrix evaluation and area accuracy evaluation, based on the data of the verification sample.

[0125] The present invention combines high-resolution optical remote sensing images with synthetic aperture radar image information from all-weather observations, and uses the XGBoost algorithm and radar data spatiotemporal filtering strategy to accurately detect and extract crop planting information in Lingnan, effectively solving the problem of low classification accuracy due to the influence of weather on optical images and the presence of speckle noise on radar images. Due to the powerful parallel processing and generalization capabilities of the XGBoost machine algorithm, the classification accuracy and efficiency are greatly improved, facilitating information management of various crops.

[0126] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and the present invention is also intended to include these modifications and modifications.

Claims

1. A method for extracting crop planting area, comprising the following steps: Acquire remote sensing image data of the target area, pre-process the acquired remote sensing image data, and then perform vectorization and rasterization to obtain field raster data of the target area; Obtain sample data of several crops in the target area, mark the types of the crops, and generate training samples of the crops in the target area; Acquire Sentinel-1A image data of the target area, pre-process the acquired Sentinel-1A image data, perform spatiotemporal filtering to obtain crop backscatter time series data, and construct a crop time series backscatter curve according to the crop backscatter time series data; The spatiotemporal filtering performed on the preprocessed Sentinel-1A image data includes spatial domain filtering and temporal domain filtering; The spatial domain filtering is to perform spatial field mean filtering on the Sentinel-1A image data of the full-year time series; the field mean filtering is to extract all observable values ​​of the training samples in the Sentinel-1A image data, calculate the average value of all observable values, and obtain the backscatter coefficient value of a single time series of each type of crop; The time domain filtering is to obtain the backscatter coefficient value of a single time series of each type of crop by the spatial domain filtering, and then take the median of the backscatter coefficient value of each type of crop at the same time to obtain the backscatter time series data under the time and space filtering; According to the variation law of the crop time series backscattering curve, the crop time series backscattering curve is divided into different stages to obtain typical phenological characteristics of the crop; Inputting the backscattering time series data and the typical phenological characteristics of crops into a preset crop classifier, and training the typical phenological characteristics of crops of the training samples to obtain a crop classification model; using the crop classification model to classify the backscattering time series data of the preprocessed Sentinel-1A image data to obtain a crop type distribution map of the target area; Using the field grid data to perform mask processing on the crop type distribution map, to obtain a field crop type distribution map of a target area; The size of the crop planting area is estimated based on the spatial resolution of the Sentinel-1A image and the number of crop pixels in the crop type distribution map of the field.

2. The method for extracting crop planting area according to claim 1, characterized in that: After the remote sensing image data of the target area is acquired, the remote sensing image data of the target area is also spliced ​​and cropped.

3. The method for extracting crop planting area according to claim 1, characterized in that: After acquiring the Sentinel-1A image data of the target area, the method further includes splicing and cropping the Sentinel-1A image data of the target area.

4. The method for extracting crop planting area according to claim 1, characterized in that: In the step of acquiring remote sensing image data of the target area, the remote sensing image data is GF-2 image data; and the preprocessing of the GF-2 image data includes radiation calibration, atmospheric correction and orthorectification.

5. The method for extracting crop planting area according to claim 1, characterized in that: The preprocessing of the Sentinel-1A image data includes thermal noise removal, orbit precision correction, radiation calibration, terrain correction and normalization.

6. The method for extracting crop planting area according to any one of claims 1 to 5, characterized in that: After obtaining the sample data of several crops in the target area, the method also includes generating verification samples of the crops in the target area; after estimating the crop planting area, two classification accuracy evaluation systems, namely error matrix evaluation and area accuracy evaluation, are constructed based on the data of the verification samples.

7. A crop planting area extraction device, characterized in that: include: A remote sensing image data acquisition module is used to acquire remote sensing image data, pre-process the remote sensing image data, and then perform vectorization and rasterization to obtain field raster data of the target area; A crop sample acquisition module is used to acquire sample data of several crops in a field in a target area, mark the types of the crops, and generate training samples of the crops in the field in the target area; A Sentinel-1A image data acquisition module, used to acquire Sentinel-1A image data, perform spatiotemporal filtering on the Sentinel-1A image data after preprocessing, and construct a crop time series backscattering curve based on the crop backscattering time series data; Wherein, the spatiotemporal filtering performed on the preprocessed Sentinel-1A image data includes spatial domain filtering and temporal domain filtering; The spatial domain filtering is to perform spatial field mean filtering on the Sentinel-1A image data of the full-year time series; the field mean filtering is to extract all observable values ​​of the training samples in the Sentinel-1A image data, calculate the average value of all observable values, and obtain the backscatter coefficient value of a single time series of each type of crop; The time domain filtering is to obtain the backscatter coefficient value of a single time series of each type of crop by the spatial domain filtering, and then take the median of the backscatter coefficient value of each type of crop at the same time to obtain the backscatter time series data under the time and space filtering; A crop typical phenological characteristics generating module is used to divide the crop time series backscattering curve into different stages according to the variation law of the crop time series backscattering curve to obtain the crop typical phenological characteristics; A crop type extraction module inputs the backscatter time series data and the typical phenological characteristics of crops into a preset crop classifier, and trains the typical phenological characteristics of crops of the training samples to obtain a crop classification model; and uses the crop classification model to classify the backscatter time series data of the preprocessed Sentinel-1A image data to obtain a crop type distribution map of the target area; A mask processing module, used for performing mask processing on the crop type distribution map using the field raster data to obtain a field crop type distribution map; The crop planting area calculation module is used to estimate the size of the crop planting area according to the spatial resolution of the Sentinel-1A image and the number of crop pixels in the field crop type distribution map.

Images