An automatic remote sensing classification method, device and equipment for drawing vegetation type wetland coverage at a national scale and a medium

Abstract

The application discloses an automatic remote sensing classification method, device and equipment for drawing vegetation type wetland coverage on a national scale and a medium. The method comprises the following steps: constructing a climate-hydrology unit for wetland classification, dividing a country into different climate-hydrology units by using a random forest classifier based on climate and hydrology auxiliary data; selecting appropriate parameters, performing Savitzky-Golay filtering pretreatment on time series data, and establishing complete dense time series NDVI data; establishing a vegetation type wetland classification system, constructing reference NDVI phenology curves of different wetland categories in each climate-hydrology unit based on dense time series NDVI; and classifying vegetation type wetlands in a region to be classified based on a feature matching method in combination with the reference NDVI phenology curves. The application establishes a vegetation type wetland classification method, and meets the demand of large-scale wetland mapping.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image information processing, and is an automatic remote sensing classification method, device, equipment and medium for drawing vegetation type wetland coverage on a national scale. BACKGROUND

[0002] In existing wetland data sets, the classification accuracy of vegetation type wetlands is often lower than that of non-vegetation type wetlands, and is easily confused with non-wetland vegetation. Due to the coverage of vegetation, vegetation type wetlands have obvious seasonal variation characteristics, which is an important reason why they are more difficult to accurately classify than other wetland types. In addition, the development of existing wetland mapping data products does not make full use of the phenological information of wetlands and relies heavily on a large number of samples.

[0003] In land use / coverage classification, the accuracy of wetlands in all land cover categories is the lowest. For example, in GlobeLand30, the mapping accuracy of forest land and farmland is 83.58% and 85.70% respectively, while the mapping accuracy of wetlands is only 74.87%; in GLC_FCS30, the mapping accuracy of farmland, forest land, bare land and water body is higher than 80%, while the classification accuracy of wetlands is only 61.8%. In wetland classification products, the classification accuracy of vegetation type wetlands is also relatively low among all wetland categories. For example, in EA_Wetlands, the mapping accuracy of woody and herbaceous swamps in inland and coastal areas is only about 70%, while the mapping accuracy of rivers and lakes is more than 85%; in GWL_FCS30 products, the classification accuracy of woody, herbaceous swamps and salt marshes is less than 80%, while the classification accuracy of permanent water bodies, salt fields and tidal flats can reach 90%. Most wetland mapping products also have deficiencies in the classification system of vegetation type wetlands, and often ignore important artificial vegetation type wetlands such as rice.

[0004] In addition, existing large-scale wetland mapping work relies heavily on samples. The sample size of CAS_Wetlands is 16496, the sample size of EA_Wetlands is 15808, and the sample size of GWL_FCS30 is even more than 200 million (globally). The production of wetland samples on a large scale is a time-consuming and laborious task, and the inter-annual stability of samples cannot be guaranteed for highly dynamic wetlands, so samples need to be repeatedly selected in multi-year long-term mapping work, which greatly affects the efficiency of updating mapping products. SUMMARY

[0005] In order to solve the above-mentioned deficiencies of the prior art and the need for rapid acquisition of large-scale wetland coverage information for scientific research and management, the purpose of the present application is to provide an automatic remote sensing classification method, device, equipment and medium for drawing vegetation type wetland coverage on a national scale.

[0006] The technical scheme of the present application is: an automatic remote sensing classification method for drawing vegetation type wetland coverage on a national scale, comprising the following steps:

[0007] Step 1, climate-hydrology unit construction for wetland classification, based on climate and hydrology auxiliary data, the country is divided into different climate-hydrology units by using a random forest classifier;

[0008] Step 2, Savitzky-Golay filtering preprocessing is performed on time series data, and complete dense time series NDVI data is established;

[0009] Step 3, a vegetation type wetland classification system is established, based on dense time series NDVI data, a wetland class reference phenology curve of different wetland classes is constructed in each climate-hydrology unit;

[0010] Step 4, based on a feature matching method, the vegetation type wetland of the region to be classified is classified in combination with the wetland class reference phenology curve.

[0011] The present application also proposes an automatic remote sensing classification device for drawing vegetation type wetland coverage on a national scale, comprising:

[0012] A climate-hydrology unit construction module is used for climate-hydrology unit construction for wetland classification, comprising: based on climate and hydrology auxiliary data, the country is divided into different climate-hydrology units by using a random forest classifier;

[0013] A filtering preprocessing module is used for Savitzky-Golay filtering preprocessing on time series data, and complete dense time series NDVI data is established;

[0014] A vegetation type wetland classification system establishment module is used for establishing a vegetation type wetland classification system, comprising: based on dense time series NDVI data, a wetland class reference phenology curve of different wetland classes is constructed in each climate-hydrology unit;

[0015] A vegetation type wetland classification module classifies the vegetation type wetland of the region to be classified in combination with the wetland class reference phenology curve based on a feature matching method.

[0016] The present application also proposes an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned automatic remote sensing classification method for drawing vegetation type wetland coverage on a national scale when executing the program.

[0017] A non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the above-mentioned automatic remote sensing classification method for drawing vegetation type wetland coverage on a national scale.

[0018] The present invention has the following beneficial effects:

[0019] This invention establishes a vegetation-type wetland classification method based on Sentinel-2 remote sensing data, along with climate and hydrological data, to meet the needs of large-scale wetland mapping. Based on the differences in climate and hydrological conditions, the country is divided into primary and secondary climate-hydrological units, ensuring relatively uniform climate and hydrological conditions within each unit. Savitzky-Golay filtering is used to process NDVI time-series data, reducing disturbances caused by external factors, and obtaining reference phenological curves for vegetation-type wetland categories within each climate-hydrological unit. Through feature matching algorithms, rapid classification and mapping of vegetation-type wetlands can be achieved. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention;

[0021] Figure 2 This is an example of the classification results of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0023] This invention discloses an automated remote sensing classification method for mapping vegetated wetland cover at the national scale. This classification method fully utilizes the temporal dimension information of dense time-series remote sensing data, enabling automated mapping of large-scale vegetated wetlands.

[0024] like Figure 1 The diagram shows a flowchart of an automated remote sensing classification method for mapping vegetation-type wetland cover at a national scale, comprising the following steps:

[0025] Step 1: Construction of Climate-Hydrological Units for Wetland Classification. Based on auxiliary data such as climate and hydrology, a random forest classifier is used to divide the country into different climate-hydrological units. Specifically, this includes:

[0026] Step 1.1, collect sample data for dividing climate-hydrological units. Use the national hydrological regionalization map of a certain year as the basis, which contains two levels of hydrological regionalization, first level and second level. A 50km wide buffer zone is established inward at each second level hydrological regionalization boundary, and sample points are randomly generated outside the buffer zone, and each sample point is assigned its first level and second level hydrological regionalization number as a class label. Randomly select 30% of all samples as validation samples and 70% as training samples.

[0027] Step 1.2, process auxiliary data. Collect publicly available national multi-year monthly precipitation, monthly average temperature, monthly potential evapotranspiration and other climate and hydrology related data, and obtain multi-year average temperature, multi-year monthly average temperature and other derived data through methods such as mean synthesis and maximum synthesis.

[0028] Step 1.3, first level climate-hydrological unit division. Use the random forest classifier, use the auxiliary data as classification data, combine the training samples, and divide the country into 11 first level climate-hydrological units.

[0029] Step 1.4, second level climate-hydrological unit division. Within each first level climate-hydrological unit, use the random forest classifier, use the auxiliary data as classification data, combine the training samples, and divide the country into 55 first level climate-hydrological units.

[0030] Step 2, Savitzky-Golay filter preprocessing of time series data:

[0031] Step 2.1, obtain Sentinel-2 images with a time range of December 1 of the previous year to January 31 of the next year, and perform cloud removal and cropping preprocessing operations on each original image. Calculate the original NDVI time series by the formula NDVI=(B8-B4) / (B8+B4), where B4 and B8 are the red and near-infrared bands of Sentinel-2;

[0032] Step 2.2, linear time interpolation of the original NDVI time series to fill in the gaps caused by cloud removal, resulting in a complete NDVI time series data with continuous space-time. The formula for linear time interpolation is:

[0033] ;

[0034] Where y is the missing value to be interpolated, y1 is the data before the missing value, y2 is the data after the missing value, and t, t1, t2 are the corresponding times.

[0035] Step 2.3, Savitzky-Golay filtering is used on the interpolated NDVI time series data, with a filter window size of 25 and a fitting function order of 3.

[0036] Step 3, a vegetation type wetland classification system is established, based on dense time series NDVI, a wetland class reference curve for different wetland classes is constructed in each climate-hydrology unit;

[0037] Step 3.1, construction of the classification system. The vegetation type wetland is divided into herbaceous marsh, woody marsh, salt marsh, mangrove, single-season rice and double-season rice, and the wetland classes contained in each secondary climate-hydrology partition are determined according to existing wetland products;

[0038] Step 3.2, sample point selection. Wetland samples are obtained using multiple existing wetland mapping products, and the samples are screened using the Mahalanobis distance method;

[0039] Step 3.3, construction of the wetland reference phenology curve. The NDVI time series curve at the sample point location is obtained, and the wetland class reference phenology curve is synthesized by averaging the curves of sample points of the same class.

[0040] Step 4, based on the feature matching method, the vegetation type wetland in the region to be classified is classified in combination with the wetland class reference phenology curve.

[0041] Step 4.1, the Euclidean distance between the NDVI time series curve of each pixel to be classified and each wetland class reference phenology curve is calculated, and a raster image with pixel values as Euclidean distance values is output;

[0042] Step 4.2, the histogram of the raster image is calculated, and the Euclidean distance value at the first inflection point on the histogram is set as the threshold value;

[0043] Step 4.3, the threshold value is used to segment the image output in 4.1, and the part less than the threshold value is retained, which is the wetland class to be extracted.

[0044] According to one embodiment of the present application, the vegetation type wetland classification system in a certain region includes woody marsh, herbaceous marsh, single-season rice and double-season rice.

[0045] According to the embodiment of the present application, the classification result is as shown in Figure 2 .

[0046] While the foregoing specific embodiments of the application have been described in some detail to provide a thorough understanding of the application, it should be apparent that the application is not limited to the specifics of the foregoings as these can, of course, vary. As can be seen, the application can be carried out by specifically constructing devices in accordance with the teaching herein or by practicing acts consistent with the principles of this application. For a better understanding of the application, its operating principles and other objects and advantages, reference should be made to the drawings and to the accompanying descriptive matter.

Claims

1. An automated remote sensing classification method for mapping vegetation type wetland cover at national scale, characterized in that, The method comprises the following steps: Step 1, climate-hydrology unit construction for wetland classification, based on climate, hydrology auxiliary data, using random forest classifier to divide the country into different climate-hydrology units; Step 2, Savitzky-Golay filtering preprocessing of time series data, to establish complete dense time series NDVI data; Step 3, establishment of vegetation type wetland classification system, based on dense time series NDVI data, to construct wetland class reference phenology curve of different wetland classes in each climate-hydrology unit; Step 4, based on feature matching method, combined with wetland class reference phenology curve, to classify the vegetation type wetland of the region to be classified; The step 1 specifically comprises: Step 1.1, collecting sample data for dividing climate-hydrology unit; using national hydrology regionalization map as the basis, which contains two levels of first-level and second-level hydrology regionalization, establishing a buffer zone with a certain width in each second-level hydrology regionalization boundary, generating sample points randomly outside the buffer zone, and assigning the first-level and second-level hydrology regionalization numbers of each sample point as the class label, randomly selecting 30% of all samples as validation samples and 70% as training samples; Step 1.2, processing auxiliary data; collecting public national multi-year monthly precipitation, monthly average temperature, monthly potential evapotranspiration and other related data, and obtaining multiple derived data including multi-year average temperature and multi-year monthly average temperature through mean synthesis and maximum synthesis method; Step 1.3, first-level climate-hydrology unit division; using the random forest classifier in Google Earth Engine cloud platform or ENVI, taking the auxiliary data as the classification data, combining with the training samples, and dividing the country into n first-level climate-hydrology units; Step 1.4, second-level climate-hydrology unit division; in each first-level climate-hydrology unit, using the random forest classifier, taking the auxiliary data as the classification data, combining with the training samples, and dividing the country into m second-level climate-hydrology units, wherein m>n; The step 4 specifically comprises: Step 4.1, calculating the Euclidean distance of each pixel to be classified NDVI time series curve and each wetland class reference phenology curve, outputting a grid image with pixel value as Euclidean distance value; Step 4.2, calculating the histogram of the grid image, and setting the Euclidean distance value at the first inflection point on the histogram as the threshold value; Step 4.3, using the threshold value to segment the output grid image, and keeping the part less than the threshold value, which is the wetland class to be extracted.

2. The automated remote sensing classification method of claim 1, wherein, The step 2 specifically comprises: Step 2.1, obtaining Sentinel-2 images with a time range of December 1 of the previous year to January 31 of the next year, performing cloud removal and cropping preprocessing operations on each original image, and calculating the original NDVI time series through the formula NDVI=(B8-B4) / (B8+B4), wherein B4 and B8 are the red and near-infrared bands of Sentinel-2; Step 2.2, linear time interpolation is performed on the original NDVI time series to fill the gaps caused by cloud removal, and a complete NDVI time series data with spatial and temporal continuity is obtained; Step 2.3, Savitzky-Golay filtering is performed on the interpolated NDVI time series data.

3. The automated remote sensing classification method of claim 2, wherein, The step 3 specifically comprises: Step 3.1, construction of classification system; the vegetation type wetland is divided into herbaceous marsh, woody marsh, salt marsh, mangrove, single-season rice and double-season rice, and the wetland categories contained in each secondary climate-hydrology division are determined according to the existing wetland products; Step 3.2, sample point selection; a plurality of existing wetland mapping products are used to obtain wetland samples, and the Mahalanobis distance and P-value between the samples are calculated, and the samples with P-value lower than 0.001 are removed; Step 3.3, construction of wetland category reference phenology curve; the NDVI time series curve at the sample point position is obtained, and the mean value of the curves of the sample points of the same category is synthesized to obtain the wetland category reference phenology curve.

4. The automated remote sensing classification method of claim 1, wherein, The mean value synthesis and maximum value synthesis method is to statistically process the data of a plurality of images at each pixel position, and to generate a new image by taking the mean value and the maximum value.

5. The automated remote sensing classification method of claim 1, wherein, The formula of linear time interpolation is: ； wherein, is a missing value that needs to be interpolated, is data before the missing value, is data after the missing value, , , is the corresponding time.

6. The automated remote sensing classification method of claim 2, wherein, In the Savitzky-Golay filtering, the filter window size is set to 25, and the fitting function order is set to 3.

7. An apparatus for automated remote sensing classification of vegetation wetland cover at national scale, implementing a method for automated remote sensing classification of vegetation wetland cover at national scale according to any one of claims 1 to 6, characterized in that, It comprises: A climate-hydrology unit construction module for constructing climate-hydrology units for wetland classification, comprising: based on climate and hydrology auxiliary data, using a random forest classifier to divide the country into different climate-hydrology units; A filtering preprocessing module for performing Savitzky-Golay filtering preprocessing on time series data to establish complete dense time series NDVI data; A vegetation type wetland classification system establishment module for establishing a vegetation type wetland classification system, comprising: based on the dense time series NDVI data, constructing wetland category reference phenology curves of different wetland categories in each climate-hydrology unit; A vegetation type wetland classification module based on feature matching method, combining the wetland category reference phenology curve to classify the vegetation type wetland of the region to be classified.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to realize the steps of the automatic remote sensing classification method for mapping vegetation type wetland coverage at national scale according to any one of claims 1-6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the automatic remote sensing classification method for mapping vegetation type wetland coverage at national scale according to any one of claims 1-6.

Images