1. The method for extracting the forest swamp based on the Sentinel satellite image and the random forest algorithm is characterized by comprising the following steps of:

screening and obtaining Sentinel-1GRD and Sentinel-2MSI images with less clouds in a research area year by using an official website of the European Bureau, and performing terrain correction, coherent spot filtering correction, radiation correction and high-resolution fusion pretreatment on the downloaded Sentinel images by using SNAP (personal network access protocol) Sentinel official processing software;

step two, acquiring Sentinel-1VV and VH polarization backscattering coefficients to obtain a VV and VH backscattering coefficient image on the space;

step three, calculating the red edge index, vegetation index and water body index characteristics of the Sentinel-2 image, wherein the calculation formula is as follows:

NDVIR1＝(B8a-B5)/(B8a+B5) (1)

NDVIR2＝(B8a-B6)/(B8a+B6) (2)

NDVIR3＝(B8a-B7)/(B8a+B7) (3)

NDR1＝(B6-B5)/(B6+B5) (4)

NDR2＝(B7-B5)/(B7+B5) (5)

NDVI＝(B8a-B4)/(B8a+B4) (6)

NDWI＝(B3-B8a)/(B3+B8a) (7)

in the formulas (1), (2), (3), (4), (5), (6) and (7), Bx represents the reflectivity of 1-12 wave bands in Sentinel-2, wherein x is 3, 4, 5, 6, 7 or 8 a;

integrating the Sentinel-1VV and VH backscattering coefficient and the index characteristics of the Sentinel-2 red edge index, vegetation index and water body index by utilizing remote sensing image space data fusion to obtain the reflectivity information of the Sentinel-1 radar and the Sentinel-2 red edge multispectral wave band fusion characteristics in the research area;

establishing a training sample and a verification point model for recognizing and extracting the type information of the forest swamp ground objects in the research area by utilizing the field actual measurement investigation data;

step six, recognizing and extracting the type information of the whole forest swamp ground objects in the research area on the training samples established in the step five by using a random forest algorithm model, acquiring the spatial distribution range of the forest swamps in the research area, and performing precision verification on the extraction result by using the verification points established in the step five;

and seventhly, identifying and extracting results according to the type information of the forest swamp ground objects in the step six, and cutting the forest swamp area in the research area to obtain the distribution range information of the forest swamps.

Background

The wetland is one of the most important land ecosystem types, and plays an extremely important role in maintaining the material circulation balance, protecting the diversity of biological species, maintaining ecological safety and the like. The forest swamp is a swamp wetland which takes woody plants with more than six meters as dominant species, has extremely rich carbon reserves and has obvious response to global climate change. Facing double severe challenges of human activities and climate change, the remote sensing monitoring of the forest swamp can timely determine the distribution boundary of the forest swamp, reflect the landscape current situation and the transformation trend of the forest swamp wetland, and has important significance for effectively protecting the health of the forest swamp ecosystem and maintaining the stability of the regional ecosystem.

In recent years, the effective extraction of the ground object type information by fusing different types of sensor remote sensing data to the maximum extent based on multi-source remote sensing data integration has become an effective method for recognizing and extracting the type of the forest swamp ground object. But the forest swamp information extraction and drawing are difficult due to poor perspective, complex species types, fussy visual interpretation and field investigation and the like. The single spectral image is used, and the range of the forest swamp below the tall and big branches of the forest cannot be effectively penetrated and identified.

Disclosure of Invention

The invention aims to solve the problem that the existing single spectrum remote sensing data cannot be accurately detected to obtain the distribution information of the forest swamps, and provides a method for extracting the forest swamps based on a Sentinel satellite image and a random forest algorithm.

The method for extracting the forest swamp based on the Sentinel satellite image and the random forest algorithm comprises the following steps:

screening and obtaining Sentinel-1GRD and Sentinel-2MSI images with less clouds in a research area year by using an official website of the European Bureau, and performing terrain correction, coherent spot filtering correction, radiation correction and high-resolution fusion pretreatment on the downloaded Sentinel images by using SNAP (personal network access protocol) Sentinel official processing software;

step two, acquiring Sentinel-1VV and VH polarization backscattering coefficients to obtain a VV and VH backscattering coefficient image on the space;

step three, calculating the red edge index, vegetation index and water body index characteristics of the Sentinel-2 image, wherein the calculation formula is as follows:

NDVIR1＝(B8a-B5)/(B8a+B5) (1)

NDVIR2＝(B8a-B6)/(B8a+B6) (2)

NDVIR3＝(B8a-B7)/(B8a+B7) (3)

NDR1＝(B6-B5)/(B6+B5) (4)

NDR2＝(B7-B5)/(B7+B5) (5)

NDVI＝(B8a-B4)/(B8a+B4) (6)

NDWI＝(B3-B8a)/(B3+B8a) (7)

in the formulas (1), (2), (3), (4), (5), (6) and (7), Bx represents the reflectivity of 1-12 wave bands in Sentinel-2, wherein x is 3, 4, 5, 6, 7 or 8 a;

integrating the Sentinel-1VV and VH backscattering coefficient and the index characteristics of the Sentinel-2 red edge index, vegetation index and water body index by utilizing remote sensing image space data fusion to obtain the reflectivity information of the Sentinel-1 radar and the Sentinel-2 red edge multispectral wave band fusion characteristics in the research area;

establishing a training sample and a verification point model for recognizing and extracting the type information of the forest swamp ground objects in the research area by utilizing the field actual measurement investigation data;

step six, recognizing and extracting the type information of the whole forest swamp ground objects in the research area on the training samples established in the step five by using a random forest algorithm model, acquiring the spatial distribution range of the forest swamps in the research area, and performing precision verification on the extraction result by using the verification points established in the step five;

and seventhly, identifying and extracting results according to the type information of the forest swamp ground objects in the step six, and cutting the forest swamp area in the research area to obtain the distribution range information of the forest swamps.

The invention realizes the recognition and spatial distribution of the forest swamp under the condition that the high and big branches cover the water body by accurate detection of the remote sensing technology. The method comprises the steps of firstly, respectively carrying out a series of preprocessing of terrain correction, speckle filtering correction, radiation correction and high-resolution fusion on obtained Sentinel-1 and Sentinel-2 remote sensing images. And then index characteristics such as Sentinel-1VV and VH backscattering coefficients, Sentinel-2 red edge indexes, vegetation indexes, water body indexes and the like are respectively obtained, and Sentinel-1 radar and Sentinel-2 red edge multispectral wave band fusion characteristic data are obtained by utilizing multi-source remote sensing data space fusion. And finally, establishing a training sample and a verification point model of the type of the forest swamp ground object in the research area based on the data of the field actual measurement investigation sampling point, and identifying and extracting the type information of the forest swamp ground object from the multisource fusion remote sensing data in the research area by using a random forest algorithm to obtain the spatial distribution range of the forest swamp ground object in the research area. Through inspection, the forest swamp information is identified based on fusion of Sentinel-1 radar and Sentinel-2 multispectral data and a random forest algorithm, and the extraction precision of the forest swamp information can reach 90.5%. The invention overcomes the difficulty brought by remote sensing interpretation of the forest swamp by branches and trunks of tall trees in the forest swamp, and solves the problem that the forest swamp of the sheltered water body is wrongly divided into forest lands. The Sentinel-1 radar and the Sentinel-2 multispectral wave band feature fusion established by the invention is quick and effective for extracting the shielded forest swamp, improves the classification precision and the credibility of the remote sensing interpretation of the forest swamp, has repeatability and scientificity, and has extremely important significance for the remote sensing mapping of the forest swamp. The method for extracting the forest marsh overcomes the problem of remote sensing interpretation of the forest marsh.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a diagram of the Sentinel-1 radar VH backscattering coefficient and the Sentinel-2 true color in the first experiment.

Detailed Description

The technical solution of the present invention is not limited to the following specific embodiments, but includes any combination of the specific embodiments.

The first embodiment is as follows: the method for extracting the forest swamp based on the Sentinel satellite image and the random forest algorithm comprises the following steps:

screening and obtaining Sentinel-1GRD and Sentinel-2MSI images with less clouds in a research area year by using an official website of the European Bureau, and performing terrain correction, coherent spot filtering correction, radiation correction and high-resolution fusion pretreatment on the downloaded Sentinel images by using SNAP (personal network access protocol) Sentinel official processing software;

step two, acquiring Sentinel-1VV and VH polarization backscattering coefficients to obtain a VV and VH backscattering coefficient image on the space;

step three, calculating the red edge index, vegetation index and water body index characteristics of the Sentinel-2 image, wherein the calculation formula is as follows:

NDVIR1＝(B8a-B5)/(B8a+B5) (1)

NDVIR2＝(B8a-B6)/(B8a+B6) (2)

NDVIR3＝(B8a-B7)/(B8a+B7) (3)

NDR1＝(B6-B5)/(B6+B5) (4)

NDR2＝(B7-B5)/(B7+B5) (5)

NDVI＝(B8a-B4)/(B8a+B4) (6)

NDWI＝(B3-B8a)/(B3+B8a) (7)

in the formulas (1), (2), (3), (4), (5), (6) and (7), Bx represents the reflectivity of 1-12 wave bands in Sentinel-2, wherein x is 3, 4, 5, 6, 7 or 8 a;

integrating the Sentinel-1VV and VH backscattering coefficient and the index characteristics of the Sentinel-2 red edge index, vegetation index and water body index by utilizing remote sensing image space data fusion to obtain the reflectivity information of the Sentinel-1 radar and the Sentinel-2 red edge multispectral wave band fusion characteristics in the research area;

establishing a training sample and a verification point model for recognizing and extracting the type information of the forest swamp ground objects in the research area by utilizing the field actual measurement investigation data;

step six, recognizing and extracting the type information of the whole forest swamp ground objects in the research area on the training samples established in the step five by using a random forest algorithm model, acquiring the spatial distribution range of the forest swamps in the research area, and performing precision verification on the extraction result by using the verification points established in the step five;

and seventhly, identifying and extracting results according to the type information of the forest swamp ground objects in the step six, and cutting the forest swamp area in the research area to obtain the distribution range information of the forest swamps.

The following experiments are adopted to verify the effect of the invention:

experiment one:

the method for extracting the forest swamp based on the Sentinel satellite image and the random forest algorithm comprises the following steps:

the method comprises the following steps: by utilizing the official website of the European and aviation administration, a Sentinel-1GRD image with less cloud amount in the annual significant wetland of Hani international wetland is screened and downloaded (S1B _ IW _ GRDH _1SDV _20200713T213839_20200713T213904_022458_02A9FB _05D9), the acquisition instrument is SAR-C, the data size is about 1.64G, and the acquisition time is 2020, 7, month and 13 days. Screening and downloading a Sentinel-2MSI image (L1C _ T51TYG _ A025783_20200530T023518) of the significant wetland of Hani, wherein the number of a track of a map is 46, the projection is UTM 51N, the cloud amount is 0.0869%, the image size is about 754.87M, and the acquisition time is 7 months and 9 days in 2020. Carrying out terrain correction, speckle filtering correction, radiation correction and high-resolution fusion pretreatment on the downloaded Sentinel image by using SNAP Sentinel official processing software;

step two: obtaining Sentinel-1VV and VH polarization backscattering coefficients to obtain a VV and VH backscattering coefficient image on the space;

step three: and calculating index characteristics such as a red edge index, a vegetation index, a water body index and the like of the Sentinel-2 image. The calculation formula is as follows:

NDVIR1＝(B8a-B5)/(B8a+B5) (1)

NDVIR2＝(B8a-B6)/(B8a+B6) (2)

NDVIR3＝(B8a-B7)/(B8a+B7) (3)

NDR1＝(B6-B5)/(B6+B5) (4)

NDR2＝(B7-B5)/(B7+B5) (5)

NDVI＝(B8a-B4)/(B8a+B4) (6)

NDWI＝(B3-B8a)/(B3+B8a) (7)

wherein Bx represents the reflectivity of 1-12 wave bands in Sentinel-2, wherein x is 3, 4, 5, 6, 7 or 8 a.

Step four: integrating the Sentinel-1VV and VH backscattering coefficient and the index characteristics of the Sentinel-2 red edge index, the vegetation index and the water body index by utilizing remote sensing image space data fusion to obtain the fusion characteristic reflectivity information of the Sentinel-1 and Sentinel-2 red edge multispectral wave band in the research area;

step five: establishing a Harni international important wetland forest swamp ground object type information identification extraction training sample and a verification point model by utilizing field actual measurement investigation data;

step six: identifying and extracting the integral forest swamp information of the Hani international important wetland by using a random forest algorithm model, carrying out precision verification on the extracted result by using verification points established in the fifth step, and identifying the forest swamp information by checking and fusing Sentinel-1 radar and Sentinel-2 multispectral data and a random forest algorithm, wherein the extraction precision of the forest swamp information can reach 90.5%;

step seven: and identifying and extracting results according to the type information of the six forest swamp ground objects, and cutting the forest swamp area in the Hani international important wetland to obtain the distribution range information of the forest swamps.