A sea ice classification method based on multispectral and SAR fused remote sensing data

Abstract

The application discloses a sea ice classification method based on multispectral and SAR fusion remote sensing data and belongs to the technical field of satellite remote sensing image processing.The sea ice classification method based on multispectral and SAR fusion remote sensing data provided by the application solves the technical problems that most of the existing sea ice classification research based on remote sensing adopts a single data source, the classification precision is limited, how to comprehensively use the rich optical characteristics of multispectrum and the advantages of SAR in detecting ground features and being less affected by weather to improve the sea ice classification precision, and the like, and adopts Sentinel-1 SAR and Sentinel-2 multispectral data to realize the target of quickly and accurately obtaining large-range sea ice classification results based on multispectral and SAR fusion remote sensing data. The method has strong applicability, can be used for sea ice monitoring, and is also helpful for climate change research, polar resource development, polar scientific exploration and navigation of the Arctic Channel.

Description

TECHNICAL FIELD

[0001] The application relates to a sea ice classification method based on multispectral and SAR fusion remote sensing data, and belongs to the technical field of satellite remote sensing image processing. BACKGROUND

[0002] Large-scale and accurate monitoring of sea ice is of great significance for climate change research, polar resource development, polar scientific research and navigation of the Arctic shipping route. Remote sensing images have the ability of large-area, long-time and rapid acquisition of surface information, and are the main data for large-scale sea ice monitoring. However, at present, most of the researches on large-scale sea ice classification based on remote sensing adopt a single data source, and the classification accuracy is limited. SUMMARY

[0003] The technical problem solved by the application is to overcome the shortcomings of the prior art and provide a sea ice classification method based on multispectral and SAR fusion remote sensing data, which realizes accurate extraction of different sea ice types based on the spectral and texture information required for sea ice classification.

[0004] The technical solution of the application is a sea ice classification method based on multispectral and SAR fusion remote sensing data, comprising:

[0005] Obtain all satellite data of a concerned area in a year in which a research time is located, and global low-resolution sea ice drift product data;

[0006] Preprocess the satellite data of the concerned area and the sea ice drift data;

[0007] Based on the preprocessed data, calculate and generate original spectrum, water index, texture index, snow index, backscattering coefficient, backscattering coefficient ratio, sea ice drift time series data, and extract monthly quantile features;

[0008] Establish a sea ice classification system according to the characteristics of the research area, and generate monthly training and verification sample sets;

[0009] Generate monthly sea ice classification probability maps according to the monthly training and verification sample sets and the quantile features;

[0010] Change monitoring is performed according to the backscattering coefficient ratio time series data to obtain sea ice data time series segmentation results;

[0011] According to the sea ice data time series segmentation results and the monthly sea ice classification probability maps, a final sea ice classification result map is obtained.

[0012] Further, the satellite data of the concerned area comprises Sentinel-1 SAR GRD data and Sentinel-2 Level-1C data.

[0013] The preprocessing includes data cropping and splicing, band extraction, pixel screening and resampling;

[0014] The band extraction includes:

[0015] The VV and VH bands of Sentinel-1 SAR GRD data, and the blue band, green band B3, red band B4, red edge band B5, red edge band B6, red edge band B7, near-infrared band B8, red edge band B8A, far-infrared band B11 and far-infrared band B12 bands of Sentinel-2 Level-1C data.

[0016] Further, the pixel screening includes:

[0017] According to the quality control band QA60 file description of Sentinel-2 Level-1C data, the values corresponding to clouds and cirrus are found;

[0018] All pixel values of the QA60 band are traversed, the pixels corresponding to the cloud and cirrus values are identified, and all band values of the pixels are set to null.

[0019] Further, the calculation and generation of original spectrum, water index, texture index, snow index, backscattering coefficient, backscattering coefficient ratio, sea ice drift time series data include:

[0020] According to the B2 band, B3 band, B4 band, B5 band, B6 band, B7 band, B8 band, B8A band, B11 band and B12 band of Sentinel-2 Level-1C data, the water index and snow index are calculated;

[0021] According to the VV, VH and ratio data of Sentinel-1 SAR GRD data, and the B2 band, B3 band, B4 band, B5 band, B6 band, B7 band, B8 band, B8A band, B11 band, B12 band, water index and snow index data of Sentinel-2 Level-1C data, the texture index is calculated;

[0022] The backscattering VV and VH and ratio data, the original spectrum data B2 band, B3 band, B4 band, B5 band, B6 band, B7 band, B8 band, B8A band, B11 band, B12 band, the generated water index, snow index and texture index and sea ice drift data are grouped according to months;

[0023] Each monthly data of each band is superimposed to form monthly time series data.

[0024] Further, the extraction of monthly quantile features includes:

[0025] Sort each pixel in each waveband monthly time series data in ascending order;

[0026] According to the sorting result, the minimum value, median and maximum quantile features are extracted.

[0027] Further, the change monitoring according to the backscattering coefficient ratio time series data to obtain the sea ice data time series segmentation result comprises: according to the backscattering coefficient ratio time series data, using a quantile stepwise regression model for change monitoring to obtain the time series segmentation result.

[0028] Further, the change monitoring according to the backscattering coefficient ratio time series data, using a quantile stepwise regression model to obtain the time series segmentation result, comprises:

[0029] Taking the backscattering ratio time series data of each pixel as input data, using a stepwise regression quantile regression model to obtain the ground feature change time point;

[0030] According to the ground feature change time point, the backscattering ratio time series of each pixel is divided into different segments.

[0031] Further, the generation of the monthly sea ice classification probability map according to the monthly training sample set and the quantile data comprises:

[0032] Extracting the backscattering VV and VH and ratio data of the training sample points, the original spectral data B2 band, B3 band, B4 band, B5 band, B6 band, B7 band, B8 band, B8A band, B11 band, B12 band, the generated water body index, snow index and texture index and sea ice drift data monthly quantile features;

[0033] Inputting the training sample data into a random forest machine learning method to generate a monthly sea ice classification probability map.

[0034] Further, the generation of the final sea ice classification result map according to the sea ice data time series segmentation result and the monthly sea ice classification probability map comprises:

[0035] Traversing the time series segmentation result and the monthly classification probability result of each pixel;

[0036] Weighted average of the probability of each class in each time series for each pixel;

[0037] The class with the maximum weighted average probability is the final class of the pixel, forming a final 10m sea ice classification result map.

[0038] Further, the water body index includes an automatic water body calculation index, a normalized water body index and an improved normalized water body index; the texture index includes variance, homogeneity and angular second moment; and the snow index includes a normalized principal component snow index.

[0039] Compared with the prior art, the present application has the following advantages:

[0040] The present application provides a sea ice classification method based on multispectral and SAR fusion remote sensing data, which solves the technical problems of limited classification accuracy of the prior art sea ice classification method based on remote sensing, and how to improve the sea ice classification accuracy by comprehensively using the rich optical characteristics of multispectral data and the advantages of SAR in detecting ground features and being less affected by weather. BRIEF DESCRIPTION OF DRAWINGS

[0041] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not intended to limit the scope of the present application. Moreover, like reference numerals designate like elements throughout the several views. In the drawings:

[0042] Fig. 1 is a sea ice classification flowchart in the present application.

[0043] Fig. 2 is a flowchart of monthly quantile feature extraction in the present application.

[0044] Fig. 3 is a flowchart of generating a final sea ice classification result based on a time series segmentation result and a monthly sea ice classification probability map in the present application. DETAILED DESCRIPTION

[0045] In order to better understand the above technical solutions, the following will make a detailed description of the technical solutions of the present application through the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solutions of the present application, but not limitations of the technical solutions of the present application. In the case of no conflict, the technical features in the embodiments and the embodiments can be combined with each other.

[0046] The sea ice classification method based on multispectral and SAR fusion remote sensing data provided by the embodiments of the present application will be further described in detail below in combination with the accompanying drawings of the specification. The specific implementation manner can include (for example,Figs. 1-3

[0047] Step 1: Obtain all Sentinel-1 SAR GRD data, Sentinel-2 Level-1C data, and global low-resolution sea ice drift product data of the study area in the year of the study time;

[0048] Step 2: Preprocess the Sentinel-1 SAR GRD, Sentinel-2 Level-1C, and sea ice drift data;

[0049] Step 3: Based on the preprocessed data, calculate and generate original spectrum, water index, texture index, snow index, backscattering coefficient, backscattering coefficient ratio, sea ice drift time series data, and extract monthly quantile features;

[0050] Step 4: Based on the data generated in Step 3, establish a sea ice classification system according to the characteristics of the study area, and generate monthly training and validation sample sets;

[0051] Step 5: Based on the monthly training sample set and quantile data, generate monthly sea ice classification probability maps using a random forest model;

[0052] Step 6: Based on the backscattering coefficient ratio time series data, use a quantile stepwise regression model for change detection to obtain time series segmentation results;

[0053] Step 7: Based on the time series segmentation results and monthly sea ice classification probability maps, obtain the final 10m sea ice classification result map.

[0054] Further, the preprocessing of the Sentinel-1 SAR GRD, Sentinel-2 Level-1C, and sea ice drift data includes data cropping and splicing, band extraction, pixel filtering, and resampling.

[0055] Further, in one possible implementation, the band extraction includes the VV and VH bands of Sentinel-1 SAR GRD, and the blue band (B2), green band (B3), red band (B4), red edge band 1 (B5), red edge band 2 (B6), red edge band 3 (B7), near-infrared band (B8), red edge band 4 (B8A), far-infrared band 1 (B11), and far-infrared band 2 (B12) bands of Sentinel-2 Level-1C.

[0056] In one possible implementation, the pixel filtering includes the following steps:

[0057] ​(1) According to the quality control band QA60 file description of Sentinel-2 Level-1C, find the corresponding value of cloud and cirrus;

[0058] (2) Traverse all pixel values of QA60 band, identify the pixel corresponding to the value of cloud and cirrus, and set all band values of the pixel to null.

[0059] In a possible implementation manner, the water body index includes an automatic water body calculation index (AWEI_nsh, AWEI_sh), a normalized water body index (NDWI), and a modified normalized water body index (MNDWI).

[0060] In a possible implementation manner, the texture index includes variance (Variance), homogeneity (Homogeneity), and angular second moment (ASM).

[0061] Further, in a possible implementation manner, the snow index includes a normalized principal component snow index (NDPCSI).

[0062] In a possible implementation manner, the calculation of the backscattering ratio value data includes the following steps:

[0063] (1) Read the VV and VH band data of Sentinel-1 SAR GRD;

[0064] (2) Calculate the ratio of the VV and VH values of each pixel.

[0065] Further, the calculation and generation of the original spectrum, the water body index, the texture index, the snow index, the backscattering coefficient, the backscattering coefficient ratio, and the sea ice drift time series data include:

[0066] (1) According to the B2, B3, B4, B5, B6, B7, B8, B8A, B11, and B12 bands of Sentinel-2 Level-1C, calculate the water body index and the snow index;

[0067] (2) According to the VV, VH, and ratio data of Sentinel-1 SAR GRD, and the B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12, water body index, and snow index data of Sentinel-2 Level-1C, calculate the texture index;

[0068] (3) Group the backscatter (VV and VH) and ratio data, raw spectral data (B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12), generated water body index, snow index and texture index, and sea ice drift data by month;

[0069] (4) Stack each of the above monthly data together to form monthly time series data.

[0070] In a possible implementation manner, the extracting the monthly quantile features comprises the following steps:

[0071] (1) Sort each pixel in the monthly time series data of each band in ascending order;

[0072] (2) According to the sorting result, extract the minimum value, median value and maximum value quantile features.

[0073] In a possible implementation manner, the monitoring the change by using the quantile stepwise regression model according to the backscatter coefficient ratio time series data to obtain time series segmentation results comprises the following steps:

[0074] (1) Taking the backscatter ratio time series data of each pixel in a year as input data, a stepwise regression quantile regression model is used to obtain a time point of change of a ground object;

[0075] (2) According to the time point of change of the ground object, the backscatter ratio time series of each pixel is divided into different segments.

[0076] Further, in a possible implementation manner, the generating monthly sea ice classification probability maps by using a random forest model according to the monthly training sample set and the quantile data comprises:

[0077] (1) Extracting the backscatter (VV and VH) and ratio data, raw spectral data (B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12), generated water body index, snow index and texture index, and sea ice drift data monthly quantile features of the training sample points;

[0078] (2) Inputting the training sample data obtained in the above step into a random forest machine learning method to generate monthly sea ice classification probability maps.

[0079] In a possible implementation manner, the obtaining the final 10m sea ice classification result map according to the time series segmentation result and the monthly sea ice classification probability map comprises:

[0080] (1) Iterating the time series segmentation result and the monthly classification probability result of each pixel;

[0081] (2) Weighted average of the probability of each class in each time series for each pixel;

[0082] (3) The class with the maximum weighted average probability is the final class of the pixel, forming the final 10m sea ice classification result map.

[0083] In the scheme provided in the embodiments of the present application, specifically includes:

[0084] Step 1: Obtain all Sentinel-1 SAR GRD data, Sentinel-2 Level-1C data of the area of interest in the year of the study time, and global low-resolution sea ice drift product data, the specific steps are as follows:

[0085] (1) Based on ArcGIS software, according to the coastline vector file of the northeast of Greenland, a 30km buffer area is generated as the study area;

[0086] (2) Import the study area vector file into the data download page to obtain all Sentinel-2 Level-1C data, Sentinel-1 SAR GRD data and global low-resolution sea ice drift product data in the study area in 2022;

[0087] Step 2: Preprocess the Sentinel-1 SAR GRD, Sentinel-2 Level-1C and sea ice drift data, the specific steps are as follows:

[0088] (1) Load the study area data, Sentinel-2 Level-1C, Sentinel-1 SAR GRD and sea ice drift data into ENVI software for clipping and splicing;

[0089] (2) Extract the VV and VH bands of Sentinel-1 SAR GRD, and the blue band (B2), green band (B3), red band (B4), red edge band 1 (B5), red edge band 2 (B6), red edge band 3 (B7), near-infrared band (B8), red edge band 4 (B8A), far-infrared band 1 (B11) and far-infrared band 2 (B12) bands of Sentinel-2 Level-1C;

[0090] (3) According to the quality control band QA60 file specification of Sentinel-2 Level-1C, find the values corresponding to clouds and cirrus;

[0091] (4) Traverse all pixel values of QA60 band, identify the pixel corresponding to cloud and cirrus values, and set all band (B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12) values of the pixel to null;

[0092] Step 3: Based on the pre-processed data, calculate and generate original spectrum, water body index, texture index, snow index, backscattering coefficient, backscattering coefficient ratio, sea ice drift time series data, and extract monthly quantile features, the specific steps are as follows:

[0093] (1) Import the pre-processed Sentinel-2 data into ENVI, calculate AWEI nsh , AWEI sh , NDWI and MNDWI, etc. water index, and NDPCSI, etc. snow index, the calculation formula is as follows:

[0094] AWEI nsh = 4 × (ρ Green - ρ SWIR1 ) - (0.25 × ρ NIR + 2.75 × ρ SWIR2 ) (Formula 1)

[0095] AWEI sh = ρ blue + 2.5 × ρ Green - 1.5 × (ρ NIR + ρ SWIR1 )- 0.25 × ρ SWIR2 (Formula 2)

[0096]

[0097]

[0098]

[0099] Wherein, ρ blue , ρ Green , ρ NIR , ρ SWIR1 and ρ SWIR2 are the reflectivity of Sentinel-2 blue band (B2), green band (B3), near-infrared band (B8), far-infrared band 1 (B11) and far-infrared band 2 (B12) band, PC brightest and PC darkest are the principal component 1 and principal component 2 analyzed by using B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12 band reflectivity of Sentinel-2;

[0100] (2) Read the VV and VH band data of Sentinel-1 SAR GRD, and calculate the ratio of VV and VH values of each pixel;

[0101] (3) Calculate the B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12 bands of Sentinel-2, AWEI nsh , AWEI sh , NDWI and MNDWI water index, snow index NDPCSI, Sentinel-1 VV, VH band and ratio data for gray level co-occurrence matrix (GLCM) to represent the probability of two pixels appearing simultaneously, to calculate three texture indexes of variance, homogeneity and angular second moment (ASM), the calculation formula is as follows:

[0102]

[0103]

[0104]

[0105] Where, p(i,j) represents the relative frequency of the given adjacent two pixels, the intensity value of one pixel is i, and the intensity value of the other pixel is j; p d,θ (i,j) represents the second-order statistical probability value of two pixels with gray values i and j at a specific offset distance d and a specific angle θ; μ is the reference pixel gray value, N g is the window size for calculating texture;

[0106] (4) Group all the bands of Sentinel-2, the generated water index and snow index data, the two backscattering bands and ratio data of Sentinel-1, and the generated texture index data and sea ice drift data according to months;

[0107] (5) Sort each pixel in each monthly time series data of each band in ascending order;

[0108] (6) According to the sorting results, extract the minimum value, median and maximum value quantile features;

[0109] Step 4: Based on the data generated in step 3, establish a sea ice classification system according to the characteristics of the study area, and generate monthly training and validation sample sets, the specific steps are as follows:

[0110] (1) Establish a sea ice classification system for the northeast of Greenland, including fixed ice, nascent ice, one-year ice and multi-year ice;

[0111] (2) According to the interpretation signs of different categories of sea ice, select the monthly training and verification sample set;

[0112] Step 5: According to the monthly training sample set and the quantile data, generate the monthly sea ice classification probability map by using the random forest model, and the specific steps are as follows:

[0113] (1) Extract all the bands of Sentinel-2, the generated water index and snow index data of all the monthly training samples of all sea ice types, the two backscattering bands and the ratio data of Sentinel-1, the generated texture index data and the monthly quantile features of sea ice drift;

[0114] (2) Train the random forest machine learning model with the training sample data features obtained in (1), input all the monthly feature values of all bands of all images as input data, and obtain the monthly sea ice classification probability map;

[0115] Step 6: According to the backscattering coefficient ratio time series data, use the quantile stepwise regression model for change monitoring to obtain the time series segmentation result, and the specific steps are as follows:

[0116] (1) Extract the backscattering ratio time series data of each pixel;

[0117] (2) Calculate the stepwise regression result of the median of the backscattering ratio time series of the pixel by using the stepwise regression function;

[0118] (3) Obtain the time series segmentation segment according to the sea ice type change time point obtained by stepwise regression;

[0119] Step 7: According to the time series segmentation result and the monthly sea ice classification probability map, obtain the final 10m sea ice classification result map, and the specific steps are as follows:

[0120] (1) Traverse the time series segmentation result and the monthly classification probability result of each pixel;

[0121] (2) Weighted average the probability of each category in each time series of each pixel;

[0122] (3) The category with the maximum weighted average probability is the final category of the pixel, and the final 10m sea ice classification result map is formed.

[0123] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

[0124] The contents not described in detail in the specification of the present application are the known technology of those skilled in the art.

Claims

1. A sea ice classification method based on multispectral and SAR fusion remote sensing data, characterized in that, include: Acquire all satellite data of interest for the region of interest in the year in which the study was conducted, as well as global low-resolution sea ice drift product data; Preprocessing of satellite data and sea ice drift data of interest; Based on the preprocessed data, the original spectrum, water index, texture index, snow index, backscattering coefficient, backscattering coefficient ratio, and sea ice drift time series data are calculated and generated, and monthly quantile features are extracted. A sea ice classification system was established based on the characteristics of the study area, and monthly training and validation sample sets were generated. Based on the monthly training and validation sample sets and quantile features, a monthly sea ice classification probability map is generated. Changes in the backscattering coefficient ratio time series data were monitored to obtain the time series segmentation results of sea ice data. Based on the time series segmentation results of sea ice data and the monthly sea ice classification probability map, the final sea ice classification result map is obtained. The calculation and generation of raw spectra, water index, texture index, snow index, backscattering coefficient, backscattering coefficient ratio, and sea ice drift time series data include: The water index and snow index are calculated based on the blue light bands B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12 of Sentinel-2 Level-1C data. The texture index is calculated based on the VV, VH and ratio data of Sentinel-1 SAR GRD data, and the B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12 bands, water index and snow index data of Sentinel-2 Level-1C data. The backscattered VV, VH and ratio data, the original spectral data in bands B2, B3, B4, B5, B6, B7, B8, B8A, B11, and B12, the generated water index, snow index, texture index, and sea ice drift data are grouped by month. The monthly data for each of the above bands are superimposed to form monthly time series data.

2. The sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 1, characterized in that, The satellite data of interest includes Sentinel-1 SAR GRD data and Sentinel-2 Level-1C data; The preprocessing includes data cropping and stitching, band extraction, pixel filtering, and resampling; The band extraction includes: The VV and VH bands of Sentinel-1 SAR GRD data, and the blue B2 band, green B3 band, red B4 band, red edge band B5 band, red edge band B6 band, red edge band B7 band, near-infrared band B8 band, red edge band B8A band, far-infrared band B11 band, and far-infrared band B12 band of Sentinel-2 Level-1C data.

3. The sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 2, characterized in that, The pixel filtering includes: Based on the QA60 document description of the quality control bands of Sentinel-2 Level-1C data, find the corresponding values ​​for clouds and cirrus clouds; Iterate through all pixel values ​​in the QA60 band, identify the pixels corresponding to cloud and cirrus cloud values, and set all band values ​​of the pixels to null.

4. The sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 2, characterized in that, The extraction of monthly quantile features includes: Sort each cell in the monthly time series data of each band in ascending order; Based on the sorting results, extract the minimum, median, and maximum quantile features.

5. The sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 2, characterized in that, The step of monitoring changes based on the backscattering coefficient ratio time series data to obtain the sea ice data time series segmentation results includes: using a quantile stepwise regression model to monitor changes based on the backscattering coefficient ratio time series data to obtain the time series segmentation results.

6. The sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 5, characterized in that, The process involves monitoring changes in the backscattering coefficient ratio time series data using a quantile stepwise regression model to obtain time series segmentation results, including: Using the annual time series data of the backscattering ratio of each pixel as input data, a stepwise regression quantile regression model is adopted to obtain the time points of ground cover changes; Based on the time points of ground feature changes, the backscattering ratio time series of each pixel is divided into different segments.

7. The sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 2, characterized in that, The step of generating a monthly sea ice classification probability map based on the monthly training sample set and quantile data includes: Backscattered VV, VH and ratio data of training sample points were extracted, along with original spectral data in bands B2, B3, B4, B5, B6, B7, B8, B8A, B11, and B12. The generated water index, snow index, texture index, and monthly quantile characteristics of sea ice drift data were also analyzed. The training sample data is input into the random forest machine learning method to generate a monthly classification probability map of sea ice.

8. A sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 2, characterized in that, The final sea ice classification result map is obtained based on the time series segmentation results of sea ice data and the monthly sea ice classification probability map, including: Iterate through the time-series segmentation results and monthly classification probability results for each pixel; A weighted average of the probabilities of each pixel for each category within each time series is calculated. The category with the highest weighted average probability is the final category of the pixel, forming the final 10m sea ice classification result map.

9. A sea ice classification method based on multispectral and SAR fusion remote sensing data according to claim 2, characterized in that, The water body indices include the automatically calculated water body index, the normalized water body index, and the improved normalized water body index; the texture indices include variance, uniformity, and angular second moment; the snow indices include the normalized principal component snow index.

Images