A remote sensing method for extracting Arctic sea ice types

Abstract

The present invention discloses a method for extracting Arctic sea ice types remotely, comprising the following steps: (1) collecting HY1C / D‑CZI DN image data, and calculating the satellite top-of-atmosphere reflectivity according to the radiation calibration coefficient and the top-of-atmosphere reflectivity; constructing a sample data set of multiple targets including clouds, water bodies, new ice, one-year ice, and multi-year ice; (2) based on the sample data set of multiple targets, calculating cloud feature quantities, inputting them into a random forest for training and verification, and performing a comprehensive evaluation on the cloud detection accuracy; (3) based on the sample data set of multiple targets, calculating sea ice feature quantities, inputting them into a random forest for training and verification, and performing a quantitative evaluation on the sea ice classification accuracy; (4) finally generating a high-resolution sea ice type remote sensing product set, and then analyzing the spatiotemporal variation characteristics of the sea ice type; the present invention provides technical support for the development of Arctic sea ice type remote sensing products of autonomous ocean color satellites.

Description

Technical Field

[0001] The invention relates to the technical field of ocean remote sensing, and in particular to a method for extracting Arctic sea ice types by remote sensing. Background Art

[0002] Polar sea ice is an important part of the Earth's climate system and an indicator and amplifier of climate and environmental changes. It affects the exchange of heat, mass, momentum, etc. between the ocean and the atmosphere, making it an important factor in the climate and weather system. In addition, the freezing and movement of sea ice have a great impact on marine activities such as shipping safety and offshore energy development.

[0003] Sea ice type identification is an important parameter for describing sea ice characteristics, and plays an important role in studying other sea ice parameters (such as sea ice area, maximum sea ice edge line, etc.). In the early days, accurate sea ice type identification information could be obtained through manual field detection, but it was very costly and the monitoring range was limited. In comparison, satellite remote sensing technology has been widely used due to its advantages of real-time, rapid, and large-area observation, and has become the main technical means for sea ice monitoring, such as optical remote sensing, microwave radiometers, and synthetic aperture radars (SAR). Among them, optical satellite remote sensing has high temporal and spatial resolutions, and most satellite data are open and free, with high accessibility, making it an effective tool for sea ice monitoring and providing important data support for sea ice type identification research.

[0004] The existing optical remote sensing monitoring methods are mainly based on foreign multispectral optical satellite sensors, such as MODIS and Landsat series satellite images. However, my country's independent ocean color satellite HY1C / D-CZI sensor (spatial resolution 50m, revisit period 3 days) is rarely used, resulting in a waste of satellite resources. In addition, there are few reports on the research of remote sensing extraction algorithms for Arctic sea ice types for HY1C / D-CZI, resulting in a shortage of domestic HY1C / D-CZI remote sensing products of Arctic sea ice types. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method for extracting Arctic sea ice types through remote sensing, so as to solve the technical problem that the recognition accuracy of Arctic sea ice types in the current ocean color satellite HY1C / D-CZI images is not high.

[0006] Technical solution: The method for extracting the type of Arctic sea ice by remote sensing according to the present invention comprises the following steps:

[0007] (1) Collect HY1C / D-CZI DN image data and calculate the satellite top-of-atmosphere reflectivity based on the radiation calibration coefficient and the top-of-atmosphere reflectivity. R toa(l) Data: Using the published weekly sea ice type products, medium-resolution MODIS sensor 06_L2 cloud products, visual interpretation and spectral features, we build sample datasets for multiple targets including clouds, water bodies and sea ice;

[0008] (2) Based on a sample data set of multiple targets, cloud features including six spectral features, four structural features, and three texture features were calculated. The obtained features were input into a random forest for training and verification, and a comprehensive evaluation of cloud detection accuracy was performed.

[0009] (3) Based on a sample data set of multiple targets, the sea ice characteristics, including eight spectral features and three texture features, were calculated. The obtained characteristics were input into the random forest for training and validation, and the sea ice classification accuracy was quantitatively evaluated.

[0010] (4) Finally, a high-resolution remote sensing product set of sea ice types is generated, and the spatiotemporal variation characteristics of the sea ice types are analyzed.

[0011] Furthermore, in step (1), the satellite top-of-atmosphere reflectivity R toa (l) The formula is as follows:

[0012] ;

[0013] Where DN(λ) is the grayscale value of the load data observed at wavelength λ; g(λ) and b (λ) is the gain and offset of the sensor radiometric calibration at wavelength λ; d is the distance between the Sun and the Earth when the satellite image passes; θ s is the solar zenith angle when the remote sensing image passes through; E sun(λ) is the mean top-of-atmosphere solar irradiance.

[0014] Furthermore, in step (1), sea ice includes: new ice, one-year ice, and multi-year ice.

[0015] Furthermore, the six spectral features include: top of atmosphere reflectivity at the blue band R TOA, blue , top of atmosphere reflectivity at green wavelengths R TOA, green , top of atmosphere reflectivity in the red band R TOA, red , the reflectivity of the top of the atmosphere in the near-infrared band R TOA, nir , Whiteness Index, Normalized Difference Vegetation Index NDVI; the Whiteness formula is as follows:

[0016] ;

[0017] in, i = blue, green, red, nir represent the blue, green, red and near-infrared bands of CZI respectively; M is the mean of the top-of-atmosphere reflectance in the four bands, whiteness is the whiteness index, and M represents the absolute value of the difference between the reflectance of each band and the mean;

[0018] The NDVI formula is as follows:

[0019] ;

[0020] in, R TOA, nir is the reflectivity of the top of the atmosphere in the near-infrared band, R TOA, red is the reflectivity of the top of the atmosphere in the red light band.

[0021] Furthermore, in step (2), the three texture features include: the mean feature calculated by the gray level co-occurrence matrix Mean GLCM , Variance characteristics calculated by gray level co-occurrence matrix GLCM , Contrast feature calculated by gray level co-occurrence matrix GLCM , the formulas are as follows:

[0022] ;

[0023] ;

[0024] ;

[0025] Among them, m, n are the number of rows and columns of the image, i and j are the position indexes of the pixels, θ is the direction, d is the distance between pixels, and μ is the mean.

[0026] Furthermore, in step (2), the four structural features include:

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] Where I is the input image, ▽ is the first-order difference operator, pis the image pixel index; w1 is the balance value 0.005, w2 is the balance value 0.01, w3 is the balance value 0.015, and w4 is the balance value 0.02; S1, S2, S3, and S4 represent the output result images corresponding to the balance values ​​w1, w2, w3, and w4 respectively; Indicates the value of variable S when it reaches the minimum value.

[0032] Furthermore, in step (3), the eight spectral features include: the top of atmosphere reflectivity at the blue light band R TOA, blue , top of atmosphere reflectivity at green wavelengths R TOA, green , top of atmosphere reflectivity in the red band R TOA, red , the reflectivity of the top of the atmosphere in the near-infrared band R TOA, nir , near infrared and blue light top of atmosphere reflectivity difference and ratio NDWI H , the difference and ratio of the top-of-atmosphere reflectivity of blue light and green light N BG , the difference and ratio of the top-of-atmosphere reflectivity of blue light and red light N BR , the difference and ratio of the top-of-atmosphere reflectivity of green light and red light N GR ;in,

[0033] ;

[0034] ;

[0035] ;

[0036] ;

[0037] in, R TOA, blue is the reflectivity of the top of the atmosphere in the blue light band, R TOA, green is the reflectivity of the top of the atmosphere in the green band, R TOA, red is the reflectivity of the top of the atmosphere in the red light band, R TOA, nir is the reflectivity of the top of the atmosphere in the near-infrared band.

[0038] Furthermore, in step (3), the three texture features include:

[0039] ;

[0040] ;

[0041] ;

[0042] Among them, Mean GLCM , Variance GLCM 、Contrast GLCM They represent the mean feature, variance feature, and contrast feature of the gray-level co-occurrence matrix respectively; m, n are the number of rows and columns of the image, and i and j is the pixel position index, θ is the direction, d is the distance between pixels, μ is the mean.

[0043] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention utilizes HY1C / D-CZI remote sensing images in combination with a random forest algorithm to classify and extract Arctic sea ice types, so as to improve the monitoring capability and efficiency of sea ice type identification, and provide technical support for the development of Arctic sea ice type remote sensing products of autonomous ocean color satellites. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the present invention;

[0045] Figure 2 Inputting a feature quantity difference map into the cloud detection model of the present invention;

[0046] Figure 3 Inputting a characteristic quantity difference map to the sea ice type model of the present invention;

[0047] Figure 4 This is the identification result diagram of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0049] like Figure 1 As shown, an embodiment of the present invention provides a method for extracting Arctic sea ice types by remote sensing, comprising the following steps:

[0050] (1) Collect HY1C / D-CZI DN image data and calculate the satellite top-of-atmosphere reflectivity based on the radiation calibration coefficient and the top-of-atmosphere reflectivity. R toa (l) Data: Using the published weekly sea ice type products, medium-resolution MODIS sensor 06_L2 cloud products, visual interpretation and spectral characteristics, we build sample data sets for multiple targets including clouds, water bodies and sea ice. Sea ice includes new ice, one-year ice and multi-year ice.

[0051] Satellite top-of-atmosphere reflectivity R toa (l) The formula is as follows:

[0052] ;

[0053] Where DN(λ) is the grayscale value of the load data observed at wavelength λ; g(λ) and b (λ) is the gain and offset of the sensor radiometric calibration at wavelength λ; d is the distance between the Sun and the Earth when the satellite image passes; θ s is the solar zenith angle when the remote sensing image passes through; E sun(λ) is the mean top-of-atmosphere solar irradiance.

[0054] (2) If Figure 2 As shown in the figure, based on a sample data set of multiple targets, the cloud features are calculated, including 6 spectral features, 4 structural features, and 3 texture features. The obtained features are input into the random forest (RF-cloud model) for training and verification, and the cloud detection accuracy is comprehensively evaluated.

[0055] The six spectral features include: top-of-atmosphere reflectivity in the blue band R TOA, blue , top of atmosphere reflectivity at green wavelengths R TOA, green , top of atmosphere reflectivity in the red band R TOA, red , the reflectivity of the top of the atmosphere in the near-infrared band R TOA, nir , Whiteness Index, Normalized Difference Vegetation Index NDVI; the Whiteness formula is as follows:

[0056] ;

[0057] in, i = blue, green, red, nir represent the blue, green, red and near-infrared bands of CZI respectively; M is the mean of the top-of-atmosphere reflectance in the four bands, whiteness is the whiteness index, and M represents the absolute value of the difference between the reflectance of each band and the mean;

[0058] The NDVI formula is as follows:

[0059] ;

[0060] in, R TOA, nir is the reflectivity of the top of the atmosphere in the near-infrared band, R TOA, red is the reflectivity of the top of the atmosphere in the red light band.

[0061] The three texture features include: the mean feature calculated by the gray-level co-occurrence matrixGLCM , Variance characteristics calculated by gray level co-occurrence matrix GLCM , Contrast feature calculated by gray level co-occurrence matrix GLCM , the formulas are as follows:

[0062] ;

[0063] ;

[0064] ;

[0065] Among them, m, n are the number of rows and columns of the image, i and j are the position indexes of the pixels, θ is the direction, d is the distance between pixels, and μ is the mean.

[0066] The four structural features include:

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] Where I is the input image, ▽ is the first-order difference operator, p is the image pixel index; w1 is the balance value 0.005, w2 is the balance value 0.01, w3 is the balance value 0.015, and w4 is the balance value 0.02; S1, S2, S3, and S4 represent the output result images corresponding to the balance values ​​w1, w2, w3, and w4 respectively; Indicates the value of variable S when it reaches the minimum value.

[0072] Figure 2 middle, R TOA,blue : Blue light top of atmosphere reflectivity

[0073] R TOA,green : Green light top-of-atmosphere reflectivity

[0074] R TOA,red : Red light top-of-atmosphere reflectivity

[0075] R TOA,nir : Near-infrared top-of-atmosphere reflectivity

[0076] Whiteness: Whiteness index: The absolute value of the difference between the reflectance of each band and the average value

[0077] NDVI: Normalized Difference Vegetation Index: The difference and ratio of near-infrared and red top-of-atmosphere reflectance

[0078] Mean GLCM : Mean characteristics of gray-level co-occurrence matrix

[0079] Variance GLCM : Variance characteristics of gray-level co-occurrence matrix

[0080] Contrast GLCM : Contrast characteristics of gray-level co-occurrence matrix

[0081] S 1: Structural image with equilibrium value of 0.005

[0082] S 2: Structural image with equilibrium value of 0.01

[0083] S 3: Structural image with equilibrium value of 0.015

[0084] S 4: Structural image with equilibrium value of 0.02

[0085] (3) If Figure 3 As shown in the figure, based on the sample data set of various targets, the sea ice characteristics are calculated, including 8 spectral features and 3 texture features. The obtained features are input into the random forest (RF-ice model) for training and verification, and the sea ice classification accuracy is quantitatively evaluated; Among them, the 8 spectral features include: the top of the atmosphere reflectivity in the blue light band R TOA, blue , top of atmosphere reflectivity at green wavelengths R TOA, green , top of atmosphere reflectivity in the red band R TOA, red , the reflectivity of the top of the atmosphere in the near-infrared band R TOA, nir , near infrared and blue light top of atmosphere reflectivity difference and ratio NDWI H , the difference and ratio of the top-of-atmosphere reflectivity of blue light and green light N BG , the difference and ratio of the top-of-atmosphere reflectivity of blue light and red light N BR , the difference and ratio of the top-of-atmosphere reflectivity of green light and red light N GR ;in,

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] in, R TOA, blue is the reflectivity of the top of the atmosphere in the blue light band, R TOA, green is the reflectivity of the top of the atmosphere in the green band, R TOA, red is the reflectivity of the top of the atmosphere in the red light band, R TOA, nir is the reflectivity of the top of the atmosphere in the near-infrared band.

[0091] The three texture features include:

[0092] ;

[0093] ;

[0094] ;

[0095] Among them, Mean GLCM , Variance GLCM 、Contrast GLCM They represent the mean feature, variance feature, and contrast feature of the gray-level co-occurrence matrix respectively; m, n are the number of rows and columns of the image, and i and j is the pixel position index, θ is the direction, d is the distance between pixels, μ is the mean.

[0096] Figure 3 middle, R TOA,blue : Blue light top of atmosphere reflectivity

[0097] R TOA,green : Green light top-of-atmosphere reflectivity

[0098] R TOA,red : Red light top-of-atmosphere reflectivity

[0099] R TOA,nir : Near-infrared top-of-atmosphere reflectivity

[0100] NDWI H : Difference and ratio of near-infrared and blue light top-of-atmosphere reflectivity

[0101] N BG : The difference and ratio of the reflectivity of blue light and green light at the top of the atmosphere

[0102] NBR : The difference and ratio of the reflectivity of the top of the atmosphere between blue light and red light

[0103] N GR : The difference and ratio of the reflectivity of green light and red light at the top of the atmosphere

[0104] Mean GLCM : Mean characteristics of gray-level co-occurrence matrix

[0105] Variance GLCM : Variance characteristics of gray-level co-occurrence matrix

[0106] Contrast GLCM : Contrast characteristics of gray-level co-occurrence matrix

[0107] (4) If Figure 4 As shown in the figure, a high-resolution sea ice type remote sensing product set is finally generated, and the spatiotemporal variation characteristics of the sea ice types are then analyzed.

[0108] Accuracy evaluation index

[0109] The present invention uses the confusion matrix method to evaluate the model accuracy, and further calculates the user accuracy (Useraccuracy), production accuracy (Producer accuracy), and overall accuracy (Overall accuracy) as evaluation indicators. The calculation formula is as follows:

[0110] ;

[0111] ;

[0112] ;

[0113] Among them, TP represents the number of positive classes correctly predicted by the model as positive classes. TN represents the number of negative classes correctly predicted by the model as negative classes. FP represents the number of negative classes incorrectly predicted by the model as positive classes, also known as the first type of error. FN represents the number of positive classes incorrectly predicted by the model as negative classes, also known as the second type of error.

[0114] Cloud detection accuracy

[0115] Based on the spectral characteristics, texture characteristics and structural characteristics of the sample set, the sample set is divided into a training set and a test set in a ratio of 7:3 and input into the random forest model for model training and verification.

[0116] Table 1 Classification accuracy of random forest training set for cloud detection

[0117] ;

[0118] Table 2 Random forest test set classification accuracy for cloud detection

[0119] .

[0120] For different sea ice types R TOA Data, calculate its spectral characteristics and texture characteristics as the input of the random forest model. Divide the sample set into training set and test set in a ratio of 7:3 and input them into the random forest model for model training and verification;

[0121] Table 3 Classification accuracy of random forest training set for sea ice

[0122] ;

[0123] Table 4 Classification accuracy of random forest test set for sea ice

[0124] ;

[0125] In order to verify the accuracy of the sea ice type identification model, the shared Arctic ship-based navigation data was used for accuracy evaluation. The accuracy evaluation is shown in Table 5 below.

[0126] Table 5 Classification accuracy of the model evaluated based on ship-based data

[0127] .

Claims

1. A method for extracting Arctic sea ice types by remote sensing, characterized in that: The following steps are involved: (1) Collect HY1C / D-CZI DN image data and calculate the satellite top-of-atmosphere reflectivity based on the radiation calibration coefficient and the top-of-atmosphere reflectivity. R toa (l) Data: Using the published weekly sea ice type products, medium-resolution MODIS sensor 06_L2 cloud products, visual interpretation and spectral features, we build sample datasets for multiple targets including clouds, water bodies and sea ice; (2) Based on a sample data set of multiple targets, the cloud features are calculated, including 6 spectral features, 4 structural features, and 3 texture features. The obtained features are input into the random forest for training and verification, and the cloud detection accuracy is comprehensively evaluated. Among them, the 6 spectral features include: the top of the atmosphere reflectivity in the blue light band R TOA, blue , top of atmosphere reflectivity in green band R TOA, green , top of atmosphere reflectivity in the red band R TOA, red , the reflectivity of the top of the atmosphere in the near-infrared band R TOA, nir , Whiteness Index, Normalized Difference Vegetation Index NDVI; the Whiteness formula is as follows: ； in, i = blue, green, red, nir represent the blue, green, red and near-infrared bands of CZI respectively; M is the mean of the top-of-atmosphere reflectance in the four bands, whiteness is the whiteness index, and M represents the absolute value of the difference between the reflectance of each band and the mean; The NDVI formula is as follows: ； in, R TOA, nir is the reflectivity of the top of the atmosphere in the near-infrared band, R TOA, red is the reflectivity of the top of the atmosphere at the red light band; The three texture features include: the mean feature calculated by the gray-level co-occurrence matrix GLCM , Variance characteristics calculated by gray level co-occurrence matrix GLCM , Contrast feature calculated by gray level co-occurrence matrix GLCM , the formulas are as follows: ； ； ； Among them, m, n are the number of rows and columns of the image, i and j are the position indexes of the pixels, θ is the direction, d is the distance between pixels, and μ is the mean; The four structural features include: ； ； ； ； Where I is the input image, ▽ is the first-order difference operator, p is the image pixel index; w1 is the balance value 0.005, w2 is the balance value 0.01, w3 is the balance value 0.015, and w4 is the balance value 0.02; S1, S2, S3, and S4 represent the output result images corresponding to the balance values ​​w1, w2, w3, and w4 respectively; Indicates the value of variable S when it reaches the minimum value; (3) Based on a sample data set of multiple targets, the sea ice characteristics are calculated, including 8 spectral features and 3 texture features. The obtained features are input into the random forest for training and verification, and the sea ice classification accuracy is quantitatively evaluated. Among them, the 8 spectral features include: the top of the atmosphere reflectivity in the blue light band R TOA, blue , top of atmosphere reflectivity in green band R TOA, green , top of atmosphere reflectivity in the red band R TOA, red , the reflectivity of the top of the atmosphere in the near-infrared band R TOA, nir , near infrared and blue light top of atmosphere reflectivity difference and ratio NDWI H , the difference and ratio of the top-of-atmosphere reflectivity of blue light and green light N BG , the difference and ratio of the top-of-atmosphere reflectivity of blue light and red light N BR , the difference and ratio of the top-of-atmosphere reflectivity of green light and red light N GR ;in, ； ； ； ； in, R TOA, blue is the reflectivity of the top of the atmosphere in the blue light band, R TOA, green is the reflectivity of the top of the atmosphere in the green band, R TOA, red is the reflectivity of the top of the atmosphere in the red light band, R TOA, nir is the reflectivity of the top of the atmosphere in the near-infrared band; The three texture features include: ； ； ； Among them, Mean GLCM 、Variance GLCM 、Contrast GLCM They represent the mean feature, variance feature, and contrast feature of the gray-level co-occurrence matrix respectively; m, n are the number of rows and columns of the image, and i and j is the pixel position index, θ is the direction, d is the distance between pixels, μ is the mean (4) Finally, a high-resolution remote sensing product set of sea ice types is generated, and the spatiotemporal variation characteristics of the sea ice types are analyzed.

2. The method for extracting Arctic sea ice types by remote sensing according to claim 1, characterized in that: In step (1), the satellite top-of-atmosphere reflectivity R toa (l) The formula is as follows: ； Where DN(λ) is the grayscale value of the load data observed at wavelength λ; g(λ) and b (λ) is the gain and offset of the sensor radiometric calibration at wavelength λ; d is the distance between the Sun and the Earth when the satellite image passes; θ s is the solar zenith angle when the remote sensing image passes through; E sun(λ) is the mean top-of-atmosphere solar irradiance.

3. The method for extracting Arctic sea ice types by remote sensing according to claim 1, characterized in that: In step (1), sea ice includes new ice, one-year ice, and multi-year ice.

Images