Sea ice drift extraction method based on optical remote sensing image

Abstract

The invention discloses a sea ice drift extraction method based on an optical remote sensing image. The method comprises the steps that the optical remote sensing image is acquired for data preprocessing; screening images from the to-be-processed data according to the space-time overlapping degree; performing histogram equalization according to the gray level distribution of the sea ice area in the target image; performing Laplace enhancement on the image after histogram equalization; respectively extracting feature points in the original image, the histogram equalization image and the Laplacian image by using an A-KAZE operator, and matching the feature points based on a feature descriptor; error filtering is carried out on the extracted preliminary sea ice drift vector, then extraction vectors in the original image, the histogram equalization image and the Laplacian image are combined, and a final sea ice drift result is generated. According to the embodiment of the invention, sea ice drift information with good space coverage and high quality can be obtained from medium-resolution optical remote sensing data such as MODIS, and the method can be widely applied to the technical field of computers.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method for extracting sea ice drift based on optical remote sensing images. Background Art

[0002] Arctic sea ice drift is an important parameter in sea ice dynamics, which directly determines the spatial redistribution of sea ice and affects the mass balance of sea ice. The movement of Arctic sea ice is complex and shows obvious differences in time and space. Therefore, the study of sea ice drift is related to the safety of Arctic shipping routes. At the same time, timely understanding of sea ice movement changes will help to rationally plan Arctic resource development activities and reduce corresponding risks and costs.

[0003] At present, microwave remote sensing is widely used in tracking Arctic sea ice movement due to its all-day and all-weather observation capabilities. The data sources of polar sea ice drift remote sensing monitoring products released by international research institutions such as NSIDC, I FREMER, and OSI-SAF are all microwave data; however, due to insufficient spatial resolution, it is difficult to monitor sea ice movement at a finer scale. In addition, sea ice drift tracking with microwave data relies on sea ice radiation brightness temperature or backscattering signals. The melting of sea ice in summer causes changes in brightness temperature and backscattering coefficient, which limits the availability of microwave data in summer sea ice drift tracking. Compared with microwave data, the sea ice spectral information and texture information provided by optical data are less affected by the melting of the sea ice surface. At the same time, thanks to its medium spatial resolution, wide swath, and short revisit period, medium-resolution optical images have considerable application potential in summer sea ice drift.

[0004] However, the practical application of extracting sea ice drift using optical remote sensing images is still limited by cloud interference and the lack of sea ice texture. Summary of the invention

[0005] The main purpose of the embodiment of the present invention is to propose a sea ice drift extraction method based on optical remote sensing images, which effectively overcomes the problems of cloud coverage and texture deficiency in optical remote sensing images, and can obtain high-quality sea ice drift information with good spatial coverage from medium-resolution optical remote sensing data such as MODIS.

[0006] To achieve the above object, one aspect of an embodiment of the present invention provides a method for extracting sea ice drift based on optical remote sensing images, comprising the following steps:

[0007] Acquiring an optical remote sensing image, and performing data preprocessing on the optical remote sensing image to obtain data to be processed;

[0008] Screening images from the data to be processed according to the temporal and spatial overlap, and selecting a target image;

[0009] Performing histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image;

[0010] Perform Laplace enhancement on the image after histogram equalization to obtain a Laplace image;

[0011] The A-KAZE operator is used to extract feature points from the original image, histogram equalization image and Laplace image, and the feature points are matched based on the feature descriptor to generate the preliminary sea ice drift vector.

[0012] The extracted preliminary sea ice drift vector is filtered, and then the extracted vectors in the original image, the histogram equalization image and the Laplace image are merged to generate the final sea ice drift result.

[0013] In some embodiments, the step of acquiring an optical remote sensing image and performing data preprocessing on the optical remote sensing image to obtain data to be processed includes the following steps:

[0014] Obtaining an optical apparent reflectance image and a corresponding sea ice classification image; wherein the optical apparent reflectance image is used for image feature extraction; and the sea ice classification image is used to determine the sea ice range in the histogram equalization process;

[0015] The various images obtained are geocoded and reprojected to the North Pole azimuth coordinate system to obtain the data to be processed.

[0016] In some embodiments, the step of screening images from the data to be processed according to the spatiotemporal overlap to select a target image is specifically as follows:

[0017] The preprocessed images are subjected to a spatiotemporal overlap analysis, wherein the generation time and spatial range of the images are read from the images themselves; images with a time difference of 18 to 30 hours and a spatial overlap of more than 60% are selected to form a matching pair to obtain the target image.

[0018] In some embodiments, performing histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image comprises the following steps:

[0019] According to the sea ice extent data, make the corresponding sea ice mask;

[0020] Calculate the grayscale probability density distribution function and cumulative probability distribution density function of the sea ice area;

[0021] The cumulative probability distribution density function of the sea ice area is used to stretch the grayscale distribution of the entire image to achieve histogram equalization;

[0022] Among them, the gray probability density distribution function Pr (r k ) is calculated as:

[0023]

[0024] Cumulative probability distribution density function T(r k ) is calculated as:

[0025]

[0026] Where MN is the number of pixels within the sea ice area, n k The gray level is r in the sea ice area k The number of pixels, L is the maximum gray value of the image; r k Represents the gray level in the sea ice area.

[0027] In some embodiments, performing Laplace enhancement on the image after histogram equalization to obtain a Laplace image includes the following steps:

[0028] Convolve the histogram equalized image with a 3x3 Laplacian operator to achieve Laplacian filtering;

[0029] Subtract the Laplace filtered image from the histogram equalized image to obtain the Laplace sharpened image;

[0030] Among them, the expression of Laplace filtering is:

[0031]

[0032] The expression of Laplace sharpening process is:

[0033]

[0034] Among them, f(x,y) is the image after histogram equalization, is the Laplacian filtered image, g(x,y) is the Laplacian sharpened image, and x,y are the pixel coordinates.

[0035] In some embodiments, the method of extracting feature points from the original image, the histogram equalization image, and the Laplace image respectively using the A-KAZE operator, and matching the feature points based on the feature descriptor to generate a preliminary sea ice drift vector includes the following steps:

[0036] A nonlinear grayscale gradient space is constructed in the image through the A-KAZE operator;

[0037] For the A-KAZE operator, the construction of the nonlinear feature space is controlled by the number of layers and the number of sub-layers in each layer in the image pyramid;

[0038] Matching is performed using a brute force matching method based on the distance of feature descriptors in the nonlinear grayscale gradient space;

[0039] Among them, the conversion formula of nonlinear gray gradient space is:

[0040]

[0041] Among them, L is the grayscale matrix of the image, div and Represent the solution functions of divergence and gradient respectively, c(x,y,t) is a conduction function; (x,y) represents the pixel position; t represents the diffusion time.

[0042] In some embodiments, filtering the extracted preliminary sea ice drift vector, and then merging the extracted vectors in the original image, the histogram equalization image, and the Laplace image to generate the final sea ice drift result includes the following steps:

[0043] Filter out errors through the nearest neighbor distance ratio test, set the nearest neighbor distance ratio test threshold to 0.75, and filter out the wrong matching point pairs with a distance ratio greater than the nearest neighbor distance ratio threshold;

[0044] Delete the vectors with too large deviations compared to the neighborhood, calculate the difference between the Euclidean distance and the cosine distance between the current drift γ0 and the simulated drift γ, and if the distance difference is within the allowable threshold, retain the current drift γ0, otherwise delete it;

[0045] Merge the extracted vectors from the original image, histogram equalized image, and Laplace image to generate the final sea ice drift results;

[0046] Among them, the formulas for the nearest neighbor ratio test and the simulated drift amount γ are as follows:

[0047] dis m <R*dis n ,

[0048] Among them, dis m and dis n They are respectively the nearest Hamming distance and the next nearest Hamming distance between the feature point to be matched and the feature point on another image, and R is the proportional coefficient.

[0049] The embodiments of the present invention include at least the following beneficial effects: the present invention provides a method for extracting sea ice drift based on optical remote sensing images, which obtains optical remote sensing images and performs data preprocessing on the optical remote sensing images to obtain data to be processed; selects images from the data to be processed according to the temporal and spatial overlap, and selects a target image; performs histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image; performs Laplace enhancement on the image after the histogram equalization to obtain a Laplace image; uses the A-KAZE operator to extract feature points in the original image, the histogram equalized image, and the Laplace image, respectively, and matches the feature points based on the feature descriptor to generate a preliminary sea ice drift vector; filters the extracted preliminary sea ice drift vector, and then merges the extracted vectors in the original image, the histogram equalized image, and the Laplace image to generate a final sea ice drift result. The embodiments of the present invention can effectively overcome the cloud coverage and texture deficiency problems existing in optical remote sensing images, and can obtain high-quality sea ice drift information with good spatial coverage from medium-resolution optical remote sensing data such as MODIS. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the overall steps provided by an embodiment of the present invention;

[0051] Figure 2 It is a flowchart of a specific implementation process provided by an embodiment of the present invention;

[0052] Figure 3 The embodiments of the present invention provide three pairs of MODIS image data, MODIS sea ice classification data, and the spatiotemporal distribution of the buoy data used for verification;

[0053] Figure 4 is a comparison diagram of sea ice drift extraction results before and after histogram equalization provided by an embodiment of the present invention;

[0054] Figure 5 is a comparison diagram of sea ice drift results before and after Laplace enhancement provided by an embodiment of the present invention;

[0055] Figure 6 It is a final extraction result diagram of three pairs of MODIS image data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The embodiment of the present invention provides a method for extracting sea ice drift based on optical remote sensing images, which relates to the field of computer technology.

[0057] Reference Figure 1 , Figure 1A flow chart of a method for extracting sea ice drift based on optical remote sensing images provided in an embodiment of the present invention, the method may include the following steps:

[0058] Acquiring an optical remote sensing image, and performing data preprocessing on the optical remote sensing image to obtain data to be processed;

[0059] Screening images from the data to be processed according to the temporal and spatial overlap, and selecting a target image;

[0060] Performing histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image;

[0061] Perform Laplace enhancement on the image after histogram equalization to obtain a Laplace image;

[0062] The A-KAZE operator is used to extract feature points from the original image, histogram equalization image and Laplace image, and the feature points are matched based on the feature descriptor to generate the preliminary sea ice drift vector.

[0063] The extracted preliminary sea ice drift vector is filtered, and then the extracted vectors in the original image, the histogram equalization image and the Laplace image are merged to generate the final sea ice drift result.

[0064] In some embodiments, the step of acquiring an optical remote sensing image and performing data preprocessing on the optical remote sensing image to obtain data to be processed includes the following steps:

[0065] Obtaining an optical apparent reflectance image and a corresponding sea ice classification image; wherein the optical apparent reflectance image is used for image feature extraction; and the sea ice classification image is used to determine the sea ice range in the histogram equalization process;

[0066] The various images obtained are geocoded and reprojected to the North Pole azimuth coordinate system to obtain the data to be processed.

[0067] In some embodiments, the step of screening images from the data to be processed according to the spatiotemporal overlap to select a target image is specifically as follows:

[0068] The preprocessed images are subjected to a spatiotemporal overlap analysis, wherein the generation time and spatial range of the images are read from the images themselves; images with a time difference of 18 to 30 hours and a spatial overlap of more than 60% are selected to form a matching pair to obtain the target image.

[0069] In some embodiments, performing histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image comprises the following steps:

[0070] According to the sea ice extent data, make the corresponding sea ice mask;

[0071] Calculate the grayscale probability density distribution function and cumulative probability distribution density function of the sea ice area;

[0072] The cumulative probability distribution density function of the sea ice area is used to stretch the grayscale distribution of the entire image to achieve histogram equalization;

[0073] Among them, the gray probability density distribution function P r (r k ) is calculated as:

[0074]

[0075] Cumulative probability distribution density function T(r k ) is calculated as:

[0076]

[0077] Where MN is the number of pixels within the sea ice area, n k The gray level is r in the sea ice area k The number of pixels, L is the maximum gray value of the image; r k Represents the gray level in the sea ice area.

[0078] In some embodiments, performing Laplace enhancement on the image after histogram equalization to obtain a Laplace image includes the following steps:

[0079] Convolve the histogram equalized image with a 3x3 Laplacian operator to achieve Laplacian filtering;

[0080] Subtract the Laplace filtered image from the histogram equalized image to obtain the Laplace sharpened image;

[0081] Among them, the expression of Laplace filtering is:

[0082]

[0083] The expression of Laplace sharpening process is:

[0084]

[0085] Among them, f(x,y) is the image after histogram equalization, is the Laplacian filtered image, g(x,y) is the Laplacian sharpened image, and x,y are the pixel coordinates.

[0086] In some embodiments, the method of extracting feature points from the original image, the histogram equalization image, and the Laplace image respectively using the A-KAZE operator, and matching the feature points based on the feature descriptor to generate a preliminary sea ice drift vector includes the following steps:

[0087] A nonlinear grayscale gradient space is constructed in the image through the A-KAZE operator;

[0088] For the A-KAZE operator, the construction of the nonlinear feature space is controlled by the number of layers and the number of sub-layers in each layer in the image pyramid;

[0089] Matching is performed using a brute force matching method based on the distance of feature descriptors in the nonlinear grayscale gradient space;

[0090] Among them, the conversion formula of nonlinear gray gradient space is:

[0091]

[0092] Among them, L is the grayscale matrix of the image, div and Represent the solution functions of divergence and gradient respectively, c(x,y,t) is a conduction function; (x,y) represents the pixel position; t represents the diffusion time.

[0093] In some embodiments, filtering the extracted preliminary sea ice drift vector, and then merging the extracted vectors in the original image, the histogram equalization image, and the Laplace image to generate the final sea ice drift result includes the following steps:

[0094] Filter out errors through the nearest neighbor distance ratio test, set the nearest neighbor distance ratio test threshold to 0.75, and filter out the wrong matching point pairs with a distance ratio greater than the nearest neighbor distance ratio threshold;

[0095] Delete the vectors with too large deviations compared to the neighborhood, calculate the difference between the Euclidean distance and the cosine distance between the current drift γ0 and the simulated drift γ, and if the distance difference is within the allowable threshold, retain the current drift γ0, otherwise delete it;

[0096] Merge the extracted vectors from the original image, histogram equalized image, and Laplace image to generate the final sea ice drift results;

[0097] Among them, the formulas for the nearest neighbor ratio test and the simulated drift amount γ are as follows:

[0098] dis m <R*dis n ,

[0099] Among them, dis m and dis nThey are respectively the nearest Hamming distance and the next nearest Hamming distance between the feature point to be matched and the feature point on another image, and R is the proportional coefficient.

[0100] The following describes the specific implementation process of the embodiment of the present invention in detail by taking a specific application scenario as an example:

[0101] In view of the problems existing in the prior art, in order to make full use of the characteristics of optical remote sensing images such as high spatial resolution and short revisit period, and to overcome the problems of cloud cover and lack of texture that are common in optical remote sensing images and are not conducive to sea ice drift extraction, the present invention can effectively solve the above problems by using a series of image enhancement technologies.

[0102] Specifically, an embodiment of the present invention provides a sea ice drift vector extraction method, which, with the support of image processing and feature matching technology, can obtain high-quality sea ice drift information with good spatial coverage from medium-resolution optical remote sensing data such as MODIS.

[0103] refer to Figure 2 The method of the embodiment of the present invention comprises the following steps:

[0104] Step 1: Obtain optical remote sensing images and perform data preprocessing;

[0105] Step 1.1: Download the optical apparent reflectance image and the corresponding sea ice classification image;

[0106] Step 1.2: Geocode the optical image and reproject it to the North Pole coordinate system (EPSG:3411).

[0107] The optical image data acquired by the embodiment of the present invention includes calibrated apparent (surface) reflectance data and corresponding sea ice extent data. Taking the MODIS series data as an example, it is MOD09&MYD09 (or MOD02&MYD02) data and MOD29&MYD29 data. The former is used for feature extraction, and the latter is used to determine the sea ice extent in the histogram equalization step. The acquired image data is geocoded and reprojected into the polar coordinate system to facilitate the subsequent sea ice drift extraction.

[0108] Step 2: For two images A and B, perform temporal and spatial overlap judgment. If the acquisition time difference between A and B is 18 to 30 hours and the spatial overlap is more than 60%, they constitute a matching image pair;

[0109] The embodiment of the present invention performs a spatiotemporal overlap analysis on the preprocessed images, wherein the generation time and spatial range of the images can be read from the images themselves. Images with a time difference of 18 to 30 hours and a spatial overlap of more than 60% are selected to form a matching pair, wherein the judgment of the spatial overlap is based on the Python Shapely module; Figure 3 The figure shows the spatial and temporal distribution of three pairs of MODIS image data, MODIS sea ice classification data and buoy data used for verification selected in the present invention. It should be noted that the purpose of constructing the matching pair in step 2 of the embodiment of the present invention is to calculate the drift vector of sea ice, which requires two images with a certain overlapping area at two adjacent times, so as to obtain the drift vector of sea ice from time A to time B.

[0110] Step 3: Perform histogram equalization on the image according to the grayscale distribution of the sea ice area in the image;

[0111] Step 3.1: Create a corresponding sea ice mask based on the sea ice extent data;

[0112] Step 3.2: Calculate the grayscale probability density distribution function (PDF) and cumulative probability distribution density function (CDF) of the sea ice area;

[0113] Step 3.3: Use the CDF of the sea ice area to stretch the grayscale distribution of the entire image to achieve histogram equalization;

[0114] Specifically, an embodiment of the present invention produces a corresponding sea ice mask based on the sea ice range data to determine the actual sea ice coverage area in the image, calculates the grayscale probability density distribution function (PDF) and the cumulative probability distribution density function (CDF) of the sea ice area, and then uses the CDF of the sea ice area to stretch the grayscale distribution of the entire image to achieve histogram equalization.

[0115] like Figure 4 As shown in the figure, the results of sea ice drift extraction by the method of the present invention are compared before and after histogram equalization. The number of sea ice drift vectors extracted by histogram equalization using the sea ice area is greater than that of the original image, and the spatial coverage is larger. The corresponding calculation formulas of PDF and CDF of the sea ice area are as follows:

[0116]

[0117] Where MN is the number of pixels within the sea ice area in the MOD09 or MYD09 image, and n k The gray level is r in the sea ice area k The number of pixels, L is the maximum grayscale value of the image, which is set to 4096.

[0118] Step 4: Perform Laplace enhancement on the image after histogram equalization;

[0119] Step 4.1: Convolve the histogram equalized image with a 3x3 Laplacian operator to achieve Laplacian filtering;

[0120] Step 4.2: Subtract the Laplace filtered image from the histogram equalized image to obtain the Laplace sharpened image;

[0121] Specifically, the Laplace filtering and Laplace sharpening of the embodiment of the present invention are used to highlight the texture features of the optical image at the edge of the sea ice. The image after histogram equalization is convolved with a 3x3 Laplace operator to implement Laplace filtering. The Laplace sharpened image is obtained by subtracting the Laplace filtered image from the histogram equalization image. Figure 5 Shown is a comparison of sea ice drift results before and after Laplace enhancement using the method of the present invention.

[0122] The corresponding formulas for Laplace filtering and Laplace sharpening are as follows:

[0123]

[0124]

[0125] Among them, f(x,y) is the image after histogram equalization, is the Laplacian filtered image, g(x,y) is the Laplacian sharpened image, and x,y are the pixel coordinates.

[0126] Step 5: Use the A-KAZE operator to extract feature points from the original image, histogram equalization image, and Laplace image, respectively, and generate a preliminary sea ice drift vector;

[0127] Specifically, the embodiment of the present invention constructs a nonlinear grayscale gradient space in the image through the A-KAZE feature tracking operator; for the A-KAZE descriptor, the construction of the nonlinear feature space is controlled by the number of layers in the image pyramid and the number of sublayers in each layer; finally, the brute force matching method is used to match according to the distance of the feature descriptor in the grayscale space. The conversion formula of the classic nonlinear diffusion space is as follows:

[0128]

[0129] Among them, L is the grayscale matrix of the image, div and Represent the solution functions of divergence and gradient respectively, c(x,y,t) is a conduction function, which is affected by the pixel position (x,y) and diffusion time t.

[0130] Step 6: Filter the extracted sea ice drift vector;

[0131] Step 6.1: Screen the points with a distance ratio greater than the nearest neighbor distance ratio threshold (0.75) as incorrect matching points;

[0132] Step 6.2: Delete vectors that deviate too much from their neighborhood;

[0133] Step 6.3: Merge the extracted vectors from the original image, histogram equalized image, and Laplace image to generate the final sea ice drift result.

[0134] Specifically, the embodiment of the present invention filters the extracted sea ice drift vector. First, the nearest neighbor distance ratio test is used for error filtering, and the nearest neighbor distance ratio test threshold is set to 0.75, and the distance ratio greater than the nearest neighbor distance ratio threshold is screened as an erroneous matching point pair. Then, the vectors with excessive deviations compared to the neighborhood are deleted, and the difference between the Euclidean distance and the cosine distance between the current drift amount γ0 and the simulated drift amount γ (the n other drift amounts within 25km of γ0 are obtained by inverse distance weighted interpolation at the starting point) is calculated. When the distance difference is within the allowable threshold, the current drift amount γ0 is retained, otherwise it is deleted; finally, the extracted vectors in the original image, histogram equalization image, and Laplace image are merged to generate the final sea ice drift result.

[0135] Among them, the formulas for the nearest neighbor ratio test and the simulated drift amount γ are as follows:

[0136] dis m <R*dis n

[0137] Among them, dis m and dis n They are respectively the nearest Hamming distance and the next nearest Hamming distance between the feature point to be matched and the feature point on another image, and R is the proportional coefficient.

[0138] like Figure 6 The figure shows the final extraction results of the method of the present invention in three pairs of MODIS image data. The results show that a large number of sea ice drift vectors are extracted in most cloud-free areas, proving that the method proposed in the present invention can effectively extract sea ice drift vectors in optical images.

[0139] Table 1 shows the deviation between the extraction results of regions I-III and the IABP buoy observation results. The vectors around the buoy are interpolated to the buoy position for verification. The total MAE and RMSE are 707m / d and 864m / d in speed, and 6.4° and 7.5° in direction, respectively. The quantitative results show that the method proposed in the present invention can extract sea ice drift information with good spatial coverage and excellent accuracy. The method proposed in the embodiment of the present invention has certain advantages in sea ice drift extraction.

[0140] Table 1 Verification results of IABP buoy observations in study areas I-III

[0141]

[0142] In summary, the present invention proposes a method for extracting sea ice drift from optical remote sensing images. With the support of image processing and feature matching technology, high-quality sea ice drift information can be obtained from medium-resolution optical remote sensing data such as MODIS. The method mainly includes the following steps: (1) optical image data preprocessing; (2) image matching pair construction based on spatiotemporal overlap analysis; (3) image histogram equalization based on sea ice range; (4) image Laplace filtering and Laplace sharpening; (5) sea ice drift vector generation supported by the A-KAZE feature tracking algorithm; (6) extraction result error filtering and post-processing; The present invention effectively overcomes the problems of cloud cover and texture deficiency that are common in optical remote sensing images and are not conducive to sea ice drift extraction. It can extract sea ice drift information with good spatial coverage and excellent accuracy, which is of great value in the study of sea ice movement in the Arctic and Antarctic.

[0143] The embodiments described in the embodiments of the present invention are intended to more clearly illustrate the technical solutions of the embodiments of the present invention, and do not constitute a limitation on the technical solutions provided by the embodiments of the present invention. Those skilled in the art can appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present invention are also applicable to similar technical problems.

[0144] Those skilled in the art will appreciate that the technical solutions shown in the figures do not limit the embodiments of the present invention and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

Claims

1. A method for extracting sea ice drift based on optical remote sensing images, characterized in that: The following steps are involved: Acquiring an optical remote sensing image, and performing data preprocessing on the optical remote sensing image to obtain data to be processed; Screening images from the data to be processed according to the temporal and spatial overlap, and selecting a target image; Performing histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image; Perform Laplace enhancement on the image after histogram equalization to obtain a Laplace image; The A-KAZE operator is used to extract feature points from the original image, histogram equalization image and Laplace image, and the feature points are matched based on the feature descriptor to generate the preliminary sea ice drift vector. The extracted preliminary sea ice drift vector is filtered, and then the extracted vectors in the original image, the histogram equalization image and the Laplace image are merged to generate the final sea ice drift result.

2. The method for extracting sea ice drift based on optical remote sensing images according to claim 1, characterized in that: The step of acquiring the optical remote sensing image and performing data preprocessing on the optical remote sensing image to obtain the data to be processed comprises the following steps: Obtaining an optical apparent reflectance image and a corresponding sea ice classification image; wherein the optical apparent reflectance image is used for image feature extraction; and the sea ice classification image is used to determine the sea ice range in the histogram equalization process; The various images obtained are geocoded and reprojected to the North Pole azimuth coordinate system to obtain the data to be processed.

3. The method for extracting sea ice drift based on optical remote sensing images according to claim 1, characterized in that: The step of screening images from the data to be processed according to the spatiotemporal overlap and selecting a target image is specifically as follows: The preprocessed images are subjected to a spatiotemporal overlap analysis, wherein the generation time and spatial range of the images are read from the images themselves; images with a time difference of 18 to 30 hours and a spatial overlap of more than 60% are selected to form a matching pair to obtain the target image.

4. The method for extracting sea ice drift based on optical remote sensing images according to claim 1, characterized in that: The step of performing histogram equalization on the target image according to the grayscale distribution of the sea ice area in the target image to obtain a histogram equalized image comprises the following steps: According to the sea ice extent data, make the corresponding sea ice mask; Calculate the grayscale probability density distribution function and cumulative probability distribution density function of the sea ice area; The cumulative probability distribution density function of the sea ice area is used to stretch the grayscale distribution of the entire image to achieve histogram equalization; Among them, the gray probability density distribution function P r (r k ) is calculated as: Cumulative probability distribution density function T(r k ) is calculated as: Where MN is the number of pixels within the sea ice area, n k The gray level is r in the sea ice area k The number of pixels, L is the maximum gray value of the image; r k Represents the gray level in the sea ice area.

5. The method for extracting sea ice drift based on optical remote sensing images according to claim 1, characterized in that: The step of performing Laplace enhancement on the image after histogram equalization to obtain a Laplace image comprises the following steps: Convolve the histogram equalized image with a 3x3 Laplacian operator to achieve Laplacian filtering; Subtract the Laplace filtered image from the histogram equalized image to obtain the Laplace sharpened image; Among them, the expression of Laplace filtering is: The expression of Laplace sharpening process is: Among them, f(x,y) is the image after histogram equalization, is the Laplacian filtered image, g(x,y) is the Laplacian sharpened image, and x,y are the pixel coordinates.

6. The method for extracting sea ice drift based on optical remote sensing images according to claim 1, characterized in that: The method of using the A-KAZE operator to extract feature points from the original image, the histogram equalization image, and the Laplace image, and matching the feature points based on feature descriptors to generate a preliminary sea ice drift vector includes the following steps: A nonlinear grayscale gradient space is constructed in the image through the A-KAZE operator; For the A-KAZE operator, the construction of the nonlinear feature space is controlled by the number of layers and the number of sub-layers in each layer in the image pyramid; Matching is performed using a brute force matching method based on the distance of feature descriptors in the nonlinear grayscale gradient space; Among them, the conversion formula of nonlinear gray gradient space is: Among them, L is the grayscale matrix of the image, div and Represent the solution functions of divergence and gradient respectively, c(x,y,t) is a conduction function; (x,y) represents the pixel position; t represents the diffusion time.

7. The method for extracting sea ice drift based on optical remote sensing images according to claim 1, characterized in that: The method of filtering the extracted preliminary sea ice drift vector, and then merging the extracted vectors in the original image, the histogram equalization image, and the Laplace image to generate the final sea ice drift result includes the following steps: Filter out errors through the nearest neighbor distance ratio test, set the nearest neighbor distance ratio test threshold to 0.75, and filter out the wrong matching point pairs with a distance ratio greater than the nearest neighbor distance ratio threshold; Delete the vectors with too large deviations compared to the neighborhood, calculate the difference between the Euclidean distance and the cosine distance between the current drift γ0 and the simulated drift γ, and if the distance difference is within the allowable threshold, retain the current drift γ0, otherwise delete it; Merge the extracted vectors from the original image, histogram equalized image, and Laplace image to generate the final sea ice drift results; Among them, the formulas for the nearest neighbor ratio test and the simulated drift amount γ are as follows: dis m <R*dis n , Among them, dis m and dis n They are respectively the nearest Hamming distance and the next nearest Hamming distance between the feature point to be matched and the feature point on another image, and R is the proportional coefficient.

Images