CN-121999378-A - Method and device for determining sea ice edge based on satellite and storage medium

CN121999378ACN 121999378 ACN121999378 ACN 121999378ACN-121999378-A

Abstract

The application discloses a method and a device for determining sea ice edges based on satellites and a storage medium. Belonging to the technical field of satellite monitoring. The method comprises the steps of obtaining remote sensing images including sea ice shot by satellites and preprocessing the remote sensing images to generate sea ice images, inputting the sea ice images into a first U-Net network to generate corresponding sea ice prediction contour maps, inputting the sea ice images into a second U-Net network to generate corresponding prediction tensors, determining snow areas, sea ice melting areas, normal sea ice areas and sea water areas in the sea ice images according to the sea ice prediction contour maps and the prediction tensors, and determining sea ice edge lines in the snow areas and/or the sea ice melting areas based on Markov chain transfer matrixes. Therefore, the sea ice edge line with high precision in the snow accumulation area and the sea ice melting area is updated to the corresponding position of the sea ice prediction outline map in a seamless way, and the sea ice real outer edge outline which is complete, accurate and anti-interference can be obtained.

Inventors

QU TIANLONG
LIU ANG
XU QIKAI

Assignees

银河航天科技(安徽)有限公司

Dates

Publication Date: 20260508
Application Date: 20260108

Claims (10)

1. A method for determining an edge of a satellite-based sea ice, comprising: Acquiring remote sensing images including sea ice shot by satellites, and preprocessing the remote sensing images to generate sea ice images meeting the input requirements of a neural network; inputting the sea ice image into a first U-Net network to generate a corresponding sea ice prediction profile; Inputting the sea ice image to a second U-Net network to generate a corresponding prediction tensor, wherein the prediction tensor is used for representing probability values that each pixel point in the sea ice image respectively belongs to a snow area, a sea ice melting area and a normal area; According to the sea ice prediction contour map and the prediction tensor, determining a snow area, a sea ice melting area, a normal sea ice area and a sea water area in the sea ice image; And determining sea ice edge lines in the snow accumulation area and/or the sea ice melting area based on a pre-constructed Markov chain transfer matrix aiming at the snow accumulation area and/or the sea ice melting area.
2. The method of claim 1, wherein the first U-Net network comprises an encoding network, a first layer-skip connection module, a second layer-skip connection module, and a decoding network comprising a first decoding module and a second decoding module, and wherein the act of inputting the sea ice image to the first U-Net network to generate a corresponding sea ice prediction profile comprises: Carrying out multi-level feature extraction on the input sea ice image through the coding network to obtain a plurality of coding feature images with different levels; Transmitting a first coding feature map output by a first middle layer of the coding network to the second decoding module through the first layer jump connection module; Transmitting a second coding feature map output by a second middle layer of the coding network to the first decoding module through the second layer jump connecting module; the first decoding module is used for carrying out feature fusion and up-sampling on the second coding feature map from the second layer jump connection module to obtain a first decoding feature map; And carrying out feature fusion and up-sampling on the first decoding feature map from the first decoding module and the first coding feature map from the first layer jump connection module through the second decoding module to obtain a second decoding feature map serving as the sea ice prediction contour map.
3. The method of claim 1, wherein the predicted tensor comprises a first prediction matrix, a second prediction matrix, and a third prediction matrix, and wherein the act of inputting the sea ice image to a second U-Net network to generate a corresponding predicted tensor comprises: inputting the sea ice image to the second U-Net network configured with three classifiers; For each pixel point in the sea ice image, synchronously outputting a first probability value, a second probability value and a third probability value of which the pixel points respectively belong to a snow area, a sea ice melting area and a normal area through the second U-Net network; and respectively re-integrating the first probability value, the second probability value and the third probability value according to the positions of the pixel points in the sea ice image so as to generate a corresponding first prediction matrix, a corresponding second prediction matrix and a corresponding third prediction matrix.
4. The method of claim 1, wherein determining normal sea ice areas, sea water areas, snow areas, and sea ice melting areas in the sea ice image based on the sea ice prediction contour map and the prediction tensor comprises: based on the sea ice prediction contour map, primarily distinguishing a sea water area from a sea ice area in the sea ice image; According to the probability value that each pixel point in the prediction tensor belongs to the snow area, judging the pixel point with the probability value larger than a first preset threshold value as the snow area; Judging the pixel points with the probability value larger than a second preset threshold value as sea ice melting areas according to the probability value of the sea ice melting areas in the prediction tensor in the pixel points of the non-snow area; And determining a normal sea ice region, a sea water region, a snow-covered region and a sea ice melting region in the sea ice image according to the sea water region and the sea ice region indicated by the sea ice prediction contour map and by combining the judging results of the snow-covered region and the sea ice melting region.
5. The method according to claim 1, characterized in that the operation of determining sea ice edge lines in the snow covered region and/or the sea ice melted region based on a pre-constructed markov chain transfer matrix for the snow covered region and/or the sea ice melted region comprises: selecting an initial pixel point at the junction of the snow area or the sea ice melting area and the normal sea ice area; Determining an initial state of the initial pixel point based on the position relation of the initial pixel point relative to the adjacent known edge pixel points; Determining a next target state according to the initial state and the state transition probability defined by the Markov chain transition matrix, and selecting a next pixel point in the neighborhood of the initial pixel point according to the next target state; taking the next pixel point as a new current pixel point, and updating the current state of the current pixel point based on the position relation between the current pixel point and the previous pixel point; Iteratively executing the steps of determining the next target state, eliminating and selecting the next pixel point until the termination condition is met, so as to obtain an ordered edge pixel point sequence; And sequentially connecting the edge pixel point sequences to form a sea ice edge line passing through the snow accumulation area or the sea ice melting area.
6. The method of claim 5, wherein selecting a next pixel within the neighborhood of the starting pixel based on the next target state comprises: determining all neighborhood pixel points of the current pixel point as initial candidate pixel points; according to the pushing direction indicated by the next target state, screening out the pixel points with the consistent direction from the initial candidate pixel points to obtain a direction candidate pixel point set; Removing pixel points outside a preset limiting area from the direction candidate pixel point set to obtain an effective candidate pixel point set, wherein the limiting area is obtained by carrying out morphological expansion processing on the initial edge contours of the snow-covered area and/or the sea ice melting area; and determining the next pixel point from the effective candidate pixel point set according to a preset selection rule.
7. The method of claim 5, wherein the markov chain transfer matrix is constructed by: Acquiring a plurality of historical sea ice images with marking information, wherein the marking information comprises a preliminary sea ice edge line, a divided snow area, a divided sea ice melting area and an accurate sea ice edge line passing through the snow area and/or the sea ice melting area; Extracting edge pixel point sequences positioned in the snow accumulation area or the sea ice melting area from the accurate sea ice edge line aiming at each historical sea ice image; For each pixel point in the edge pixel point sequence, determining a pre-transfer state according to the position relation between the pixel point and the previous pixel point, and determining a post-transfer state according to the position relation between the pixel point and the next pixel point, wherein the states comprise inward pushing of the edge pixel point, current position keeping and outward pushing of the edge pixel point; Counting the occurrence times of each pre-transfer state to each post-transfer state in all the samples of the historical sea ice images; And calculating to obtain each state transition probability parameter in the Markov chain transition matrix based on the occurrence times obtained through statistics.
8. A storage medium comprising a stored program, wherein the method of any one of claims 1 to 7 is performed by a processor when the program is run.
9. A satellite-based sea ice edge determination apparatus, comprising: The sea ice image generation module is used for acquiring remote sensing images including sea ice shot by satellites, preprocessing the remote sensing images and generating sea ice images meeting the input requirements of the neural network; The prediction contour map generation module is used for inputting the sea ice image into a first U-Net network to generate a corresponding sea ice prediction contour map; the prediction tensor generation module is used for inputting the sea ice image into a second U-Net network to generate a corresponding prediction tensor, wherein the prediction tensor is used for representing probability values that each pixel point in the sea ice image respectively belongs to a snow area, a sea ice melting area and a normal area; the region determining module is used for determining a snow area, a sea ice melting area, a normal sea ice area and a sea water area in the sea ice image according to the sea ice prediction contour map and the prediction tensor; And the sea ice edge line determining module is used for determining the sea ice edge line in the snow accumulation area and/or the sea ice melting area based on a pre-constructed Markov chain transfer matrix aiming at the snow accumulation area and/or the sea ice melting area.
10. A satellite-based sea ice edge determination apparatus, comprising: processor, and A memory, coupled to the processor, for providing instructions to the processor to process the following processing steps: Acquiring remote sensing images including sea ice shot by satellites, and preprocessing the remote sensing images to generate sea ice images meeting the input requirements of a neural network; inputting the sea ice image into a first U-Net network to generate a corresponding sea ice prediction profile; Inputting the sea ice image to a second U-Net network to generate a corresponding prediction tensor, wherein the prediction tensor is used for representing probability values that each pixel point in the sea ice image respectively belongs to a snow area, a sea ice melting area and a normal area; According to the sea ice prediction contour map and the prediction tensor, determining a snow area, a sea ice melting area, a normal sea ice area and a sea water area in the sea ice image; And determining sea ice edge lines in the snow accumulation area and/or the sea ice melting area based on a pre-constructed Markov chain transfer matrix aiming at the snow accumulation area and/or the sea ice melting area.

Description

Method and device for determining sea ice edge based on satellite and storage medium Technical Field The present application relates to the field of satellite monitoring technologies, and in particular, to a method and apparatus for determining an edge of sea ice based on a satellite, and a storage medium. Background In the fields of offshore engineering, shipping safety and marine ecological protection, sea ice state monitoring has important practical significance. The sea ice in the coastal area in winter not only directly affects the port navigation, offshore operation and coastal infrastructure safety, but also changes the sea gas exchange process, and has profound effects on regional climate and marine ecosystem. Traditional monitoring means such as manual observation, shipborne radar or buoy station and the like are limited by factors such as narrow space coverage, low data updating frequency, difficult operation in severe weather and the like, and the large-range, high-precision, continuous and dynamic business monitoring requirements are difficult to realize. With the rapid development of satellite remote sensing technology, the satellite remote sensing technology has the advantages of macroscopic, objective and high-frequency observation, and provides a reliable technical approach for real-time inversion of sea ice range, type, thickness and motion parameters. However, this technique still faces the following inherent difficulties in extracting the sea ice edges (i.e., the boundary between sea ice and open sea water): 1. Sea ice surface heterogeneity causes "false edge" interference in that the sea ice surface in its natural state is not a uniform medium. Snow is often covered on the sea ice surface, and extremely high reflectivity (high gray value) is shown in a visible light-near infrared image, meanwhile, a 'melting pool' formed by melting may exist on the sea ice surface, and the reflection characteristic of the sea ice surface is similar to that of open sea water (low gray value). The extremely high and extremely low gray areas inside the ice surface are distributed in a staggered way, so that strong local gray contrast is formed. Conventional edge detection operators (e.g., sobel, canny operators, etc.) based on gray gradients are intended to detect gray abrupt changes in images. In practical applications, these operators not only can respond at the real sea water-sea ice junction, but also can respond equally strongly or even more strongly at the junctions of snow-melting pool, snow-bare ice, melting pool-bare ice and the like inside the ice surface. These edge signals generated by the internal textures of the ice surface constitute a large number of false edges, which are mixed with the external contours of the real sea ice, so that the automatic extraction algorithm is difficult to accurately distinguish and separate the real sea ice outer edge lines, and the accuracy and reliability of edge extraction are seriously reduced. 2. The sea ice edge blending pixel causes boundary blurring and positioning uncertainty, namely, in the transition zone where sea ice meets sea water, there is usually an area where crushed ice, floating ice cubes and sea water are blended. In this region, the signals received by one instantaneous field of view (i.e., one pixel) of the satellite sensor originate from the contributions of various features (ice and water), forming a "hybrid pixel". The radiation value or backscatter coefficient of the picture element is not the value of pure sea water or pure sea ice, but a weighted average between the two. Thus, the gray or scattering characteristic change from open sea water to continuous sea ice is not a sharp step (jump), but a continuous, gradual transition. This gradual nature results in a gradual change in the gradient of the image in this region. The method for dividing by adopting the fixed threshold value faces fundamental dilemma in that when the threshold value is set too high, part of the mixed pixels are judged to be seawater, so that the extracted sea ice range is smaller and the edge is contracted inwards, and when the threshold value is set too low, the mixed pixels are judged to be sea ice, so that the sea ice range is larger and the edge is expanded outwards. Small variations in threshold selection can directly lead to significant swings in sea ice edge line position, introduce systematic errors and subjective uncertainties that are difficult to eliminate, and limit the accuracy of estimation of sea ice range. Aiming at the technical problems that in the sea ice edge extraction in the prior art, the real limit between sea ice and open sea water is difficult to accurately and stably position under the double influences of sea ice surface heterogeneity and edge mixed pixels, no effective solution is proposed at present. Disclosure of Invention The embodiment of the disclosure provides a satellite-based sea ice edge determining method, device and storage