A sea ice melt pond detection method based on ICESat-2 lidar point cloud data

Abstract

The application discloses a kind of sea ice melt pool detection methods based on ICESat-2 laser radar point cloud data, to solve the problem of accurate separation of double surface caused by low signal-to-noise ratio of photon in summer sea ice melt pool area.The method is segmented by data preprocessing to ICESat-2 photon data and generates sliding window to locate sea ice and melt pool height range;Using Gaussian fitting to divide the ice / under-ice photon set, combined with dynamic height difference threshold iteration and terrain adaptive elliptical density double threshold for noise removal;Based on the smooth height histogram of denoising photon, double-peak detection and saddle point analysis are carried out, and the height layer range of the top and bottom of the melt pool is dynamically divided.The application realizes the full-automatic accurate separation of double surface of sea ice melt pool in low signal-to-noise ratio environment through multi-level dynamic threshold optimization and terrain adaptive elliptical density calculation, and provides reliable technical support for global polar environment monitoring and climate change research.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing technology and environmental monitoring technology, and in particular to a method for detecting sea ice melt pools based on ICESat-2 lidar point cloud data. Background Technology

[0002] Sea ice is a sensitive indicator of global climate change, playing a crucial role in global heat balance, atmospheric circulation, ocean water cycle, and temperature-salinity balance. Affected by global warming, sea ice is undergoing significant melting, profoundly impacting Arctic shipping development, biodiversity survival, and climate system stability. Sea ice thickness monitoring has evolved over a century, progressing from single-point borehole measurements to a comprehensive space-air-ground observation system. Altimeter satellites (such as ICESat-2 and CryoSat-2) measure the height of the sea ice surface relative to sea level, combining this with hydrostatic models to estimate ice thickness, and have become the primary method for large-scale sea ice thickness monitoring. Currently, international data centers release satellite data products on sea ice thickness covering October to April of the following year; however, retrieving summer sea ice thickness still faces technical challenges, primarily due to the influence of melt pools developing during the summer on the identification of satellite altimeter echo signals.

[0003] Launched in 2018, the ICESat-2 laser altimeter satellite carries an advanced topographic laser altimeter system that emits 532nm green laser light with high penetration capability into water. However, due to limitations imposed by the instrument itself and the observation environment, ICESat-2 photon point cloud data contains considerable noise, necessitating denoising to extract signal photons. Current denoising methods for ICESat-2 photon point cloud data mainly fall into three categories: 1) Clustering methods, which separate signal from noise by analyzing the similarity of photon points. Typical clustering algorithms include density-based spatial clustering algorithms with noise and algorithms that identify cluster structures by ranking points. Much research has focused on improving the parameters of clustering algorithms to address the inherent characteristics of ICESat-2 point cloud data; however, clustering algorithms are highly sensitive to parameter variations. 2) Methods based on statistics and probability, which analyze the distribution patterns of photon point cloud data and remove noise based on photon distribution patterns or probabilities. Common methods include histogram analysis and Gaussian distribution fitting. These methods avoid subjective biases caused by manually setting thresholds, but they often face the problem of high computational cost when estimating parameters. 3) Based on photon geometric features, noise is usually identified by utilizing the spatial distribution and geometric features of point clouds. The most common method is density-based, which assumes that the densities of signal photons and noise photons are different, i.e., signal photons are more clustered, while noise photons are relatively dispersed. This method has great potential for denoising problems in cases of uneven point cloud density distribution.

[0004] Several algorithms exist for melting pool detection and depth measurement based on ICESat-2 photon point cloud data. The University of Maryland's sea ice ridge detection algorithm manually selects the melting pool region and extracts the elevations of the top and bottom of the melting pool based on the photon distribution histogram. Building upon this, an adaptive density-dimensional sea ice bifacial bifurcation detection algorithm achieves fully automatic tracking of the melting pool's two surfaces, simultaneously obtaining information on the melting pool's width and depth. However, the adaptive density-dimensional sea ice bifacial bifurcation detection algorithm only considers horizontal weights, ignoring topographic variations at the bottom of the melting pool. This leads to a mixture of signal and noise photons at the bottom, severely affecting the accurate determination of the melting pool's two surfaces. Furthermore, while this algorithm considers photon density differences along the track direction, it ignores the significant density differences between photons on the melting pool surface and at the bottom, thus failing to effectively handle noise within the melting pool and interfering with the determination of the two surfaces. To address these issues, a new ICESat-2 photon denoising method that considers both topographic variations and density inhomogeneities is needed. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by providing a method for detecting sea ice melt pools based on ICESat-2 lidar point cloud data.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for detecting sea ice melt pools based on ICESat-2 lidar point cloud data, characterized by the following steps:

[0008] Step 1: Divide the ICESat-2 global geolocation photon data into segments of regular size, generate a sliding window for each segment and count the histogram within the window, and preliminarily locate the height range that may contain sea ice melt pools based on the height of the interval with the highest count in the histogram.

[0009] Step 2: Bin the signal layer in each data segment and generate a height histogram. Use Gaussian fitting to fit the histogram to classify photons into two categories: above ice and below ice.

[0010] Step 3: Calculate the height difference of photons on adjacent ice surfaces and generate a histogram. After fitting the histogram using an exponential decay model, remove noisy photons through dynamic threshold iteration.

[0011] Step 4: Construct a search neighborhood using an adaptively rotated ellipse, and combine Gaussian weighted density calculation and local-global dual threshold filtering to remove noise photons under the ice;

[0012] Step 5: Identify the peak-saddle point relationship of the two surfaces based on the filtered signal photon height histogram, dynamically divide the height range of the top and bottom of the molten pool, and generate a schematic diagram of the two surfaces of the molten pool.

[0013] Further, step 1 includes:

[0014] Step 11: Divide the relative along-track distance in the ICESat-2 global geolocation photon data into segments according to regular size intervals;

[0015] Step 12: Generate a sliding window for each data segment and calculate a height histogram within each window;

[0016] Step 13: Find the interval with the highest count in the histogram, and the center of the interval is regarded as the height center;

[0017] Step 14: Generate a signal layer with the height center as the axis to initially locate the height range that may include sea ice topography and melt pool topography.

[0018] According to the method of claim 1, step 2 includes:

[0019] Step 21: Binn the signal layer of each data segment and generate the corresponding height histogram;

[0020] Step 22: Using the peak value of the height histogram as the center of the ice surface, perform Gaussian fitting on the histograms of the local areas above and below the ice surface;

[0021] Step 23: Use the parameters of the fitted Gaussian function to set a threshold, and use it as the boundary between the ice surface and the ice underside;

[0022] Step 24: The above-ice photon subset and below-ice photon subset of each sub-bin need to be merged separately to finally obtain two photon subsets of the signal layer in each segment of data: above-ice photon subset and below-ice photon subset.

[0023] Further, step 3 includes:

[0024] Step 31: Calculate the height difference between adjacent photons in the photon set on the ice and generate a histogram;

[0025] Step 32: Fit the attenuation model to the exponential decay distribution in the height difference histogram;

[0026] Step 33: With the constant term of the fitted decay function approaching zero as the goal, a dynamic threshold is set in the iteration and noise photons with large height differences but sparse distribution between adjacent photons are removed through cyclic iteration until the model converges and meets the termination condition.

[0027] Further, step 4 includes:

[0028] Step 41: Initialize the dynamic rotation ellipse for the photons under ice, and adjust the direction and length of the major axis according to the terrain slope;

[0029] Step 42: Optimize the maximum density direction by traversing the direction in stages, and combine the density check on both sides of the ellipse to eliminate the interference of noise on the slope calculation;

[0030] Step 43: Calculate the photon weighted density value within the ellipse using the Gaussian kernel function;

[0031] Step 44: Extract the photon subset of the signal under ice by using a dual filtering method of local maximum inter-class variance threshold and global quantile threshold.

[0032] Further, step 5 includes:

[0033] Step 51: Generate a height histogram for the signal photons and smooth the data using a binomial filter;

[0034] Step 52: Detect the double peaks and saddle points in the smoothed histogram to verify whether it is a double surface of the molten pool;

[0035] Step 53: Dynamically define the height range of the top and bottom of the melting pool based on the bimodal-saddle point position;

[0036] Step 54: Distribute the signal photons to the dual-surface aggregate according to the height range, and output a schematic diagram of the molten pool dual-surface. Further, the exponential decay model in step 3 is:

[0037] y = a*exp{-b*x}+c

[0038] The dynamic threshold iterates to bring the constant term c close to zero, and the noise determination condition is:

[0039] y valid ={y i ||y i -c|<10 -3}

[0040] Where x is the height difference, y is the frequency of the height difference histogram, and a, b, and c represent the amplitude, attenuation rate, and constant term of the fitted attenuation model, respectively. i The height difference d obtained statistically i Frequency, y valid This represents all conditions that satisfy |y i -c|<10 -3 The set of y values.

[0041] Furthermore, in step 4, the major semi-axis a′ of the ellipse is dynamically adjusted by the terrain slope θ, satisfying:

[0042]

[0043] The formula for determining photon distance within an elliptical domain is:

[0044]

[0045] Where θ represents the terrain slope, a ′ This represents the dynamic adjustment value of the semi-major axis of the ellipse under sloping terrain, where 'a' represents the initial value of the semi-major axis. `dist(p,q)` represents the rotation distance between any point `p` inside the ellipse and the center point `q`, with a rotation angle of θ and ΔX. θ and Δh θ Calculation via coordinate rotation:

[0046]

[0047] Among them, X p ,h p These represent the relative distance and elevation along the track from point p, respectively, and X. q ,h q q represents the relative distance along the track and the elevation value, respectively.

[0048] Furthermore, the local threshold segmentation in step 4 employs the maximum inter-class variance method, with the optimal threshold t. otsu satisfy:

[0049]

[0050] Where ω0,ω1 represent the category percentages, and μ0,μ1 represent the category means. The variance is between classes.

[0051] Furthermore, the binomial filter formula in step 5 is:

[0052] H new (x)=H(x-2)*0.0625+H(x-1)*0.25

[0053] +H(x)*0.375+H(x+1)*0.25+H(x+2)*0.0625

[0054] Where H new H is the height of the histogram after filtering, H is the height of the histogram before filtering, and x is the index of the height interval.

[0055] This invention also discloses a sea ice melt pool detection system, which can be used to implement the above-mentioned sea ice melt pool detection method, specifically including:

[0056] The data preprocessing module is configured to segment ICESat-2 global geolocation photon data into segments of regular size, generate sliding windows, and statistically analyze the height histogram within the windows to locate the height range that may include sea ice melt pools.

[0057] The photon point cloud classification module is configured to bin the signal layer of segmented data and generate a height histogram, and classify photons into ice-surface photon sets and ice-subsurface photon sets by Gaussian fitting.

[0058] The ice-based photon denoising module is configured to calculate the height difference between adjacent photons in the ice-based photon set and dynamically iterate to remove noisy photons based on an exponential decay model.

[0059] The under-ice photon denoising module is configured to construct a search neighborhood by adaptively rotating an ellipse, and combine Gaussian weighted density calculation and local-global dual threshold filtering to remove under-ice noise photons;

[0060] The melt pool detection module is configured to perform height histogram analysis on the denoised photons above and below the ice, dynamically divide the height range of the top and bottom of the melt pool based on the peak-saddle point relationship, and generate a schematic diagram of the dual surfaces of the melt pool.

[0061] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described sea ice melt pool detection method.

[0062] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described sea ice melt pool detection method.

[0063] Compared with the prior art, the advantages of the present invention are as follows:

[0064] 1. By employing a multi-level dynamic height difference threshold iterative optimization method and an elliptical density denoising technique that considers topographic variations, fully automated dual-surface detection of sea ice melt pools is achieved without human intervention. This method is adaptable to varying topographic complexity, noise distribution, and lighting conditions, and is suitable for large-scale data processing in global polar sea ice regions, significantly reducing labor costs and operational barriers.

[0065] 2. To address the characteristics of low signal-to-noise ratio and complex noise distribution in photon point clouds in summer melting pool areas, a dual filtering mechanism of dynamic height difference threshold iteration and elliptic density is used to effectively separate signal photons from noise photons.

[0066] 3. An elliptical search neighborhood technique based on dynamic adjustment of terrain slope, combined with Gaussian weighted density calculation, overcomes the shortcomings of traditional methods that ignore vertical density differences and terrain changes.

[0067] 4. An exponential decay model is adopted for fast convergence and a phased directional traversal strategy to reduce redundant calculations.

[0068] 5. Achieve precise separation and dynamic altitude range division of the two surfaces of the summer melting pool, breaking through the technical bottleneck that existing satellite data products are only applicable to winter. Attached Figure Description

[0069] Figure 1 This is a flowchart of the sea ice melt pool detection method according to an embodiment of the present invention.

[0070] Figure 2 This is a photon distribution map of an embodiment of the present invention; wherein (a) is the original photon distribution map of the 318000-319000m segment in the ICESat-2 global geolocation photon test data; and (b) is the total signal photon distribution map of the 318000-319000m segment data after layered denoising.

[0071] Figure 3 This is a schematic diagram of the double-surface molten pool according to an embodiment of the present invention; wherein (a) is Figure 2 (b) is the double-surface distribution map of the melt pool obtained after all signal photons are detected by the melt pool; (b) is the double-surface distribution map of the melt pool obtained after all signal photons are detected by the melt pool after the same 318000-319000m segmented data is denoised by the sea ice double-surface bifurcation detection algorithm based on adaptive density dimension. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0073] This invention provides a method for detecting sea ice melt pools based on ICESat-2 lidar point cloud data, comprising the following steps:

[0074] Step 1: Data preprocessing. Extract the track distance and height data from the target area file of the ICESat-2 global geolocation photon data to be used for melt pool detection, and segment it according to the size rules. Then, generate a sliding window and count the histogram within the window. Based on the height of the interval with the highest count in the histogram, preliminarily locate the height range that may contain sea ice melt pools.

[0075] Step 2: Bind the signal layer data from the initial location and generate a height histogram. Based on the assumption that the lateral distribution of photon clusters returned by ICESat-2 in flat areas is close to a Gaussian distribution, use formula (1) to simulate the quasi-Gaussian shape of photons in high-density regions. Data exceeding three standard deviations are considered low-density regions, i.e., photons are classified according to formula (2).

[0076]

[0077] Where, a is the amplitude of the fitted Gaussian function; μ is the mean of the fitted Gaussian peak; σ is the width of the fitted Gaussian peak; h ph Let be the ellipsoidal height value of the photon. The entire estimation process is calculated using the iterative nonlinear least squares method.

[0078] Step 3: Denoising of photons on ice based on the adaptive height difference threshold method. Calculate the height difference between all adjacent photons in the ice photon set obtained in Step 2 and generate a histogram for a fixed interval. Then, fit the histogram using an exponential decay model (Equation 3), aiming to make the constant term of the decay function approach zero. Set a dynamic threshold to iteratively remove noise photons with large height differences but sparse distribution between adjacent photons until the model converges. The dynamic threshold is determined based on the height difference corresponding to the highest frequency value of the y-value obtained from Equation (4).

[0079] y = a*exp{-b*x}+c (3)

[0080] y valid ={y i ||y i -c|<10 -3} (4)

[0081] Where x is the height difference, y is the frequency of the height difference histogram, and a, b, and c represent the amplitude, attenuation rate, and constant term of the fitted attenuation model, respectively. i The height difference d obtained statistically i Frequency, y valid This represents all conditions that satisfy |y i -c|<10 -3 The set of y values.

[0082] Step 4: Denoising of sub-ice photons based on the adaptive elliptical density threshold method. An adaptive rotation ellipse is defined for all photons p in the sub-ice photon set obtained in Step 2. First, the ellipse is initialized, with the initial major and minor axes set according to empirical values ​​for the sea ice region. The major axis a is dynamically determined by formula (5), while the minor axis b is a fixed value. A distance threshold is used to determine whether the remaining points q are within the elliptical domain of the current center point p; the specific process is shown in formulas (6-7).

[0083]

[0084] Where θ represents the terrain slope (tangent), a ′ This represents the dynamic adjustment value of the semi-major axis of the ellipse under sloping terrain, where 'a' represents the initial value of the semi-major axis. `dist(p,q)` represents the rotation distance between any point `p` inside the ellipse and the center point `q`, with a rotation angle of θ. θ and Δh θ It can be obtained from formula (8):

[0085]

[0086] Where X p ,h pThese represent the relative distance and elevation along the track from point p, respectively, and X. q ,h q q represents the relative distance along the track and the elevation value, respectively.

[0087] Then, the horizontal direction is traversed in stages. First, the direction from 0 to 180° is traversed with coarse angle intervals, and then the identification of the maximum density direction θ is optimized with fine angle intervals. In addition, the density check on both sides of the ellipse is combined to avoid noise interference with the slope calculation, so as to ensure that the maximum density direction θ truly reflects the terrain slope.

[0088] The photon density calculation process is as follows: Each photon p becomes the center point. A Gaussian kernel function is used to weight the photons within the current ellipse, including the center point. Finally, the weighted sum of all photons within the ellipse is defined as the density value of the center point of the ellipse.

[0089] The photon set under ice is binned, and the optimal threshold t is calculated for each bin using the Otsu's method. otsu Formulas 9-10 are used to separate high-density signals from low-density noise. Then, a global density threshold t is added. quantile The q quantile of the density value of the signal photons obtained initially is used as the global threshold to further eliminate discrete noise photons.

[0090]

[0091] Where ω0 represents the proportion of signal photons in the current bin, with a corresponding average photon density of μ0; ω1 represents the proportion of signal photons in the current bin, with a corresponding average photon density of μ1. The inter-class variance of the two classes is denoted as . When selecting the optimal threshold t otsu At this point, the inter-class variance reaches its maximum value.

[0092] Step 5: Combine the denoised photon sets above and below the ice into a final signal photon set. This set will be used in subsequent step 6.

[0093] Step 6: Molten Pool Detection. Data binning is performed on the final signal photon data to generate height histograms. A binomial filter (Equation 11) is applied to each histogram to smooth local fluctuations and enhance the dual-surface signal characteristics. Double peaks and saddle points between them are detected in the smoothed histograms. If double peaks exist, the peak values ​​are verified to meet the conditions; if they do, it is determined to be a dual-surface molten pool. Otherwise, a single peak is used to determine a single surface. If only a single peak exists, it is directly determined to be a single surface. Based on the detected double peak and saddle point positions in the smoothed height histogram, the height range of the top and bottom of the molten pool is dynamically defined. The final signal photons are dynamically allocated to two sets according to the height range of the dual surfaces, achieving precise division of the dual surfaces. For single surfaces, the signal photons are copied to both sets. A schematic diagram of the dual-surface molten pool is finally output based on the photons in the two sets.

[0094] H new (x)=H(x-2)*0.0625+H(x-1)*0.25

[0095] +H(x)*0.375+H(x+1)*0.25

[0096] +H(x+2)*0.0625(11)

[0097] Where H new H is the height of the histogram after filtering, H is the height of the histogram before filtering, and x is the index of the height interval.

[0098] This invention addresses a key challenge in current sea ice melt pool detection technology: denoising large-scale ICESat-2 photon point cloud data under complex photon distribution patterns. It also overcomes the problem of existing adaptive density threshold denoising methods ignoring uneven density distribution in the vertical direction and changes in sub-ice topography. The invention provides a sea ice melt pool detection method based on ICESat-2 lidar point cloud data.

[0099] The sea ice melt pool detection method based on ICESat-2 lidar point cloud data described in the above embodiments of the present invention fully considers the density non-uniformity and topographic features in the ICESat-2 photon point cloud data, enabling fully automated, large-scale, and high-volume detection and depth measurement of Arctic sea ice melt pools, and is suitable for low signal-to-noise ratio conditions. Furthermore, this method is highly operable and scalable, thus contributing to the production of satellite data products on global summer sea ice thickness.

[0100] In implementing this invention, ICESat-2 global geolocation photon data collected on June 22, 2019, was selected as the test data. This data covers an area from 79.94°N to 87.91°N latitude and spans from 86.13°W to 73.17°E longitude. Due to the large geographical area covered by this data, only typical segments of data were selected for the experiment. The specific implementation steps of this invention are as follows:

[0101] Step 1: In the segmented data of 318000-319000m, such as Figure 2 As shown in (a), a sliding window is used to statistically analyze height histograms at certain intervals, and then the interval with the highest count is found within each histogram. A signal layer is generated with the center of this height as the axis, with the aim of initially locating the height range that may include sea ice topography and melt pool topography.

[0102] Step 2: Use formulas (1) and (2) to classify the signal layer obtained in step 1 to obtain the set of photons above ice and the set of photons below ice.

[0103] Step 3: The ice-based photon subset obtained in Step 2 is denoised using formulas (3) and (4) based on an adaptive height difference threshold. The denoising results are as follows: Figure 2 As shown in (b).

[0104] Step 4: Using the photon set obtained in Step 2, calculate the Gaussian density values ​​of all photon points using formulas (5) to (8), and then separate high-density signal photons into low-density signal photons according to formulas (9) and (10). The denoising process for the photons under ice is as follows: Figure 2 As shown in (b).

[0105] Step 5: Combine the signal photons obtained in Step 3 and Step 4 to obtain the final signal photon set.

[0106] Step 6: Perform molten pool detection on the final signal photon set obtained in Step 5. (See schematic diagram of the molten pool's dual surface, e.g.) Figure 3 You can see the division of the two surfaces and the boundary identification of the molten pool.

[0107] To more clearly illustrate the effects of the present invention, this embodiment also conducted a melt pool detection experiment based on the classic sea ice dual-surface bifurcation detection algorithm. Figure 3 (a) is the result obtained based on the method proposed in this invention. Figure 3 (b) shows the results based on the classic sea ice bifacial bifurcation detection algorithm (non-melt pool areas are only displayed in blue, while in the melt pool area, blue represents the top and orange represents the bottom). Figure 3 (a) As can be seen, the photon reflection intensity processed by the method of this invention is significantly improved compared to the result processed by the sea ice bifacial bifurcation detection algorithm, while a large number of data holes appear in the result processed by the sea ice bifacial bifurcation detection algorithm. This indicates that under low signal-to-noise ratio conditions, the method proposed in this invention can more robustly capture terrain than the sea ice bifacial bifurcation detection algorithm, and retains more effective signal photons. Figure 3 (a)). Furthermore, in contrast... Figure 3 (a) and Figure 3 (b) It can be seen that the proposed method can more accurately capture the bottom topography of the melt pool and identify the melt pool boundary than the sea ice dual-surface bifurcation detection algorithm. It also improves the identification of ice-top topography (such as ice ridges) to a certain extent, and the topography continuity is stronger. As can be seen from the above examples, the method of the present invention is very effective for melt pool detection of ICESat-2 photon point cloud data and can handle low signal-to-noise ratio situations.

[0108] In another embodiment of the present invention, a sea ice melt pool detection system is provided. This system can be used to implement the above-described sea ice melt pool detection method, specifically including:

[0109] The data preprocessing module is configured to segment ICESat-2 global geolocation photon data into segments of regular size, generate sliding windows, and statistically analyze the height histogram within the windows to locate the height range that may include sea ice melt pools.

[0110] The photon point cloud classification module is configured to bin the signal layer of segmented data and generate a height histogram, and classify photons into ice-surface photon sets and ice-subsurface photon sets by Gaussian fitting.

[0111] The ice-based photon denoising module is configured to calculate the height difference between adjacent photons in the ice-based photon set and dynamically iterate to remove noisy photons based on an exponential decay model.

[0112] The under-ice photon denoising module is configured to construct a search neighborhood by adaptively rotating an ellipse, and combine Gaussian weighted density calculation and local-global dual threshold filtering to remove under-ice noise photons;

[0113] The melt pool detection module is configured to perform height histogram analysis on the denoised photons above and below the ice, dynamically divide the height range of the top and bottom of the melt pool based on the peak-saddle point relationship, and generate a schematic diagram of the dual surfaces of the melt pool.

[0114] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a sea ice melt pool detection method.

[0115] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0116] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the sea ice melt pool detection method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by a processor.

[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the implementation methods of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the present invention.

Claims

1. A method for detecting sea ice melt pools based on ICESat-2 lidar point cloud data, characterized in that, Includes the following steps: Step 1: Divide the ICESat-2 global geolocation photon data into segments of regular size, generate a sliding window for each segment and count the histogram within the window, and preliminarily locate the height range that may contain sea ice melt pools based on the height of the interval with the highest count in the histogram. Step 2: Bin the signal layer in each data segment and generate a height histogram. Use Gaussian fitting to fit the histogram to classify photons into two categories: above ice and below ice. Step 3: Calculate the height difference of photons on adjacent ice surfaces and generate a histogram. After fitting the histogram using an exponential decay model, remove noisy photons through dynamic threshold iteration. The exponential decay model is as follows: ， The dynamic threshold iterates to bring the constant term c close to zero, and the noise determination condition is: ， in, It's a height difference. The frequency of the height difference histogram. , , These represent the amplitude, attenuation rate, and constant term of the fitted attenuation model, respectively. The height difference obtained through statistics frequency, Represents all satisfaction of A set of values; Step 3 includes the following sub-steps: Step 31: Calculate the height difference between adjacent photons in the photon set on the ice and generate a histogram; Step 32: Fit the attenuation model to the exponential decay distribution in the height difference histogram; Step 33: With the constant term of the fitted decay function approaching zero as the goal, a dynamic threshold is set in the iteration and noise photons with large height differences but sparse distribution between adjacent photons are removed through cyclic iteration until the model converges and the termination condition is met. Step 4: Construct a search neighborhood using an adaptively rotated ellipse, and combine Gaussian weighted density calculation and local-global dual threshold filtering to remove noise photons under the ice; Step 5: Identify the peak-saddle point relationship of the two surfaces based on the filtered signal photon height histogram, dynamically divide the height range of the top and bottom of the molten pool, and generate a schematic diagram of the two surfaces of the molten pool.

2. The sea ice melt pool detection method according to claim 1, characterized in that: Step 1 includes: Step 11: Divide the relative along-track distance in the ICESat-2 global geolocation photon data into segments according to regular size intervals; Step 12: Generate a sliding window for each data segment and calculate a height histogram within each window; Step 13: Find the interval with the highest count in the histogram, and the center of the interval is regarded as the height center; Step 14: Generate a signal layer with the height center as the axis to initially locate the height range that may include sea ice topography and melt pool topography.

3. The method for detecting sea ice melt pools according to claim 1, characterized in that: Step 2 includes: Step 21: Binn the signal layer of each data segment and generate the corresponding height histogram; Step 22: Using the peak value of the height histogram as the center of the ice surface, perform Gaussian fitting on the histograms of the local areas above and below the ice surface; Step 23: Use the parameters of the fitted Gaussian function to set a threshold, and use it as the boundary between the ice surface and the ice underside; Step 24: The above-ice photon subset and below-ice photon subset of each sub-bin need to be merged separately to finally obtain two photon subsets of the signal layer in each segment of data: above-ice photon subset and below-ice photon subset.

4. The method for detecting sea ice melt pools according to claim 1, characterized in that: Step 4 includes: Step 41: Initialize the dynamic rotation ellipse for the photons under ice, and adjust the direction and length of the major axis according to the terrain slope; Step 42: Optimize the maximum density direction by traversing the direction in stages, and combine the density check on both sides of the ellipse to eliminate the interference of noise on the slope calculation; Step 43: Calculate the photon weighted density value within the ellipse using the Gaussian kernel function; Step 44: Extract the photon subset of the signal under ice by using a dual filtering method of local maximum inter-class variance threshold and global quantile threshold.

5. The method for detecting sea ice melt pools according to claim 1, characterized in that: Step 5 includes: Step 51: Generate a height histogram for the signal photons and smooth the data using a binomial filter; Step 52: Detect the double peaks and saddle points in the smoothed histogram to verify whether it is a double surface of the molten pool; Step 53: Dynamically define the height range of the top and bottom of the melting pool based on the bimodal-saddle point position; Step 54: Distribute the signal photons to the dual-surface set according to the height range, and output a schematic diagram of the molten pool dual-surface.

6. The method for detecting sea ice melt pools according to claim 4, characterized in that: In step 4, the semi-major axis a′ of the ellipse is dynamically adjusted by the terrain slope θ, satisfying: ， The formula for determining photon distance within an elliptical domain is: ， in, Indicates the slope of the terrain. This represents the dynamic adjustment value of the semi-major axis of the ellipse under sloping terrain. This represents the initial value of the semi-major axis of the ellipse. This represents the rotational distance between any point p inside the ellipse and the center point q of the ellipse, with the rotation angle being... , and Calculation via coordinate rotation: ， in, , Let P be the relative distance along the track and its elevation value, respectively. , q represents the relative distance along the track and the elevation value, respectively.

7. The method for detecting sea ice melt pools according to claim 6, characterized in that: The local threshold segmentation in step 4 uses the maximum inter-class variance method to determine the optimal threshold. satisfy: ， ， in , For category proportions, , The mean of the categories. .

8. The method for detecting sea ice melt pools according to claim 5, characterized in that: The formula for the binomial filter in step 5 is: ， ， in H is the height of the histogram after filtering, H is the height of the histogram before filtering, and x is the index of the height interval.

Images