A polar icebreaking planning method and system based on multi-source remote sensing spatio-temporal collaboration

Abstract

The present application provides a kind of polar ice breaking planning method and system based on multi-source remote sensing space-time cooperation, it is related to ship intelligent navigation and polar environment remote sensing monitoring technical field, the method of the present application includes: obtaining historical remote sensing image and extracting approximate rank measure feature and optical density feature, training sea ice feasible area discrimination model;The received delay satellite remote sensing image is input into sea ice feasible area discrimination model to obtain polar ice breaking feasibility score, constructs polar navigation resistance cost map and generates initial route;Based on thickness gradient field extraction optimal ice breaking starting point and optimal cut-in heading;Local image is registered to delay satellite remote sensing image to form overlap area, complete real-time absolute correction of overlap area;Extract the reference space-time drift vector of overlap area, calculate the extrapolated drift vector of non-overlapping area in combination with spatial attenuation weight function and perform reverse mapping, generate global space-time correction navigation resistance cost map, reconstruct local dynamic route and perform navigation.

Description

Technical Field

[0001] This invention relates to the field of intelligent ship navigation and polar environment remote sensing monitoring technology, specifically to a polar icebreaking planning method and system based on multi-source remote sensing spatiotemporal coordination. Background Technology

[0002] With the increasing normalization of polar scientific expeditions and commercial shipping routes, the safe and efficient navigation of icebreakers in ice-covered waters has become a crucial aspect of polar exploration. Polar seas are covered year-round by sea ice of various forms, and the ice distribution is highly dynamic, influenced by a combination of natural factors such as strong polar winds, ocean currents, and tides. To reduce navigation resistance in ice-covered areas, conserve fuel, and avoid structural damage, icebreakers must rely on precise environmental monitoring data for scientifically sound route planning before entering ice-covered regions.

[0003] Currently, polar environment monitoring and route planning primarily rely on spaceborne remote sensing technology to obtain large-scale ice condition data. However, due to the limited frequency of polar-orbiting satellite transits and communication link limitations in data downlink, analysis, processing, and distribution, satellite remote sensing images received by icebreakers typically have a time delay of several hours or even longer. Driven by complex polar hydro-meteorological conditions, the drift, compression, accumulation, and deformation of sea ice can be significant within a few hours. Routes planned based on time-lag satellite images often fail to accurately reflect the real-time ice conditions when ships actually arrive in the area, posing a potential risk of guiding ships into newly formed heavy ice zones or ice ridges. Furthermore, due to limitations in spatial resolution and the perennial cloud cover in polar regions, satellite remote sensing struggles to accurately identify small-scale interglacial lakes, crevasses, and other vulnerable areas conducive to navigation.

[0004] To compensate for the limitations of satellite remote sensing in terms of localized observation and real-time capabilities, some conventional technologies have attempted to incorporate shipborne UAVs or airborne radar for close-range reconnaissance. However, the polar environment is characterized by extreme cold, frequent strong gusts of wind, and anomalies in the geomagnetic field. Under these harsh polar weather conditions, frequent deployments of UAVs for large-area, long-duration continuous patrols not only face the risk of drastically reduced endurance due to a sharp drop in battery activity at extremely low temperatures, but also carry a very high risk of loss of flight control and equipment loss. Constrained by stringent limitations on takeoff and landing windows and safety redundancy, UAVs in real polar operational scenarios can only conduct very limited, short-duration, localized reconnaissance, and cannot be relied upon alone to undertake comprehensive, real-time updates of the large-scale flight environment.

[0005] On the other hand, the riskiest phase in polar icebreaking operations typically occurs the instant a vessel cuts through a thick ice floe from a clear water or thin ice zone. Existing course planning systems usually only provide the bridge with a sequence of track coordinates, generally lacking planning for the local attitude and mechanical environment at the icebreaking initiation point. In actual operations, if the icebreaker fails to accurately locate the weak points of the ice edge, or enters the ice edge at a non-orthogonal or non-optimal tangential heading angle, the bow is highly susceptible to severe lateral slippage upon contact with the hard sea ice. This uncontrollable slippage not only causes the vessel to deviate from the preset course and significantly reduces forward icebreaking efficiency, but also subjects the hull sides to enormous asymmetric instantaneous compressive stress, seriously threatening the structural safety of the vessel.

[0006] In summary, existing polar icebreaking navigation technologies inherently contradict the severe time lag in processing satellite remote sensing data and the strict limitations imposed by the extreme polar environment on local reconnaissance methods, making it difficult to obtain real-time ice conditions through high-frequency, large-area scanning. Furthermore, existing plans lack precise parameterized guidance on icebreaking initiation conditions when transitioning from route to local icebreaking operations, making it difficult to meet the actual needs of icebreakers for safe and efficient navigation in harsh polar sea conditions. Summary of the Invention

[0007] This invention provides a polar icebreaking planning method based on multi-source remote sensing spatiotemporal coordination, the method comprising:

[0008] Historical remote sensing images were acquired and approximate rank metric features and optical density features were extracted to train a sea ice feasible area discrimination model.

[0009] The received time-delayed satellite remote sensing images are input into the sea ice feasible area discrimination model to obtain the polar icebreaking feasibility score, construct a polar navigation resistance cost map and generate an initial route.

[0010] Trigger the shipborne UAV to acquire local images, and extract the optimal icebreaking starting point and optimal cutting direction based on the thickness gradient field;

[0011] The local image is registered to the time-lapse satellite remote sensing image to form an overlapping area. The features are reconstructed and input into the sea ice feasible area discrimination model to complete the real-time absolute correction of the overlapping area.

[0012] Extract the reference spatiotemporal drift vector of the overlapping area, calculate the extrapolated drift vector of the non-overlapping area by combining the spatial attenuation weight function and perform reverse mapping, generate a global spatiotemporal corrected navigation resistance cost map, reconstruct the local dynamic route and perform navigation.

[0013] Historical remote sensing images are divided into segments of size [size missing]. The remote sensing image patch is subjected to affine rotation transformation according to each angle in a preset set of rotation angles. The covariance matrix of the transformed grayscale matrix is ​​calculated, and eigenvalue decomposition is performed to obtain a descending sequence of eigenvalues. subscript Indicates the preset rotation direction;

[0014] Set preset energy threshold Calculate the number of eigenvalues ​​that satisfy the cumulative energy percentage requirement. The computational logic must meet the following requirements:

[0015]

[0016] Number of generated in each direction Arrange them sequentially to construct an approximate rank metric feature vector for the remote sensing image patch.

[0017] Constructing a polar navigation resistance cost map includes: using an exponential penalty mapping function to map the grid. Polar icebreaking feasibility score Converted into the cost of polar navigation resistance :

[0018]

[0019] In the formula, This is the set basic navigation energy consumption constant for clear water areas. The ice drag weighting coefficient is determined for the current icebreaker power rating. The drag nonlinear amplification constant; utilizing the polar navigation drag cost of all grids. Generate discrete maps of the resistance costs of polar navigation.

[0020] The acquisition of local images by the shipborne UAV is triggered using a dynamic triggering mechanism that incorporates communication time delay. Specifically, this includes calculating the communication time delay of the delayed satellite remote sensing image. Establish and communicate with time delay Positively correlated dynamic safe takeoff distance threshold :

[0021]

[0022] in, The basic safety braking distance constant for icebreakers. This is a scalar measure of the maximum sea ice drift speed in the current sea area based on historical statistics. This represents the icebreaker's current real-time speed. The mechanical warm-up and system initialization time required for the drone to take off from the time the system issues the command;

[0023] When the calculated distance between the ship's current physical coordinates and the boundary of the sudden ice change ahead is less than or equal to the dynamic safe takeoff distance threshold... At that time, a takeoff trigger command is generated.

[0024] After extracting the optimal icebreaking starting point based on the thickness gradient field, the optimal cut-in speed is calculated based on energy conservation. The specific calculation formula is as follows:

[0025]

[0026] in, For the current displacement quality of the icebreaker, This serves as the reference uniaxial compressive strength constant for polar sea ice. The geometric characteristic coefficient represents the current icebreaking efficiency of the bow hull of an icebreaker. This is the estimated absolute sea ice thickness corresponding to the optimal icebreaking starting point obtained through inversion.

[0027] Registering local images to time-lapse satellite remote sensing images to form overlapping areas includes: extracting the physical pose parameters of the UAV at the exposure time, including the absolute center geographic latitude. Absolute center geographical longitude Absolute flight altitude and the yaw angle of the aircraft For arbitrary pixel coordinates in a local image Based on the physical focal length constant of the camera pod With the Earth's average radius constant The true latitude is calculated using a high-latitude coordinate mapping equation that includes Earth curvature compensation. With actual geographical longitude :

[0028]

[0029]

[0030] The obtained latitude and longitude coordinates are back-projected to obtain the grid index to which the local image pixels belong, and the set of overlapping grids in the multi-source images to achieve spatial coverage is extracted.

[0031] Extracting the baseline spatiotemporal drift vector of the overlapping region includes: retrieving the historical polar icebreaking feasibility score matrix. Polar icebreaking feasibility score in overlapping areas generated in real time from local images Feasibility score matrix for breaking ice in historical polar regions Define the matching window in the middle, and define the two-dimensional translation offset variable as follows: Construct a spatiotemporal feature similarity evaluation function based on two-dimensional normalized cross-correlation. Solve for the spatiotemporal feature similarity evaluation function. The optimal pixel offset that yields the global maximum value is then defined as the reference spatiotemporal drift vector for the overlapping region. :

[0032]

[0033] The extrapolated drift vector of the non-overlapping region is calculated by combining the spatial decay weighting function, including: for any mesh to be corrected in the mesh set of the non-overlapping region. Calculate the distance from the plane Euclidean mesh to the geometric center of the overlapping region. Introducing the spatially relevant length constant of polar sea ice deformation. Construct a spatial decay weight function to characterize the uniform decay of ice volleyball movement. :

[0034]

[0035] The reference spacetime drift vector Multiply by the spatial decay weighting function This yields the extrapolated drift vector specific to the mesh to be corrected. The formula is:

[0036]

[0037] Perform a reverse mapping to generate a globally spatiotemporally corrected navigation resistance cost map, including: utilizing extrapolated drift vectors. Reverse mapping calculation of historical data source grid floating-point coordinates :

[0038]

[0039]

[0040] Historical Polar Icebreaking Feasibility Score Matrix The extrapolated and corrected feasibility score is calculated by performing bilinear spatial interpolation. Extrapolate and adjust the feasibility score. Substitute the physical travel cost into the exponential penalty mapping function to reconstruct the physical travel cost, and then stitch the extrapolated update data of the non-overlapping areas with the absolute correction data of the overlapping areas to generate a global spatiotemporal corrected navigation resistance cost map.

[0041] This invention also provides a polar icebreaking planning system based on multi-source remote sensing spatiotemporal coordination, the system comprising:

[0042] Offline model training module: acquires historical remote sensing images and extracts approximate rank metric features and optical density features to train a sea ice feasible area discrimination model;

[0043] Initial route generation module: Input the received delayed satellite remote sensing images into the sea ice feasible area discrimination model to obtain the polar icebreaking feasibility score, construct the polar navigation resistance cost map and generate the initial route;

[0044] Correction module: Triggers shipborne UAV to acquire local images, extracts the optimal icebreaking starting point and optimal cutting direction based on the thickness gradient field; registers the local image to the time-lapse satellite remote sensing image to form an overlapping area, reconstructs features and inputs them into the sea ice feasible area discrimination model to complete the real-time absolute correction of the overlapping area;

[0045] Reconstruction Module: Extracts the reference spatiotemporal drift vector of the overlapping area, calculates the extrapolated drift vector of the non-overlapping area by combining the spatial decay weight function and performs reverse mapping, generates a global spatiotemporal corrected navigation resistance cost map, reconstructs the local dynamic route and performs navigation.

[0046] This invention provides a polar icebreaking planning method and system based on multi-source remote sensing spatiotemporal collaboration. It effectively resolves the inherent contradiction between the severe communication delays in wide-area spaceborne remote sensing data in polar regions and the limitations of local reconnaissance methods such as UAVs under extreme weather conditions. Through deep collaboration between macroscopic delayed data and microscopic real-time data, this invention overcomes the limitation of traditional polar navigation relying solely on static, outdated maps. It achieves high-fidelity reconstruction of the polar ice-covered navigation environment, significantly improving the passage efficiency and operational safety of icebreakers under complex and dynamic ice conditions.

[0047] Within local polar sea areas, driven by co-originating wind fields and ocean currents, the drift motion of adjacent ice floes exhibits strong spatial autocorrelation. This invention utilizes local real-time high-fidelity images acquired by UAVs within extremely short safe flight windows, spatially mapping and cross-correlation matching these images with historical time-lapse satellite images to accurately extract the true physical drift vector of local sea ice during communication delays. Subsequently, fully leveraging the plastic deformation dissipation characteristics caused by the increased spatial distance of polar ice, a spatial consistency attenuation mechanism is introduced to radiate and extrapolate the extracted local true drift vector to the vast, undetected blind areas. This reverse mapping reconstruction enables the system to complete global physical drift correction of the entire macroscopic environmental map with several hours of time lag, using only extremely limited local aerial data, with zero additional take-off and landing risks and extremely low UAV power consumption, thereby eliminating the threat of hidden heavy ice areas caused by ocean currents.

[0048] The reconstructed UAV data directly updates the absolute ice conditions within the overlapping area, forming a high-confidence real-time safety view. Simultaneously, this invention deeply aligns with the mechanical operation scenario of ship icebreaking, enabling attitude and kinetic energy planning for the icebreaking initiation point. The system automatically identifies the weakest point of the ice edge through thickness gradient analysis and forcibly constrains the ship's bow to cut into the ice edge with a strictly orthogonal heading. This attitude control maximizes the conversion of the ship's longitudinal kinetic energy into vertically downward icebreaking shear force, preventing severe lateral slippage caused by tilting impacts against hard ice ridges, greatly avoiding asymmetric instantaneous compression damage to the ship's weak sidewalls, and ensuring structural safety for polar ocean operations.

[0049] Based on the latest time-delay-free resistance cost map corrected by global spatiotemporal extrapolation and precise entry attitude, the system can reconstruct local micro-dynamic routes in real time, guiding ships to smoothly pass through ice-blocking channels with the least real resistance at the current moment. Furthermore, during actual icebreaking propulsion, the system can simultaneously collect data on the mechanical energy consumed by the ship to overcome ice resistance, using this data as a feedback label with absolutely real physical properties to update the underlying sea ice feasible area discrimination model online. Attached Figure Description

[0050] Figure 1 This is a flowchart of the polar icebreaking planning based on multi-source remote sensing spatiotemporal coordination according to the present invention.

[0051] Figure 2 This experiment compares the cost of icebreaking resistance in polar operating environments. Detailed Implementation

[0052] This embodiment provides a polar icebreaking planning system based on multi-source remote sensing spatiotemporal collaboration. The system is applied to the ice-breaking navigation environment of icebreakers in polar seas (such as Prydz Bay in Antarctica and high-latitude shipping lanes in the Arctic). It focuses on hardware-level configuration and architecture design for physical scenarios such as high polar cold, year-round ice cover, limited satellite communication links, and high transmission latency.

[0053] The system's overall hardware architecture is deployed on the icebreaker itself and in the polar environment monitoring data center. It mainly includes a space-based remote sensing monitoring subsystem, a shipborne broadband communication subsystem, a shipborne unmanned aerial vehicle (UAV) reconnaissance subsystem, and a shipborne intelligent computing and planning subsystem. These subsystems are interconnected via wired and wireless networks, forming closed-loop data transmission links for both local and wide-area coverage.

[0054] The satellite-based remote sensing monitoring subsystem includes a data aggregation and processing server deployed in the polar environment monitoring network, and a remote sensing satellite constellation composed of multi-orbit polar-orbiting satellites and geostationary satellites. Preferably, the remote sensing satellite constellation includes MODIS satellites, Landsat satellites, and synthetic aperture radar satellites equipped with medium-resolution imaging spectrometers. The data aggregation and processing server receives large-scale raw polar remote sensing image data transmitted from the aforementioned satellite constellation, and performs format conversion and geographic coordinate projection conversion on the raw remote sensing image data to generate remote sensing images with time-stamped information and latitude and longitude coordinate information.

[0055] The shipborne broadband communication subsystem is deployed on the communication deck and bridge of the icebreaker, and is used to establish a data interaction link between the icebreaker and the satellite-based remote sensing monitoring subsystem. The specific deployment includes a shipborne satellite communication antenna array, a modem, and a local area network switch. The shipborne broadband communication subsystem employs paired carrier multiple access (PCMA) frequency reuse, combined with time division multiple access (TDMA) or code division multiple access (CDMA) multiple access methods. Preferably, the shipborne satellite communication antenna array uses a satellite terminal supporting the Inmarsat 5 Global Xpress (GX) standard, achieving global coverage signal reception in the Ka band via geostationary satellites, with a maximum uplink communication rate of 5 Mbps and a maximum downlink communication rate of 50 Mbps. The shipborne broadband communication subsystem periodically receives remote sensing images sent by the data aggregation and processing server. Due to objective limitations in data processing and polar communication transmission, the remote sensing images carry a communication delay greater than a preset time threshold when they reach the shipborne terminal.

[0056] The shipborne UAV reconnaissance subsystem is deployed on the flight deck or forward deck area of ​​the icebreaker, and includes a multi-rotor or fixed-wing UAV, an automatic landing pad, a radio telemetry and control data link terminal, and a high-precision positioning module. The UAV is equipped with a low-temperature adaptable battery pack with a heating and insulation module, and has a high-definition camera pod with visible light and infrared dual-light fusion mounted on its fuselage. The high-precision positioning module is configured as an RTK positioning receiver supporting dual-frequency GPS and BeiDou satellite navigation systems, used to simultaneously record six-degree-of-freedom pose data and precise latitude and longitude coordinates at the exposure time of each frame of aerial image. The radio telemetry and control data link terminal operates in the microwave band, used to establish a point-to-point high-speed local image transmission link between the icebreaker and the UAV, transmitting the local high-definition aerial images collected by the UAV back to the icebreaker in real time.

[0057] The shipborne intelligent computing and planning subsystem is located in the local computer room and bridge of the icebreaker, serving as the computing and display control hub for the system. This subsystem includes a shipborne workstation, a massive storage disk array, and a human-machine interface display terminal. Preferably, the shipborne workstation is equipped with computing nodes with at least 12GB of physical memory and pre-installed with a data mining and image processing engine supporting a Python environment. The shipborne workstation is physically connected to the modem of the shipborne broadband communication subsystem and the radio telemetry and control data link terminal via an internal Ethernet bus.

[0058] The shipborne workstation is internally equipped with independent image processing, feature extraction, and route solving units. The image processing unit is used to perform gridded segmentation and coordinate system registration mapping on satellite remote sensing images and local UAV aerial images; the feature extraction unit is used to perform multi-directional spectral decomposition on the registered grid image blocks to generate approximate rank metric feature vectors; the route solving unit embeds a support vector machine classifier and a heuristic optimization module, which is used to score the feasibility of ice areas based on the extracted feature vectors and generate dynamic route data containing the target waypoint sequence.

[0059] Furthermore, the shipborne intelligent computing and planning subsystem is also equipped with a standard NMEA 0183 / NMEA2000 industrial interface, physically connected to the icebreaker's integrated navigation system (INS) and electronic chart display and information system (ECDIS). The initial route, local dynamic route, and icebreaking starting point parameters output by the route solving unit are converted into standard navigation messages via the aforementioned industrial interface, directly injected into the electronic chart display and information system for route overlay display, and then sent to the icebreaker's autopilot and power control system for command execution.

[0060] Next, this embodiment details the training process of the sea ice feasible area discrimination model running in the polar intelligent computing and planning subsystem, based on the hardware system. Since polar seas are often shrouded in clouds and fog, and ice floes exhibit complex texture deformation and accumulation characteristics under the combined influence of ocean currents and strong polar winds, the training in this embodiment focuses on modeling and classifying polar optical attenuation and anisotropic fracturing textures of sea ice.

[0061] The training of the sea ice feasible region discrimination model based on historical remote sensing images includes the following steps performed in sequence:

[0062] Step S101: Acquire historical remote sensing images and perform label mapping and slicing processing;

[0063] A set of historical remote sensing images containing geographic coordinates and timestamps is obtained through a data aggregation and processing server. Addressing the issue that the long-term cloud cover in polar environments prevents optical satellite images from effectively penetrating thick cloud layers, this step, based on the principle of absolute safety during polar navigation, performs label reduction and reorganization: Thick ice areas marked manually are forcibly merged with thick cloud areas where the underlying ice condition cannot be determined, and both are labeled as Category 1 (defined as infeasible areas); thin ice areas are merged with thin cloud areas with high light transmittance, and labeled as Category 2 (defined as conditionally feasible areas); ice-free seawater areas are labeled as Category 3 (defined as absolutely feasible areas).

[0064] The historical remote sensing images with completed label mapping are segmented into a grid, and the extracted data is kept at a fixed size. The remote sensing image patch. In this embodiment, a pixel dimension constant is set based on the icebreaker's hull size and the computational load allocation for the search. .

[0065] Step S102: Extract multi-directional approximate rank metric features based on polar sea ice texture;

[0066] Because tidal forces and thermal expansion in polar regions cause multidirectional ice cracks and ridges to form within ice floes, a single-view image matrix cannot effectively capture such linear textures. This step performs multi-angle rotation and spectral decomposition on a single remote sensing image patch.

[0067] Define a set of preset rotation angles The total number of directions And the corresponding rotation angles are respectively The two-dimensional grayscale data of a single remote sensing image patch is subjected to an affine rotation transformation according to each angle in the preset rotation angle set to generate... A rotated grayscale matrix, labeled as subscript Represents a set The first in One rotation direction, and .

[0068] For each Calculate its covariance matrix and perform eigenvalue decomposition on it. Sort the calculated eigenvalues ​​in descending order of size, and denote the sequence as the eigenvalue sequence. The magnitude of the eigenvalues ​​directly reflects the principal component distribution of ice surface texture energy in the direction of rotation of the remote sensing image patch.

[0069] Set preset energy threshold In this embodiment, it is preferably set to Calculate the number of eigenvalues ​​that satisfy the cumulative energy percentage requirement. Its computational logic must meet the following requirements:

[0070]

[0071] By finding the smallest integer that makes the above inequality true... This maps a large grayscale matrix into a single digital quantity, which is defined as the approximate rank at the current angle. The approximate rank generated in each direction is arranged in order to construct the approximate rank metric feature vector of the remote sensing image patch. , expressed as:

[0072]

[0073] Step S103: Extract optical density features based on the extinction characteristics of polar clouds and snow;

[0074] The reflectivity and absorptivity of solar radiation vary significantly among ice layers of different thicknesses, ice ridges, and overhead clouds in the polar regions. To quantify these thickness differences as tractable features, optical density calculations are performed on all pixels within the remotely sensed image patch to generate an optical density matrix. .

[0075] Define the optical density matrix any element in The calculation formula is:

[0076]

[0077] in, For remote sensing image patches in spatial coordinates The pixel grayscale value at that location; The relative incident light intensity constant is set to the maximum pixel grayscale value in the current historical remote sensing image.

[0078] Calculate the optical density matrix respectively The statistical mean and variance were used to extract the mean optical density. and optical density variance Mean optical density The optical density variance represents the overall average ice thickness within the current image patch grid. This indicates the degree of dramatic fluctuations in local thickness caused by ice pack compression.

[0079] The approximate rank metric eigenvector obtained in step S102 With the mean optical density The optical density variance Perform dimension concatenation to generate a complete feature vector for model input. Its form is expressed as:

[0080]

[0081] Step S104: Train the sea ice feasible region discrimination model based on support vector machine.

[0082] Obtain the complete feature vectors of all historical remote sensing image patches. The global training set is constructed using the category labels of the first, second, or third category generated by the mapping in step S101.

[0083] The amount of sampled data from ice-free seawater or large areas of thin ice (absolutely feasible and conditionally feasible regions) is usually much larger than that from thick ice deposits or ice ridges (infeasible regions) under specific environments, resulting in an extremely imbalanced data distribution in the training set. To prevent the classification decision boundary from shifting towards the majority class, this step uses a support vector machine equipped with a radial basis function kernel as the core classifier to achieve nonlinear mapping and large-margin classification in the high-dimensional feature space.

[0084] The global training set is input into the support vector machine for iterative optimization training. During training, the optimal classification hyperplane is found and key support vector samples that determine the decision boundary position are extracted, establishing a non-linear mapping relationship between feature vectors and class labels.

[0085] After the classifier training converges, a probability mapping function is further invoked to transform the geometric distance of the decision boundary output by the classifier into a probability value interval. Based on this, for any input feature vector, the model can output a probability value interval strictly within the specified range. Polar icebreaking feasibility score for continuous intervals The closer the score is to 1, the higher the probability that the corresponding grid area is thin ice or clear water, and the higher the safety of ships passing through ice.

[0086] Next, this embodiment details the process of feasible area delineation and initial route generation based on time-lapse satellite imagery. Due to the limited transit windows of polar low-Earth orbit satellites and the limited bandwidth of downlink data and maritime satellite relay communication, the remote sensing images received by the ship objectively exhibit significant time lag. Furthermore, polar icebreakers are large vessels, and there is a significant nonlinear mechanical difference between straight-line resistance and turning resistance within ice-covered areas. The specific steps of this embodiment are as follows:

[0087] Step S201: Receive delayed satellite remote sensing images and perform global grid feature extraction;

[0088] The shipborne broadband communication subsystem receives large-scale raw remote sensing images of the polar regions from the data aggregation and processing server. It also records the icebreaker's current local real-time system time. It also analyzes the satellite Earth observation imaging time contained in the remote sensing image metadata. Calculate the communication delay of the image. :

[0089]

[0090] In real polar ocean operations, this It usually lasts for several hours.

[0091] The shipborne intelligent computing and planning subsystem transforms the received time-lapse satellite remote sensing images into a coordinate projection system and divides them into rows. The number of columns is The grid matrix. Define any grid index currently being processed as... ,in and For each grid in the grid matrix Calculate its multi-directional approximate rank metric features and optical density features, and generate corresponding meshes. Complete feature vector .

[0092] Step S202: Construct a polar navigation resistance cost map based on the discriminant model;

[0093] Complete feature vectors of all grids Each input is fed into the sea ice feasible region discrimination model, which is in standby mode, and the corresponding polar icebreaking feasibility score is calculated and output. .

[0094] To transform the probabilistic safety factor into a mechanical cost constraint guiding path planning, the polar icebreaking feasibility score needs to be converted into a polar navigation resistance cost. Considering the nonlinear, sharp increase in the propulsion drag of icebreakers due to polar sea ice thickness—that is, when the feasibility score drops to a certain level, the accumulated thick ice ridges will cause the hull to become completely stuck—an exponential penalty mapping function is used to construct the drag cost:

[0095]

[0096] in, The set basic navigation energy consumption constant for clear water areas; This refers to the ice resistance weighting coefficient calibrated for the current icebreaker power. is the resistance nonlinear amplification constant.

[0097] In addition, set safe passage thresholds. When it exists At that time, force the grid to be Assigning a value of infinity indicates that the region was already an absolutely infeasible area at the time of its historical generation (such as the edge of a glacier or an area of ​​extremely thick heavy ice), forcing the planning algorithm to avoid such grids. After traversing all grids, a discrete map of polar navigation resistance costs is generated.

[0098] Step S203: Optimize the initial route by incorporating information on the limited maneuverability of polar vessels;

[0099] Extract the grid corresponding to the current position of the icebreaker as the starting node. Extract the grid corresponding to the endpoint coordinates set in the driver's cab as the target node. A heuristic A* optimization algorithm is used to perform node path search on the polar navigation resistance cost map.

[0100] Icebreakers are designed with their bows specifically for cutting ice head-on, while their sidewalls have relatively weak resistance to compression, and their turning radius is extremely large in polar ice. Therefore, frequent, high-angle turns not only result in significant kinetic energy loss but also greatly increase the risk of the ship getting stuck. Based on this constraint, a path accumulation cost is constructed. and heuristic prediction costs Total agency value assessment function .

[0101] When starting from the current search node to adjacent candidate nodes During expansion, the actual path cumulative cost The recursive formula is defined as follows:

[0102]

[0103] in, The turning penalty cost for inter-mesh connections is calculated using the following formula:

[0104]

[0105] in, The constant for the steering penalty coefficient of large polar ships; For the icebreaker to sail from the previous node to the current search node The course, and the direction from the current search node Drive to the adjacent alternative node The scalar of the spatial deflection angle between the expected headings.

[0106] Heuristic prediction cost The calculation formula is:

[0107]

[0108] In the formula, Adjacent candidate nodes With the target node The Euclidean distance between them.

[0109] The optimization algorithm maintains a priority queue and continuously expands it to improve the overall agent value evaluation function. Starting from the smallest node, expand until the target node is reached. Then, by tracing back to the parent node, a grid node connection sequence with the minimum global resistance cost is extracted. This grid node connection sequence is then used to generate the initial flight path. Its mathematical expression is a continuous set of coordinates:

[0110]

[0111] The initial route The large areas of heavy icing within the field of view of the time-lapse satellite remote sensing image were initially avoided, but due to the aforementioned communication delay... The existence of ice conditions ahead of the route still presents a high degree of uncertainty, requiring subsequent shipborne UAV reconnaissance subsystems to perform local spatiotemporal feature corrections and navigation parameter reconstruction.

[0112] Next, this embodiment is based on the initial route generated above. This document details the specific calculation process for trigger-based local data acquisition and icebreaking initiation point parameter extraction using a shipborne UAV. In polar physics scenarios, extremely low temperatures drastically reduce the discharge activity of UAV batteries, making all-weather, large-area routine patrols impossible. Simultaneously, when an icebreaker cuts into thicker ice floes from clear water or thin ice, if the cutting angle deviates from the ice edge normal, severe lateral slippage can easily occur, causing pressure on the ship's sides and deviation from its intended course. Therefore, a dynamic triggering mechanism is needed to precisely launch the UAV and calculate the mechanical attitude parameters for icebreaking initiation based on the acquired high-resolution local images. The specific steps include the following sequential execution:

[0113] Step S301: Dynamic triggering determination of UAV based on flight path and communication time delay;

[0114] To conserve the power of drones in extremely cold conditions, the icebreaker followed the initial route. During navigation, the shipborne intelligent computing and planning subsystem calculates the ship's current physical coordinates in real time. Distance of the flight path along the boundary of the sudden change in ice conditions ahead of the route .

[0115] Define the boundary node of ice condition change In order to be in The first sequence to meet its polar icebreaking feasibility score. The grid nodes, where This is the set critical feasibility constant for the thick ice region.

[0116] Considering the satellite image communication time delay calculated in step S201 This can lead to uncertain drift in sea ice position, necessitating the establishment of a dynamic safe takeoff distance threshold that is positively correlated with time lag. :

[0117]

[0118] in, This is the basic safety braking distance constant for icebreakers; This is a scalar representing the maximum sea ice drift velocity in the current sea area based on historical statistics. This is the icebreaker's current real-time speed. This refers to the mechanical warm-up and system initialization time required for the drone to take off from the time the system issues the command.

[0119] When the system detects Immediately upon activation, a takeoff trigger command is generated, driving the shipborne UAV on the flight deck to take off via the radio telemetry and control data link terminal, and then proceeding along... The forward course is conducting close-range reconnaissance.

[0120] Step S302: Local image acquisition and ice thickness calculation by UAV;

[0121] Shipborne drones fly to After traversing the local airspace, a high-definition camera pod equipped with a visible light and infrared dual-light fusion system, mounted on the fuselage of the aircraft, is used to perform downward orthogonal imaging to acquire a local high-resolution image matrix. The pixel coordinates of the local image are defined as follows: .

[0122] Because the heat exchange between the thin polar ice layer and the relatively warmer seawater underneath is more intense, its surface infrared radiation temperature is significantly higher than that of the thick deposited ice. Using infrared radiation thermal field data transmitted back by UAVs, combined with a polar ice-water thermodynamic conduction model, the infrared images were converted into a local sea ice thickness field matrix. .matrix any element in This represents the estimated absolute sea ice thickness at the corresponding coordinate point.

[0123] Step S303: Extraction of the optimal ice-breaking starting point based on the thickness gradient field;

[0124] To avoid direct impact between icebreakers and strong, hard ice ridges, it is necessary to locate the thinnest point or a localized weak point with ice cracks in the ice-water interface as the icebreaking starting point. The physical field matrix for this local sea ice thickness... Calculate the two-dimensional spatial gradient to obtain the thickness gradient field. :

[0125]

[0126] Calculate the magnitude of the gradient vector. Extracting the desired result All pixels constitute the geometric point set of the ice water edge. ,in This is a preset thickness gradient threshold used to distinguish the ice-water boundary.

[0127] For sets For all candidate edge points, construct an objective function for optimizing the starting point that considers ice thickness drag and route deviation penalty. :

[0128]

[0129] in, and These are the weights for ice thickness drag penalty and route deviation penalty, respectively. Let be the local Euclidean distance from the candidate edge point to the initial macroscopic flight path projection line. Traverse the set. , so that the objective function The pixel coordinates that yield the minimum value are the optimal ice-breaking starting points. This point represents a physical node that is both close to the planned flight path and located in the area where the local ice conditions are most vulnerable.

[0130] Step S304: Calculation of cutting attitude and kinetic energy parameters;

[0131] To ensure that the icebreaker's bow hits the optimal icebreaking starting point. To maximize the conversion of longitudinal thrust into vertically downward icebreaking shear force and avoid dangerous lateral slippage, the icebreaker's physical heading must be strictly perpendicular to the local tangent of the ice edge.

[0132] Using the gradient vector at the optimal icebreaking starting point obtained in step S303 The positive direction of the thickness gradient is always perpendicular to the isothility line (i.e., the ice edge tangent) and points inward into the ice layer. Therefore, the optimal tangent heading angle... The definition is directly equivalent to the azimuth angle of the thickness gradient vector at that point:

[0133]

[0134] Furthermore, to ensure that the icebreaker's kinetic energy at the moment of entry is sufficient to crush the ice layer of the corresponding thickness, while avoiding excessive impact overload that could damage the hull structure due to excessive speed, an initial icebreaking kinetic energy matching equation based on energy conservation is established:

[0135]

[0136] in, The current displacement quality of the icebreaker; The optimal entry speed is to be determined. This serves as the baseline uniaxial compressive strength constant for polar sea ice. This represents the geometric characteristic coefficient of the current icebreaking efficiency of the bow hull of an icebreaker. In the practical engineering application of this embodiment, the mechanical properties of polar sea ice are strongly influenced by the volume of brine and extremely low temperatures. The reference uniaxial compressive strength constant of the polar sea ice is... The specific settings are as follows: (i.e., 2.5 MPa), this value represents the conservative upper limit of compressive strength for hard ice deposits in Prydz Bay, Antarctica, and typical Arctic shipping routes, ensuring that the ship's kinetic energy is sufficient to cope with the worst ice conditions; the geometric characteristic coefficient of the icebreaker's bow hull icebreaking efficiency. The geometric characteristic coefficients are mainly determined by the bow angle and hull friction coefficient of the icebreaker. Taking the hull design of typical polar icebreakers such as the Xue Long as an example, this embodiment will use the aforementioned geometric characteristic coefficients. Specifically calibrated as a constant .

[0137] By rearranging and taking the square root of the above equation, the optimal cut-in speed can be calculated and output. :

[0138]

[0139] At this point, the shipborne intelligent computing and planning subsystem has successfully extracted and output the core control parameter set for icebreaking initiation. This parameter set will be directly sent to the shipborne integrated navigation system to guide the icebreaker to intervene in the thick ice area with the best safe attitude and power. At the same time, with the continuous transmission of local high-resolution images from the UAV, the system will enter the spatiotemporal collaborative correction stage for the delay error of the remote sensing images.

[0140] This embodiment, based on the aforementioned generated icebreaking initiation control parameters and UAV local acquisition data, details the process of multi-source image spatiotemporal registration and real-time correction of overlapping areas. In polar ice condition monitoring, time-lapse satellite remote sensing images suffer from communication delays. Furthermore, polar ocean currents and strong winds cause significant sea ice drift, rendering the ice ridge or crack locations recorded in the grid ineffective in representing the current real-world environment. Simultaneously, the extremely high spatial resolution of local high-resolution UAV images makes direct comparison with coarse-resolution satellite grids impossible. Therefore, a rigorous high-latitude coordinate mapping and scale alignment algorithm must be constructed to achieve real-time data fusion and absolute correction.

[0141] Step S401: Coordinate registration based on high-latitude Earth curvature compensation;

[0142] Because the polar regions are located at high latitudes, meridians converge sharply, causing significant distortion errors when using conventional planar geometric projections. To accurately fit the local high-resolution images acquired by the UAV into the grid matrix containing the time-lapse satellite remote sensing images, the shipborne intelligent computing and planning subsystem needs to extract the real-time six-degree-of-freedom pose parameters synchronously recorded by the UAV's high-precision positioning module.

[0143] The physical pose parameters of the drone at the moment of exposure include: absolute center geographic latitude. Absolute center geographical longitude Absolute flight altitude and the yaw angle of the aircraft The physical focal length constant of the camera pod is set to... The Earth's average radius constant is For arbitrary pixel coordinates in a local high-resolution image. Construct a high-latitude coordinate mapping equation that includes Earth curvature compensation, and calculate the true latitude corresponding to the pixel. With actual geographical longitude :

[0144]

[0145]

[0146] The obtained latitude and longitude coordinates are then projected back into the grid matrix coordinate system defined by the time-lapse satellite remote sensing image. Let the geographic latitude of the reference origin at the upper left corner of the time-lapse satellite remote sensing image be... Geographic longitude is And the preset grid space span resolution is Then any drone image pixel The grid index to which it belongs The calculation formula is:

[0147]

[0148]

[0149] In the formula, This indicates rounding down. It iterates through all pixels in the local high-resolution image, indexing all grid points that produce valid mappings. Perform a union operation to extract a set of overlapping grid areas from the multi-source images to achieve spatial coverage. .

[0150] Step S402: High-resolution feature reconstruction of the grid under scale alignment;

[0151] For the overlapping region mesh set Each covered grid in At this time, a large amount of high-resolution pixel data from the drone has accumulated inside. Because the sea ice feasible area discrimination model in standby mode is based on a fixed size... The remote sensing image patches are partitioned into support vector space. If high-resolution images are directly input, the matrix dimension will collapse.

[0152] To address this scale gap issue, this step involves each covered grid cell. Within the boundary, perform bilinear interpolation resampling on all drone pixels it contains, forcibly downsampling and normalizing its pixel dimensions to the standard. In dimensional space, generate resampled high-resolution image patches.

[0153] The feature extraction algorithm chain is invoked to perform multi-angle rotation and spectral decomposition on the resampled high-resolution image patch, extracting the approximate rank metric feature vector of the UAV based on the current real ice layer physical state. Simultaneously perform an assessment of the extinction characteristics of polar clouds and snow, and calculate the corresponding average optical density of the UAV. With the variance of optical density of drones The above features are concatenated dimensionally to generate a reconstructed complete feature vector specifically for real-time correction. :

[0154]

[0155] Step S403: Absolute confidence correction based on the latest environmental characteristics;

[0156] Due to the reconstruction of the complete feature vector The real-time aerial photography from drones completely avoids communication delays. This introduces errors in ice condition evolution. The overlapping area grid set... Reconstructing the complete feature vector of all grids in the middle Each value is re-input into the sea ice feasible region discrimination model running in the shipborne intelligent computing and planning subsystem. After high-dimensional mapping using radial basis function kernel function and decision function calculation, the model outputs a real-time polar icebreaking feasibility score representing the current absolutely realistic physical environment. .

[0157] Using this real-time calculation result, the original polar icebreaking feasibility score carrying time delay error in step S202 is forcibly overwritten and replaced in global memory. The results are then substituted into the polar navigation resistance cost index penalty mapping function to calculate the corrected polar navigation resistance cost. :

[0158]

[0159] Through the aforementioned calculations and data overwriting, this system has created a real-time data security window with extremely high spatiotemporal fidelity within the limited physical area covered by the UAV's field of view on the flight cost map, thus completing the real-time absolute correction of the overlapping area. Subsequently, based on this overlapping area correction result, the system will further calculate the extrapolation of the spatiotemporal motion field of the uncovered area.

[0160] This embodiment details the calculation process for extrapolating and correcting non-overlapping areas based on local motion vectors, using the generated overlapping area correction results. In polar operating environments, the range and field of view of UAVs are limited, resulting in a limited set of overlapping area meshes. It occupies only a very small portion of the global map. For non-overlapping areas outside the drone's field of view, due to communication delays... The presence of polar wind stress and surface ocean currents has driven significant sea ice drift. If icebreakers directly enter unrenewed areas, they are likely to encounter thick ice ridges pushed by wind currents, leading to severe ice entrapment or hull damage. Considering the strong spatial autocorrelation of the drift of large-scale polar ice floes within local sea areas—that is, the motion vectors of adjacent ice floes are highly consistent, and their motion consistency exhibits plastic dissipation with increasing distance—this system employs motion field extrapolation and reconstruction to address the aforementioned blind zone problem.

[0161] Step S501: Extraction of drift vectors in overlapping regions based on spatiotemporal cross-correlation matching;

[0162] In order to obtain communication delay Within the time window, the actual physical displacement of polar sea ice is determined by the shipborne intelligent computing and planning subsystem, which retrieves two sets of data from memory: one set is the historical polar icebreaking feasibility score matrix before it is forcibly overwritten by step S403, defined as... One set represents the distribution of ice condition characteristics before the time lag; the other set is the polar icebreaking feasibility score of the overlapping area generated in real time by the UAV in step S403, defined as... This represents the current, absolutely true distribution of ice condition characteristics.

[0163] exist A matching window is defined in the matrix, centered on the overlapping area and extending outwards with a preset search radius. The two-dimensional translation offset variable is defined as follows. Construct a spatiotemporal feature similarity evaluation function based on two-dimensional normalized cross-correlation. :

[0164]

[0165] in, This is the arithmetic mean of the real-time scores of all grid drones within the overlapping area; This is the arithmetic mean of the historical scores within the corresponding offset window.

[0166] By traversing the search region, the optimal pixel offset that makes the above relevance evaluation function achieve its global maximum value is found:

[0167]

[0168] This optimal offset represents the local ice floe, including the UAV's field of view, in the past The overall rigid body translation within a time period. This is defined as the reference spatiotemporal drift vector of the overlapping region. :

[0169]

[0170] Step S502: Spatial autocorrelation drift field extrapolation;

[0171] Polar ice sheets are not perfectly rigid bodies. The farther the sea ice is from the known overlapping area, the greater the disturbance it is caused by local sea surface uplift or eddies, leading to its drifting motion... A deviation occurs.

[0172] Define the set of meshes in the global mesh after removing overlapping regions. The remaining mesh set is the set of non-overlapping mesh areas. .

[0173] Calculate the mesh set of the overlapping region Geometric center coordinates :

[0174]

[0175]

[0176] in, This represents the total number of grids contained in the overlapping area.

[0177] For non-overlapping region mesh sets Any grid to be corrected in Calculate the distance from the geometric center to the planar Euclidean mesh. :

[0178]

[0179] Constructing a spatial decay weighting function to characterize the uniform decay of ice volleyball movement :

[0180]

[0181] in, This is the spatially relevant length constant for polar sea ice deformation. Its value reflects the density of ice floes in the current polar season. The denser the ice surface, the farther the deformation is transmitted, and the larger this value is set.

[0182] The reference spacetime drift vector Multiplying by the aforementioned spatial attenuation weighting function, the extrapolated drift vector specific to the non-overlapping region mesh is calculated. :

[0183]

[0184] Step S503: Reconstruction of resistance costs based on reverse mapping of historical ice condition characteristics;

[0185] After obtaining the grid-level extrapolated drift vector, a reverse mapping must be performed to update the current real-world map. Logically, the current state is within the grid. The sea ice inside is actually sea ice that drifted from the past, specifically from the upstream offset position at the time of satellite imaging.

[0186] against Any grid to be corrected in The extrapolation drift vector is used to calculate the floating-point coordinates of the historical data source grid. :

[0187]

[0188]

[0189] Because ocean currents have non-integer physical properties, the reverse calculation is required. The values ​​are floating-point numbers. The historical polar icebreaking feasibility score matrix... In the process, four adjacent integer grid points enclosing the floating-point coordinates are selected, and bilinear spatial interpolation is performed to calculate the extrapolated and corrected feasibility score. This score eliminates the misalignment of ice conditions caused by communication delays.

[0190] Will Substituting this into the polar navigation resistance cost exponential penalty mapping function, the physical passage cost of the non-overlapping grid is recalculated and written into the system update register:

[0191]

[0192] By traversal The process involves processing all grids and performing the aforementioned reverse mapping. The extrapolated update data for non-overlapping areas is then globally stitched together with the real-time absolute correction data for overlapping areas generated in step S403, successfully generating a fully covered global spatiotemporally corrected navigation resistance cost map. This map is subsequently fed into the route solving unit to trigger an emergency reconstruction of the icebreaker's local dynamic route.

[0193] This embodiment, based on the aforementioned output global spatiotemporally corrected navigation resistance cost map and icebreaking initiation core control parameters, details the final implementation process of dynamic route reconstruction and navigation execution. In polar ice navigating, the actual ice conditions after spatiotemporal offset correction often deviate significantly from satellite observations from several hours prior. The original initial macroscopic route has likely already crossed newly formed ice ridges or high-resistance accumulation zones. Therefore, it is necessary to reconstruct the local path based on the latest environmental map without time lag and control the ship's physical engagement attitude to ensure absolute safety during icebreaking operations. Specifically, the following steps are executed sequentially:

[0194] The shipborne intelligent computing and planning subsystem extracts the latest generated global spatiotemporal corrected navigation resistance cost map and retrieves the optimal icebreaking starting point, optimal entry course, and optimal entry speed calculated in the aforementioned steps.

[0195] The system uses the optimal icebreaking starting point as the new local planning starting point and the safe node or final target node located inside the ice zone in the original initial macro-path as the local planning ending point. During the path re-optimization process, an initial attitude constraint is forcibly applied, requiring that the tangent direction of the first planned path segment starting from the starting point must be strictly equal to the optimal cutting-in heading, thereby eliminating the risk of the ship cutting into the ice edge at an inclined angle at the algorithm level.

[0196] Under the premise of satisfying the aforementioned mandatory initial attitude constraints, a heuristic optimization algorithm is used to re-expand nodes on the global spatiotemporally corrected navigation resistance cost map. Since the map has now undergone absolute correction for overlapping areas and extrapolation correction for drift vectors in non-overlapping areas, it can avoid hidden heavy ice areas brought by ocean currents and make full use of newly formed small interglacial lakes or weak ice belts to generate a continuous grid node sequence with minimum resistance and smoothest turning angle. This updated local safe path sequence generates the local dynamic flight path.

[0197] After completing the local dynamic route reconstruction, the shipborne intelligent computing and planning subsystem encodes the local dynamic route, the precise geographical coordinates of the optimal icebreaking starting point, the optimal entry heading, and the optimal entry speed into a standard industrial navigation message.

[0198] The navigation message is directly injected into the icebreaker's Electronic Chart Display and Information System (ECDIS) and Integrated Navigation System (INS) via the ship's internal Ethernet bus. The electronic chart display terminal overlays the updated local dynamic route on the bridge screen in a highlighted form, while simultaneously erasing dangerous sections of the invalid initial route.

[0199] Simultaneously, navigation control commands are sent to the icebreaker's autopilot and propulsion control system. Guided by the autopilot, the icebreaker approaches the target ice edge at a constant optimal cutting speed, strictly maintaining the bow direction aligned with the optimal cutting course. At the moment the bow contacts the hard ice, the ship's longitudinal kinetic energy is perfectly converted into a vertically downward icebreaking shear force through an orthogonal force-bearing posture, completely avoiding sidewall damage caused by lateral slippage. After cutting into the ice floe, the icebreaker continuously icebreaks along the local dynamic route of least resistance within the ice zone.

[0200] As the icebreaker advances along a local dynamic route, the draft deformation sensor on the bottom of the icebreaker, the shaft power monitor of the power system, and the hull vibration sensor collect the actual mechanical energy consumed by the ship to overcome the resistance of sea ice in real time, and convert it into real physical navigation resistance data.

[0201] The shipborne intelligent computing and planning subsystem matches and packages the real physical navigation resistance data with the spatiotemporally corrected feature vectors of the currently traversed grid, creating a verification sample with real mechanical labels, which is then transmitted back to the sea ice feasible area discrimination model in real time. Addressing the challenge of exhaustively exhaustive polar environment samples, the system employs a positive and unlabeled sample learning (PU learning) mechanism. Utilizing the absolutely real physical feedback obtained from actual shipboard sea ice compaction, it fine-tunes the decision boundaries and penalty weights of the support vector machine classifier online. Through this closed-loop linkage between engineering physics and algorithmic models, the discrimination model's assessment of complex ice conditions in polar regions becomes increasingly accurate, providing more robust and reliable algorithmic support for icebreaking planning in subsequent voyages and future polar expeditions.

[0202] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0203] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0204] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A polar icebreaking planning method based on multi-source remote sensing spatio-temporal collaboration, characterized in that, The method includes: Historical remote sensing images were acquired and approximate rank metric features and optical density features were extracted to train a sea ice feasible area discrimination model. The received time-delayed satellite remote sensing images are input into the sea ice feasible area discrimination model to obtain the polar icebreaking feasibility score, construct a polar navigation resistance cost map and generate an initial route. Trigger the shipborne UAV to acquire local images, and extract the optimal icebreaking starting point and optimal cutting direction based on the thickness gradient field; The local image is registered to the time-lapse satellite remote sensing image to form an overlapping area. The features are reconstructed and input into the sea ice feasible area discrimination model to complete the real-time absolute correction of the overlapping area. Extract the reference spatiotemporal drift vector of the overlapping area, calculate the extrapolated drift vector of the non-overlapping area by combining the spatial attenuation weight function and perform reverse mapping, generate a global spatiotemporal corrected navigation resistance cost map, reconstruct the local dynamic route and perform navigation; Acquire historical remote sensing images and extract approximate rank metric features, including: segmenting the historical remote sensing images into segments of size [missing information]. The remote sensing image patch is subjected to affine rotation transformation according to each angle in a preset set of rotation angles. The covariance matrix of the transformed grayscale matrix is ​​calculated, and eigenvalue decomposition is performed to obtain a descending sequence of eigenvalues. subscript Indicates the preset rotation direction; Setting a preset energy threshold , calculating the number of feature values satisfying the cumulative energy proportion requirement , the calculation logic needs to meet the following requirements: ； The number of generated directions The approximate rank measurement feature vector of the remote sensing image block is sequentially arranged and constructed.

2. The method according to claim 1, characterized in that, Constructing a polar navigation resistance cost map includes: using an exponential penalty mapping function to map the grid. Polar icebreaking feasibility score Converted into the cost of polar navigation resistance : ； wherein, is the set ice-free zone base navigation energy consumption constant, is the ice resistance weight coefficient for the current icebreaker power rating, is the resistance non-linear amplification constant; the polar navigation resistance cost of all grids is used A discrete polar navigation resistance cost map is generated.

3. The method according to claim 1, characterized in that, The acquisition of local images by the shipborne UAV is triggered using a dynamic triggering mechanism that incorporates communication time delay. Specifically, this includes calculating the communication time delay of the delayed satellite remote sensing image. Establish and communicate with time delay Positively correlated dynamic safe takeoff distance threshold : ； wherein, is the icebreaker base safety braking distance constant, is the maximum sea ice drift speed scalar of the current sea area historical statistics, is the current real-time speed of the icebreaker, is the mechanical preheating and system initialization time required for the UAV from the system to issue an order to take off; When the calculated distance between the ship's current physical coordinates and the boundary of the sudden ice change ahead is less than or equal to the dynamic safe takeoff distance threshold... At that time, a takeoff trigger command is generated.

4. The method according to claim 1, characterized in that, After extracting the optimal icebreaking starting point based on the thickness gradient field, the optimal cut-in speed is calculated based on energy conservation. The specific calculation formula is as follows: ； in, For the current displacement quality of the icebreaker, This serves as the reference uniaxial compressive strength constant for polar sea ice. The geometric characteristic coefficient represents the current icebreaking efficiency of the bow hull of an icebreaker. This is the estimated absolute sea ice thickness corresponding to the optimal icebreaking starting point obtained through inversion.

5. The method according to claim 1, characterized in that, Registering local images to time-lapse satellite remote sensing images to form overlapping areas includes: extracting the physical pose parameters of the UAV at the exposure time, including the absolute center geographic latitude. Absolute center geographical longitude Absolute flight altitude and the yaw angle of the aircraft For arbitrary pixel coordinates in a local image Based on the physical focal length constant of the camera pod With the Earth's average radius constant The true latitude is calculated using a high-latitude coordinate mapping equation that includes Earth curvature compensation. With actual geographical longitude : ； ； The obtained latitude and longitude coordinates are back-projected to obtain the grid index to which the local image pixels belong, and the set of overlapping grids in the multi-source images to achieve spatial coverage is extracted.

6. The method according to claim 1, characterized in that, Extracting the baseline spatiotemporal drift vector of the overlapping region includes: retrieving the historical polar icebreaking feasibility score matrix. Polar icebreaking feasibility score in overlapping areas generated in real time from local images Feasibility score matrix for breaking ice in historical polar regions Define the matching window in the middle, and define the two-dimensional translation offset variable as follows: Construct a spatiotemporal feature similarity evaluation function based on two-dimensional normalized cross-correlation. Solve for the spatiotemporal feature similarity evaluation function. The optimal pixel offset that yields the global maximum value is then defined as the reference spatiotemporal drift vector for the overlapping region. : 。 7. The method according to claim 6, characterized in that, The extrapolated drift vector of the non-overlapping region is calculated by combining the spatial decay weighting function, including: for any mesh to be corrected in the mesh set of the non-overlapping region. Calculate the distance from the plane Euclidean mesh to the geometric center of the overlapping region. Introducing spatially relevant length constants for polar sea ice deformation. Construct a spatial decay weight function to characterize the uniform decay of ice volleyball movement. : ； The reference spacetime drift vector Multiply by the spatial decay weighting function This yields the extrapolated drift vector specific to the mesh to be corrected. The formula is: 。 8. The method according to claim 7, characterized in that, Perform a reverse mapping to generate a globally spatiotemporally corrected navigation resistance cost map, including: utilizing extrapolated drift vectors. Reverse mapping calculation of historical data source grid floating-point coordinates : ； ； Historical Polar Icebreaking Feasibility Score Matrix The extrapolated and corrected feasibility score is calculated by performing bilinear spatial interpolation. Extrapolate and adjust the feasibility score. Substitute the physical travel cost into the exponential penalty mapping function to reconstruct the physical travel cost, and then stitch the extrapolated update data of the non-overlapping areas with the absolute correction data of the overlapping areas to generate a global spatiotemporal corrected navigation resistance cost map.

9. A polar icebreaking planning system based on multi-source remote sensing spatiotemporal coordination, characterized in that, The system includes: Offline model training module: acquires historical remote sensing images and extracts approximate rank metric features and optical density features to train a sea ice feasible area discrimination model; Initial route generation module: Input the received delayed satellite remote sensing images into the sea ice feasible area discrimination model to obtain the polar icebreaking feasibility score, construct the polar navigation resistance cost map and generate the initial route; Correction module: Triggers shipborne UAV to acquire local images, extracts the optimal icebreaking starting point and optimal cutting direction based on the thickness gradient field; registers the local image to the time-lapse satellite remote sensing image to form an overlapping area, reconstructs features and inputs them into the sea ice feasible area discrimination model to complete the real-time absolute correction of the overlapping area; Reconstruction Module: Extracts the reference spatiotemporal drift vector of the overlapping area, calculates the extrapolated drift vector of the non-overlapping area by combining the spatial decay weight function and performs reverse mapping, generates a global spatiotemporal corrected navigation resistance cost map, reconstructs the local dynamic route and performs navigation; Acquire historical remote sensing images and extract approximate rank metric features, including: segmenting the historical remote sensing images into segments of size [missing information]. The remote sensing image patch is subjected to affine rotation transformation according to each angle in a preset set of rotation angles. The covariance matrix of the transformed grayscale matrix is ​​calculated, and eigenvalue decomposition is performed to obtain a descending sequence of eigenvalues. subscript Indicates the preset rotation direction; Set preset energy threshold Calculate the number of eigenvalues ​​that satisfy the cumulative energy percentage requirement. The computational logic must meet the following requirements: ； Number of generated in each direction Arrange them sequentially to construct an approximate rank metric feature vector for the remote sensing image patch.

Images