A spatiotemporal fusion method for intelligent perception and tracking of oceanic eddies

Abstract

The present application relates to the technical field of ocean eddy intelligent perception and tracking, and particularly relates to a spatio-temporal fusion method for ocean eddy intelligent perception and tracking, comprising the following steps: S1: multi-source data acquisition and preprocessing, collecting ocean environment data including sea surface temperature, sea surface height anomaly and flow velocity field data through satellite remote sensing, ocean buoy and acoustic detection equipment, and performing spatio-temporal alignment, noise filtering and format standardization processing on the data to generate input data in a unified format; the present application introduces fluid dynamics constraints through an improved optical flow method, constructs a physically driven segmentation model based on 3D U-Net, and combines a bidirectional feedback fusion mechanism to effectively solve the tracking distortion and segmentation distortion in the prior art, and in the optical flow tracking stage, the accuracy, real-time performance and generalization capability of eddy analysis under complex flow fields and multi-source data are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing and tracking technology for ocean eddies, specifically a spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies. Background Technology

[0002] The descriptions in this section are provided only as background information relating to this disclosure and do not constitute prior art.

[0003] Currently, the perception and tracking of ocean eddies mainly rely on methods such as satellite remote sensing, buoy observation, and numerical simulation.

[0004] The "Method, Apparatus, Device, and Readable Storage Medium for Predicting Ocean Eddies" disclosed in application number "CN117408147A" is also an increasingly mature technology. It "includes detecting eddy regions and non-eddy regions within a preset time series from ocean current field data; extracting first spatiotemporal features of the current field before the formation of eddy regions and second spatiotemporal features of the current field before the formation of non-eddy regions from the ocean current field data; constructing a dataset using eddy regions and the first spatiotemporal features of the current field as positive samples and non-eddy regions and the second spatiotemporal features of the current field as negative samples; constructing a neural network model and building a model loss function based on the spatiotemporal feature loss of the current field; training and testing the neural network model using the positive and negative samples in the dataset; and obtaining a trained eddy prediction model when the model loss function converges. This method achieves accurate prediction of future eddies by fusing spatiotemporal features, improving the storage efficiency of ocean current field data."

[0005] In terms of dynamic tracking, although traditional optical flow methods can estimate motion trajectories, they lack physical constraints and are easily affected by complex flow fields, leading to trajectory distortion and inaccurate motion estimation. In terms of three-dimensional segmentation, most methods still remain at the two-dimensional level, making it difficult to effectively reconstruct the vertical structure of vortices. Furthermore, deep learning models often ignore physical laws, resulting in segmentation results that do not match the actual fluid behavior.

[0006] Furthermore, existing methods often treat motion tracking and morphology segmentation as independent modules, lacking an effective coordination mechanism. This results in problems such as inconsistencies between trajectories and boundaries, limited overall system accuracy, and poor real-time performance. Existing systems lag significantly in the coordination of motion tracking and morphology segmentation, failing to achieve true closed-loop optimization and limiting the accuracy and real-time performance of vortex trajectory prediction and morphology evolution. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies, solving the problems mentioned in the background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies, comprising the following steps:

[0009] S1: Multi-source data acquisition and preprocessing. Marine environmental data, including sea surface temperature, sea surface height anomaly, and current field data, are collected through satellite remote sensing, ocean buoys, and acoustic detection equipment. The data is then processed for spatiotemporal alignment, noise filtering, and format standardization to generate input data in a unified format.

[0010] S2: Improved optical flow method for vortex motion tracking. Based on the optical flow dynamic tracking module, it uses hydrodynamic constraints to calculate the optical flow field, it estimates the motion trajectory and velocity field of the vortex by iteratively optimizing the optical flow vector field, and introduces a dynamic region focusing mechanism to reduce the segmentation range based on motion prediction and improve tracking efficiency.

[0011] S3: 3D U-Net vortex segmentation driven by deep learning. It uses a 3D convolutional neural network model to segment the boundaries of vortices. The model embeds a physical constraint loss function to ensure that the segmentation results conform to the laws of fluid dynamics. It outputs the 3D morphological parameters of the vortex, including the vortex boundary, center and intensity.

[0012] S4: Bidirectional feedback combined with computation, through the bidirectional iterative fusion optimization module, the optical flow tracking results and image segmentation results are mutually corrected and iteratively optimized until the tracking trajectory and segmentation boundary converge to a stable state, realizing the synchronous driving of motion tracking and boundary segmentation;

[0013] S5: Vortex life history synthesis and output, generating standardized data products, including a life history report integrating vortex trajectory, morphological evolution, and physical parameters, and outputting it to users through a visual interface.

[0014] Preferably, steps two and S3 achieve dynamic accuracy improvement through bidirectional iterative optimization. The perception of ocean eddies relies on multiple observation methods. The two types of data are aligned and complemented under a unified spatiotemporal coordinate system. Remote sensing data is used to outline the macroscopic two-dimensional distribution and movement trajectory of the eddies. At the same time, sparse field data is used to correct the deviation of the remote sensing inversion results and to provide numerical values ​​for the vertical structure of the eddies, thereby constructing a more three-dimensional eddy information from the sea surface to underwater.

[0015] Preferably, the multi-source data acquisition and preprocessing steps specifically include: integrating remote sensing data from multiple satellites, including sensor data and acoustic current profiler data; unifying data from different sources into the same spatiotemporal grid; using interpolation methods to process missing values ​​and ensuring consistent temporal resolution; applying wavelet transform and Kalman filtering to remove high-frequency noise and outliers; retaining vortex-related signals for easy retrieval by subsequent modules; and evaluating the quality of the preprocessed data by calculating the signal-to-noise ratio and consistency index to ensure the quality of the input data.

[0016] Preferably, the improved optical flow method for vortex motion tracking specifically includes: calculating the initial optical flow field based on the optical flow algorithm, optimizing the optical flow vector by minimizing the energy function, identifying high-probability vortex regions based on the optical flow prediction results, narrowing the segmentation range, reducing the computational load, using a multi-scale pyramid structure for optical flow calculation, gradually refining the motion field from coarse to fine, and solving the optical flow equation using the Gauss-Newton method to generate a vortex motion trajectory diagram and a time-series velocity field. The optical flow method also incorporates particle image velocimetry technology to improve tracking accuracy in complex flow fields.

[0017] Preferably, the vortex 3D segmentation step specifically adopts a 3D U-Net model with an encoder-decoder architecture. The encoder part extracts features through 3D convolution and pooling layers, and the decoder part restores spatial resolution through deconvolution and skip connections. A hydrodynamic constraint term is added to the standard cross-entropy loss function to ensure that the segmentation result is physically realistic. The model is trained using historical vortex data and synthetic data, and the generalization ability is improved through data augmentation. The model outputs a 3D binary mask of the vortex, including boundary, center, and intensity parameters. Morphological operations are applied to remove noise, and the segmentation results on the time series are connected to generate a continuous vortex shape.

[0018] Preferably, the bidirectional feedback operation step specifically includes: obtaining the initial motion trajectory and boundary segmentation results from the optical flow tracing module and the 3D U-Net segmentation module respectively; projecting the motion vector of the optical flow tracing onto the segmentation boundary; adjusting the optical flow field or segmentation mask; iterating and updating the optical flow tracing and segmentation results alternately until the difference between the trajectory and the boundary is less than a preset threshold; using the mean square error and intersection-union ratio as convergence indicators to generate the optimized vortex motion trajectory and three-dimensional shape; the collaborative fusion mechanism also introduces adaptive weight adjustment to dynamically adjust the contribution of optical flow and segmentation according to data quality.

[0019] Preferably, the optimized motion trajectory, three-dimensional morphology, and physical parameters (such as vorticity and energy) are integrated into a unified database to automatically generate a standardized vortex life history report. The morphological evolution animation and physical parameter time series application display the results, support user interactive query, and support export in NetCDF, JSON, and CSV formats for easy subsequent analysis. The life history report also includes uncertainty assessment, and confidence intervals are calculated based on error propagation during the fusion process.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] This invention provides a spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies:

[0022] 1. This invention introduces fluid dynamic constraints through an improved optical flow method and constructs a physics-driven segmentation model based on 3D U-Net. Combined with bidirectional feedback fusion, it solves tracking and segmentation distortion. In the optical flow tracking and 3D segmentation stages, by adding a vorticity term to the loss function, it ensures the physical consistency of the segmentation results, thereby accurately restoring the vortex's 3D structure. Through a bidirectional iterative fusion mechanism, it achieves dynamic correction and collaborative optimization of the motion trajectory and segmentation boundary, effectively eliminating inconsistencies between modules and significantly improving the overall accuracy of the system. Through spatiotemporal alignment and noise filtering of multi-source data, it provides high-quality input for subsequent processing, reducing error accumulation from the source. It exhibits stronger generalization and real-time processing capabilities in complex marine environments, providing reliable technical support for the full life cycle monitoring of marine vortices.

[0023] 2. This invention fundamentally improves the three-dimensionality and accuracy of eddy sensing by constructing a unified multi-source data preprocessing platform and a two-way feedback fusion mechanism. Through spatiotemporal grid alignment, wavelet denoising, and Kalman filtering, it effectively integrates satellite remote sensing and field observation data to form high-quality, multi-dimensional input, providing a reliable foundation for subsequent analysis. At the algorithm level, through bidirectional iterative optimization of optical flow tracing and 3D U-Net segmentation, it achieves dynamic complementarity and correction of motion and morphological information, significantly improving the consistency of eddy three-dimensional boundaries and motion trajectories. This ensures high fusion accuracy and stability under different data quality environments, overcoming key technical bottlenecks in multi-source data fusion and providing a complete solution for the three-dimensional and real-time monitoring of ocean eddies. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0027] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0028] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "configuration" should be interpreted broadly. For example, they can refer to a fixed connection or configuration, a detachable connection or configuration, or an integral connection or configuration. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0029] like Figure 1 As shown, the present invention proposes a spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies, comprising the following steps:

[0030] S1: Multi-source data acquisition and preprocessing. Marine environmental data, including sea surface temperature, sea surface height anomaly, and current field data, are collected through satellite remote sensing, ocean buoys, and acoustic detection equipment. The data is then processed for spatiotemporal alignment, noise filtering, and format standardization to generate input data in a unified format.

[0031] S2: Improved optical flow method for vortex motion tracking. Based on the optical flow dynamic tracking module, it uses hydrodynamic constraints to calculate the optical flow field, it estimates the motion trajectory and velocity field of the vortex by iteratively optimizing the optical flow vector field, and introduces a dynamic region focusing mechanism to reduce the segmentation range based on motion prediction and improve tracking efficiency.

[0032] S3: 3D U-Net vortex segmentation driven by deep learning. It uses a 3D convolutional neural network model to segment the boundaries of vortices. The model embeds a physical constraint loss function to ensure that the segmentation results conform to the laws of fluid dynamics. It outputs the 3D morphological parameters of the vortex, including the vortex boundary, center and intensity.

[0033] S4: Bidirectional feedback combined with computation, through the bidirectional iterative fusion optimization module, the optical flow tracking results and image segmentation results are mutually corrected and iteratively optimized until the tracking trajectory and segmentation boundary converge to a stable state, realizing the synchronous driving of motion tracking and boundary segmentation;

[0034] S5: Vortex life history synthesis and output, generating standardized data products, including a life history report integrating vortex trajectory, morphological evolution, and physical parameters, and outputting it to users through a visual interface.

[0035] Among them, S2 and S3 achieve dynamic accuracy improvement through bidirectional iterative optimization. The perception of ocean eddies relies on multiple observation methods. The two types of data are aligned and complemented under a unified spatiotemporal coordinate system. Remote sensing data is used to outline the macroscopic two-dimensional distribution and movement trajectory of the eddies. At the same time, sparse field data is used to correct the deviation of the remote sensing inversion results and provide numerical values ​​for the vertical structure of the eddies, thereby constructing a more three-dimensional eddy information from the sea surface to underwater.

[0036] In practical applications, a bidirectional iterative fusion of optical flow tracing and image segmentation is used to improve dynamic tracking accuracy and boundary segmentation real-time performance. In the multi-source data acquisition and preprocessing stage, satellite remote sensing and field observation data are integrated to ensure the diversity and reliability of input data. The preprocessing steps eliminate noise and spatiotemporal inconsistencies, providing high-quality input for subsequent analysis. In the optical flow tracing stage, the improved optical flow method introduces hydrodynamic constraints, and the vortex motion trajectory is estimated by optimizing the optical flow field. The dynamic region focusing mechanism reduces the computational range. In the 3D U-Net segmentation stage, the segmentation results are ensured to conform to real physical laws, and the accurate vortex 3D morphology is output. The bidirectional feedback mechanism iteratively corrects the optical flow tracing and segmentation results, ultimately avoiding tracking distortion and segmentation lag, and improving generalization ability and real-time performance.

[0037] The multi-source data acquisition and preprocessing steps specifically include: integrating remote sensing data from multiple satellites, including sensor data and acoustic current profiler data; unifying data from different sources into the same spatiotemporal grid; using interpolation methods to handle missing values ​​and ensuring consistent temporal resolution; applying wavelet transform and Kalman filtering to remove high-frequency noise and outliers; and retaining vortex-related signals for easy use by subsequent modules. The preprocessing steps also include data quality assessment, which ensures the reliability of the input data by calculating the signal-to-noise ratio and consistency index.

[0038] In practical use, by integrating heterogeneous data sources and standardizing their processing, the reliability of subsequent modules is ensured. During the data acquisition phase, multiple satellite and field device data are selected, covering different spatiotemporal scales and physical parameters. Spatiotemporal alignment is achieved by interpolating data to unify them to the same grid, avoiding errors caused by resolution differences. High-frequency noise and outliers are removed, while vortex-related signals are preserved. Format standardization converts data to NetCDF or HDF5 format for easy storage and retrieval. Data quality assessment is performed by calculating the signal-to-noise ratio and consistency indicators. Through multi-source fusion and preprocessing, a high-quality, consistent data foundation is provided for optical flow tracing and segmentation, reducing error accumulation in subsequent modules.

[0039] The improved optical flow method for vortex motion tracking specifically includes: calculating the initial optical flow field based on the optical flow algorithm; optimizing the optical flow vector by minimizing the energy function; identifying high-probability vortex regions based on the optical flow prediction results; narrowing the segmentation range to reduce computational load; using a multi-scale pyramid structure for optical flow calculation; gradually refining the motion field from coarse to fine; and solving the optical flow equation using the Gauss-Newton method to generate vortex motion trajectory diagrams and time-series velocity fields. The optical flow method also incorporates particle image velocimetry technology to improve tracking accuracy in complex flow fields.

[0040] In practical use, dynamic region focusing identifies high-probability vortex regions based on optical flow prediction results, narrowing the segmentation range, reducing unnecessary calculations, and improving efficiency. Iterative optimization employs a multi-scale pyramid structure, progressively refining the motion field from coarse to fine. The output includes motion trajectory and velocity field. Through physical constraints and dynamic focusing, the accuracy and efficiency of optical flow tracing in complex flow fields are improved, providing reliable motion information for bidirectional fusion.

[0041] The 3D U-Net vortex segmentation steps driven by deep learning specifically include: using a 3D U-Net model with an encoder-decoder architecture, the encoder extracts features through 3D convolution and pooling layers, the decoder restores spatial resolution through deconvolution and skip connections, a hydrodynamic constraint term is added to the standard cross-entropy loss function to ensure the physical realism of the segmentation results, the model is trained using historical vortex data and synthetic data, and the generalization ability is improved through data augmentation, the model outputs a 3D binary mask of the vortex, including boundary, center and intensity parameters, morphological operations are applied to remove noise, and the segmentation results on the time series are connected to generate continuous vortex shapes.

[0042] In practical use, the 3D U-Net vortex 3D segmentation steps driven by deep learning are limited. Its working principle is based on a 3D convolutional neural network and a physical constraint loss function. The network structure adopts an encoder-decoder architecture. The encoder extracts features through 3D convolution and pooling layers, and the decoder restores spatial resolution through deconvolution and skip connections to ensure that segmentation details are preserved and that the model output conforms to the laws of fluid dynamics. The training data is used to improve the model's generalization ability through data augmentation. The segmentation output generates a 3D binary mask to identify the vortex boundary and center. Morphological operations are applied to remove noise and connect the time series results to achieve high-precision, physically realistic vortex 3D segmentation, providing accurate morphological information for bidirectional fusion.

[0043] The bidirectional feedback-assisted computation steps specifically include: obtaining initial motion trajectories and boundary segmentation results from the optical flow tracing module and the 3D U-Net segmentation module respectively; projecting the motion vectors of the optical flow tracing onto the segmentation boundary; adjusting the optical flow field or segmentation mask; and iteratively updating the optical flow tracing and segmentation results alternately until the difference between the trajectory and the boundary is less than a preset threshold. The mean square error and intersection-union ratio are used as convergence indicators to generate optimized vortex motion trajectories and three-dimensional shapes. The collaborative fusion mechanism also introduces adaptive weight adjustment, dynamically adjusting the contribution of optical flow and segmentation according to data quality.

[0044] In practical use, bidirectional feedback is used in conjunction with the calculation steps. In the initialization phase, initial results are obtained from two modules. In the mutual correction phase, the optical flow motion vector is projected onto the segmentation boundary, and the optical flow field or segmentation mask is adjusted. In the iterative optimization phase, the tracking and segmentation results are updated alternately until the difference is less than the threshold. The convergence judgment uses the mean square error and intersection-union ratio to ensure the stability of the results. The optimized trajectory and shape are output. Through bidirectional feedback and iterative optimization, the inconsistency between tracking and segmentation is eliminated, and dynamic accuracy improvement and real-time boundary segmentation are achieved.

[0045] The optimized motion trajectory, 3D morphology, and physical parameters (such as vorticity and energy) are integrated into a unified database to automatically generate a standardized vortex life history report. The morphological evolution animation and physical parameter time series application display the results, support user interactive query, and support export in NetCDF, JSON, and CSV formats for easy subsequent analysis. The life history report also includes uncertainty assessment, and confidence intervals are calculated based on error propagation during the fusion process.

[0046] In practical use, the eddy life history synthesis and output process works by integrating all optimized results and generating a standardized report. During the data integration phase, motion trajectories, three-dimensional morphology, and physical parameters are integrated into a unified database. During the report generation phase, a life history report is automatically created, including trajectory diagrams, morphological evolution animations, and parameter time series. The visualization output is displayed through a web or desktop interface, supporting user interactive queries. Uncertainty assessment is based on the calculation of confidence intervals during error propagation in the fusion process, improving the credibility of the results. Through standardized output and visualization, it provides users with comprehensive and intuitive eddy life history information, supporting marine research and applications.

[0047] The system employs a modular architecture to implement the workflow. A multi-source data acquisition and preprocessing module processes the input data. A physically guided optical flow dynamic tracking module performs motion estimation. A 3D U-Net vortex segmentation module completes boundary segmentation. A bidirectional iterative fusion optimization module performs correction and optimization. A vortex lifecycle synthesis and output module generates a report. The system coordinates its work through inter-module data flow; for example, the preprocessing module outputs data to the tracking and segmentation modules, and the fusion module receives the outputs from both for optimization. A user interface module allows for parameter input and result viewing. The system's working principle is to achieve efficient and automated vortex sensing and tracking through integrated modular design, improving overall performance and ease of use.

[0048] The modular design enhances tracking accuracy and efficiency. The optical flow computation unit executes an improved optical flow algorithm. The physical constraint unit embeds fluid dynamics equations to ensure the physical realism of the optical flow field. The dynamic focusing unit reduces the processing area based on motion prediction, reducing computational load. The iterative optimization unit optimizes the optical flow field through a multi-scale method. The GPU acceleration unit improves computational speed, achieving high-precision motion tracking and providing reliable input for other modules in the system.

[0049] Based on deep learning models and distributed computing, the module employs a network training unit to optimize model parameters, a physical loss unit to calculate constraint loss, a segmentation execution unit to perform 3D segmentation on the input data, a post-processing unit to smooth and connect the results, and a distributed computing unit to process large-scale data. This module works by leveraging efficient network structures and computing resources to achieve fast and accurate vortex segmentation, supporting real-time processing requirements.

[0050] When the processor runs the stored program, it sequentially performs data acquisition, optical flow tracing, 3D U-Net segmentation, bidirectional fusion, and life history output steps. Storage media such as SSDs or cloud platforms ensure reliable storage and fast access to the program, enabling the technical solution to be deployed in a variety of computing environments, improving its applicability and scalability.

[0051] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0052] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies, characterized in that, Includes the following steps: S1: Multi-source data acquisition and preprocessing. Marine environmental data, including sea surface temperature, sea surface height anomaly, and current field data, are collected through satellite remote sensing, ocean buoys, and acoustic detection equipment. The data is then processed for spatiotemporal alignment, noise filtering, and format standardization to generate input data in a unified format. S2: Improved optical flow algorithm for vortex motion tracking. Based on the optical flow dynamic tracking module, it uses hydrodynamic constraints to calculate the optical flow field, it iteratively optimizes the optical flow vector field, estimates the motion trajectory and velocity field of the vortex, and introduces a dynamic region focusing mechanism to narrow the segmentation range based on motion prediction, thereby improving tracking efficiency. S3: 3D U-Net vortex segmentation driven by deep learning. It uses a 3D convolutional neural network model to segment the boundaries of vortices. The model embeds a physical constraint loss function to ensure that the segmentation results conform to the laws of fluid dynamics. It outputs the 3D morphological parameters of the vortex, including the vortex boundary, center and intensity. S4: Bidirectional feedback combined with computation, through the bidirectional iterative fusion optimization module, the optical flow tracking results and image segmentation results are mutually corrected and iteratively optimized until the tracking trajectory and segmentation boundary converge to a stable state, realizing the synchronous driving of motion tracking and boundary segmentation; S5: Vortex life history synthesis and output, generating standardized data products, including a life history report integrating vortex trajectory, morphological evolution, and physical parameters, and outputting it to users through a visual interface.

2. The spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies according to claim 1, characterized in that: The S2 and S3 achieve dynamic accuracy improvement through bidirectional iterative optimization. The perception of ocean eddies relies on multiple observation methods. The two types of data are aligned and complemented under a unified spatiotemporal coordinate system. Remote sensing data is used to outline the macroscopic two-dimensional distribution and movement trajectory of the eddies. At the same time, sparse field data is used to correct the deviation of the remote sensing inversion results and provide numerical values ​​for the vertical structure of the eddies, thereby constructing a more three-dimensional eddy information from the sea surface to underwater.

3. The spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies according to claim 1, characterized in that: The multi-source data acquisition and preprocessing steps specifically include: integrating remote sensing data from multiple satellites; preprocessing data quality assessment, including sensor data and acoustic current profiler data; unifying data from different sources onto the same spatiotemporal grid; using interpolation methods to handle missing values ​​and ensuring consistent temporal resolution; applying wavelet transform and Kalman filtering to remove high-frequency noise and outliers; and retaining vortex-related signals for easy retrieval by subsequent modules. The preprocessing data quality assessment ensures the quality of input data by calculating the signal-to-noise ratio and consistency index.

4. The spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies according to claim 1, characterized in that: The improved optical flow algorithm for vortex motion tracking specifically includes: calculating the initial optical flow field based on the optical flow algorithm; optimizing the optical flow vector by minimizing the energy function; identifying high-probability vortex regions based on the optical flow prediction results; narrowing the segmentation range to reduce computational load; using a multi-scale pyramid structure for optical flow calculation; gradually refining the motion field from coarse to fine; and solving the optical flow equation using the Gauss-Newton method to generate vortex motion trajectory diagrams and time-series velocity fields. The optical flow algorithm also incorporates particle image velocimetry technology to improve tracking accuracy in complex flow fields.

5. A spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies according to claim 1, characterized in that: The vortex 3D segmentation step specifically adopts a 3D U-Net model with an encoder-decoder architecture. The encoder extracts features through 3D convolution and pooling layers, while the decoder restores spatial resolution through deconvolution and skip connections. A hydrodynamic constraint term is added to the standard cross-entropy loss function to ensure the physical realism of the segmentation results. The model is trained using historical vortex data and synthetic data, and its generalization ability is improved through data augmentation. The model outputs a 3D binary mask of the vortex, including boundary, center, and intensity parameters. Morphological operations are applied to remove noise, and the segmentation results from the time series are connected to generate a continuous vortex shape.

6. A spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies according to claim 1, characterized in that: The bidirectional feedback operation steps specifically include: obtaining initial motion trajectories and boundary segmentation results from the optical flow tracing module and the 3D U-Net segmentation module respectively; projecting the motion vectors of the optical flow tracing onto the segmentation boundary; adjusting the optical flow field or segmentation mask; iterating and updating the optical flow tracing and segmentation results alternately until the difference between the trajectory and the boundary is less than a preset threshold; using mean square error and intersection-union ratio as convergence indicators to generate optimized vortex motion trajectories and three-dimensional shapes; and the collaborative fusion mechanism also introduces adaptive weight adjustment to dynamically adjust the contribution of optical flow and segmentation according to data quality.

7. The spatiotemporal fusion method for intelligent sensing and tracking of ocean eddies according to claim 1, characterized in that: The S3-optimized motion trajectory, 3D morphology, and physical parameters are integrated into a unified database to automatically generate a standardized vortex life history report. The morphological evolution animation and physical parameter time series application display the results, support user interactive query, and support export in NetCDF, JSON, and CSV formats for easy subsequent analysis. The life history report also includes uncertainty assessment, which calculates confidence intervals based on error propagation during the fusion process.

Images