A method for automatically tracking ocean temperature front and extracting characteristic parameter information

Abstract

This invention relates to the field of marine information and pattern recognition technology, and discloses a method for automatic tracking and feature parameter extraction of ocean temperature fronts. The method includes: acquiring and preprocessing continuous sea surface temperature remote sensing data; constructing an adaptive multi-scale temperature gradient tensor field; automatically identifying initial seed points using a joint criterion of local extremum response and global structural saliency; generating a preliminary front through bidirectional chain growth along the main gradient direction; optimizing the front trajectory by fusing ocean dynamics priors and spatiotemporal continuity constraints; and finally extracting feature parameters such as position, intensity, direction, curvature, and lifespan. Through the above technical solution, this invention achieves fully automatic and highly robust temperature front tracking and accurate parameter extraction, significantly improving cross-scenario generalization capabilities and operational application value.

Description

Technical Field

[0001] This invention belongs to the field of marine information and pattern recognition technology, specifically relating to a method for automatic tracking of ocean temperature fronts and extraction of feature parameter information. Background Technology

[0002] Oceanic temperature fronts, as narrow transitional zones with significant temperature gradients in the ocean, are crucial interfaces for the interaction of ocean dynamic processes, ecological distribution, and the climate system, and have important applications in marine environmental monitoring, fisheries resource management, and climate prediction. With the development of satellite remote sensing and field observation technologies, high spatiotemporal resolution sea surface temperature data are becoming increasingly abundant, providing a data foundation for the refined identification and dynamic tracking of temperature fronts. However, temperature fronts exhibit highly nonlinear spatial morphology, strong time-varying characteristics, and are influenced by multiple factors such as eddies, flow fields, and seasonal variations, resulting in complex and variable structures that pose significant challenges to automated tracking.

[0003] Automatic tracking methods for ocean temperature fronts typically rely on image processing techniques such as edge detection, gradient thresholding, or curve fitting, combined with temporal correlation based on continuity constraints. While these methods can extract front locations under specific conditions, their core parameters (such as gradient threshold, curvature tolerance, and temporal correlation window) often require manual setting based on experience, lacking adaptability to the physical characteristics of different sea areas and seasonal evolution patterns. When applied to cross-regional or multi-seasonal scenarios, fixed parameters can easily lead to problems such as frontal breakage, false connections, or trajectory drift, severely affecting the continuity and physical plausibility of the tracking results.

[0004] In the process of dynamically tracking temperature fronts, there are common drawbacks such as rigid parameters, lack of prior constraints on ocean dynamics, and difficulty in balancing frontal geometry and physical consistency. Especially under complex flow field disturbances or weak frontal conditions, traditional methods cannot dynamically adjust the identification and tracking strategy, leading to distortion in the extraction of characteristic parameters (such as front strength, direction, and curvature), making it difficult to support high-precision ocean process analysis and operational applications. Summary of the Invention

[0005] This invention provides a method for automatic tracking and feature parameter extraction of ocean temperature fronts, aiming to solve the technical problems of existing tracking algorithms that rely on human experience and are difficult to adapt to dynamic changes in different sea areas and seasons. The method constructs an adaptive multi-scale temperature gradient field model, combines a spatiotemporally context-aware intelligent identification mechanism for front initial seed points, and employs a dynamic optimization strategy for front evolution paths based on a fusion of physical constraints and data-driven approaches. This enables fully automatic and robust tracking of ocean temperature fronts, and simultaneously extracts key feature parameters such as their location, intensity, direction, curvature, and lifespan.

[0006] As one embodiment of the present invention, the method for automatic tracking and feature parameter extraction of ocean temperature fronts includes the following steps: acquiring continuous time-series sea surface temperature remote sensing data covering the target sea area; preprocessing the sea surface temperature remote sensing data to generate a standardized spatiotemporally consistent sea surface temperature field; constructing a multi-scale temperature gradient tensor field based on the standardized sea surface temperature field; automatically identifying the initial seed point set of the temperature front using the multi-scale temperature gradient tensor field through a joint criterion of local extreme response and global structural saliency; generating preliminary frontal segments by bidirectional chain growth along the main direction of the temperature gradient, starting from the initial seed point set; applying ocean dynamics prior constraints and spatiotemporal continuity regularization terms to the preliminary frontal segments, and solving for the optimal frontal trajectory through an iterative optimization algorithm; and calculating and outputting the geometric and physical feature parameters of the temperature front based on the optimal frontal trajectory.

[0007] Furthermore, the acquisition of continuous time-series sea surface temperature remote sensing data covering the target sea area specifically includes: receiving raw brightness temperature data collected by an infrared radiometer carried by a polar-orbiting or geostationary ocean satellite; performing atmospheric correction, cloud pollution removal, and orbital stitching processing on the raw brightness temperature data; and resampling the processed data to a unified latitude and longitude grid with a time resolution of not less than once per day and a spatial resolution of not less than 1 / 4 degree.

[0008] Furthermore, the preprocessing of sea surface temperature remote sensing data to generate a standardized spatiotemporally consistent sea surface temperature field specifically includes: performing outlier detection and repair on the resampled sea surface temperature data, using a sliding window mean filter to identify and replace data points that deviate from more than 3 times the interquartile range; performing spatiotemporal interpolation on missing regions, using an anisotropic diffusion equation for smooth filling, with the diffusion coefficient adaptively adjusted according to the direction of the local temperature gradient; and performing mean-variance standardization on the entire time series, so that the time series mean of each grid point is 0 and the variance is 1.

[0009] Furthermore, the construction of the multi-scale temperature gradient tensor field specifically includes: on the standardized sea surface temperature field, performing convolution smoothing using Gaussian kernel functions with scale parameters of 0.5 degrees, 1 degree, and 1.5 degrees respectively to generate three smoothed temperature fields at different scales; for each smoothed temperature field, calculating its first-order partial derivative to obtain the temperature gradient components in the east-west and north-south directions; combining the two gradient components at the same scale into a two-dimensional gradient vector and calculating their outer product to form the second-order temperature gradient tensor at that scale; and fusing the gradient tensors of the three scales by scale weight, with the weights determined by the signal-to-noise ratio of the local structure at each scale, and assigning higher weights to scales with higher signal-to-noise ratios.

[0010] Furthermore, the automatic identification of the initial seed point set for temperature fronts using a joint criterion of local extreme response and global structural saliency specifically includes: calculating the gradient magnitude and gradient direction consistency index for each grid point in the fused multi-scale gradient tensor field; the gradient magnitude is defined as the Euclidean norm of the gradient vector, and the gradient direction consistency index is defined as the reciprocal of the standard deviation of the gradient direction in the neighborhood; setting gradient magnitude thresholds and direction consistency thresholds, both of which are automatically determined from the global distribution using the non-parametric Otsu method; grid points that simultaneously satisfy both threshold conditions are marked as candidate seed points; non-maximum suppression is applied to the candidate seed points, retaining the points with the largest local gradient magnitudes to form the final initial seed point set.

[0011] Furthermore, the process of generating preliminary frontal segments by bidirectional chain growth along the main direction of the temperature gradient, starting from the initial set of seed points, specifically includes: for each initial seed point, the gradient direction at its location is used as the initial growth direction; during the growth process, at each step, the grid point with the largest gradient magnitude and whose direction makes an angle of less than 45 degrees with the current growth direction is searched forward within the neighborhood as the next node; the growth step size is fixed at one grid cell; when no next node that meets the conditions can be found in three consecutive steps, or when an existing frontal segment is encountered, the growth in that direction is stopped; growth in both positive and negative directions is performed on each seed point, and all nodes are connected to form closed or open preliminary frontal segments.

[0012] Furthermore, the application of ocean dynamics prior constraints and spatiotemporal continuity regularization terms to the initial frontal line segment, and the solution of the optimal frontal trajectory through an iterative optimization algorithm, specifically includes: constructing an objective function, which consists of three terms: the first term is a data fidelity term, which measures the sum of the temperature gradient amplitudes at each point on the optimized frontal trajectory and requires maximization; the second term is a dynamic smoothing term, which introduces geostrophic constraints, requiring that the deviation between the frontal direction and the background geostrophic direction does not exceed 30 degrees, with a heavier penalty for larger deviations; the third term is a spatiotemporal continuity term, requiring that the rate of change of the frontal position at adjacent times does not exceed the typical propagation speed of a mesoscale eddy in the ocean, i.e., not exceeding 50 kilometers per day; minimizing the objective function using a sequential quadratic programming algorithm, with the latitude and longitude coordinates of all nodes on the frontal trajectory as the variables; iterating until the change in the objective function value is less than a preset convergence threshold, or reaching the maximum number of iterations.

[0013] Furthermore, the calculation and output of the geometric and physical characteristic parameters of the temperature front based on the optimal frontal trajectory specifically includes: the position parameter being the latitude and longitude coordinate sequence of all nodes on the frontal trajectory; the intensity parameter being the average value of the temperature gradient amplitude at each node on the trajectory; the orientation parameter being the principal axis direction of the entire trajectory, calculated using principal component analysis; the curvature parameter being the reciprocal of the local radius of curvature at each node on the trajectory, calculated using a 3-point circle fitting method; and the lifespan parameter being the time span from the first detection of the front to its last tracking, in days.

[0014] In a preferred embodiment of the present invention, when constructing a multi-scale temperature gradient tensor field, the scale parameter of the Gaussian kernel function is dynamically adjusted according to the typical mesoscale eddy scale of the target sea area. Specifically, by analyzing the power spectral density of historical sea surface temperature data, the wavelength corresponding to the peak of the energy spectrum is selected as the reference scale, and the other scales are set at 50% and 150% of the reference scale.

[0015] In a preferred embodiment of the present invention, when applying prior constraints on ocean dynamics, the background geostrophic data is derived from remote sensing products of sea surface height anomalies from the same period. It is obtained by calculating the gradient of sea surface height anomalies and applying the geostrophic balance equation for inversion, and the spatial resolution is consistent with the sea surface temperature field.

[0016] In a preferred embodiment of the present invention, the typical propagation speed of the ocean mesoscale eddy in the spatiotemporal continuity regularization term is corrected according to latitude, taking 80 km / day in the equatorial region, 50 km / day in the mid-latitude region, and 30 km / day in the high-latitude region.

[0017] In a preferred embodiment of the present invention, when calculating the temperature front intensity parameter, the standard deviation of the intensity is output simultaneously to characterize the non-uniformity of the front intensity along the front; when calculating the curvature parameter, the number of curvature sign change points is calculated simultaneously to identify the inflection point or vortex embedding structure of the front.

[0018] In a preferred embodiment of the present invention, the method further includes a front type identification step: based on the temperature difference sign on both sides of the optimal front trajectory, if the warm water side is located on the right side of the trajectory's forward direction, it is determined to be a cold front; if the warm water side is located on the left side, it is determined to be a warm front; if the absolute value of the temperature difference on both sides is less than 0.5 degrees Celsius, it is determined to be a weak front or a mixed front.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention abandons the reliance on fixed thresholds, fixed scales or fixed growth rules in traditional methods, and achieves a unified characterization of temperature fronts of different intensities and widths through adaptive fusion of multi-scale gradient tensor fields. By using the joint criterion of local extrema and global significance, the starting position of the front can be accurately captured in complex background noise without the need to manually set the seed point identification threshold. By introducing prior knowledge of ocean dynamics and spatiotemporal continuity constraints, the front tracking results not only conform to the local characteristics of the observation data, but also follow the laws of ocean physics, significantly improving the generalization ability and tracking stability in cross-sea and cross-seasonal scenarios; The output feature parameter system is complete, well-defined, and has clear physical meaning. It can directly serve operational application scenarios such as marine environmental monitoring, fishery resource assessment, and numerical model assimilation, and solves the problems of missed detection, false detection, and feature distortion caused by parameter rigidity in existing technologies. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall technical solution architecture of the method for automatic tracking and feature parameter information extraction of ocean temperature fronts proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the adaptive multi-scale temperature gradient field model and the intelligent identification mechanism of frontal seed points in this invention. Figure 3 This is a logical flowchart of the sea surface temperature data preprocessing and standardized spatiotemporal consistent temperature field construction in this invention; Figure 4 This is a logical flowchart of the bidirectional chain-like frontal growth and preliminary line segment generation based on the gradient principal direction in this invention. Figure 5 This is a schematic diagram of the frontal trajectory optimization framework that integrates ocean dynamics priors and spatiotemporal continuity constraints in this invention; Figure 6 This is a flowchart illustrating the logical process of extracting and classifying the geometric and physical characteristic parameters of the temperature front in this invention. Detailed Implementation

[0021] Please refer to Figures 1 to 6 This invention provides a method for automatic tracking and feature parameter extraction of ocean temperature fronts. Its core lies in constructing an adaptive multi-scale temperature gradient field model, combining a spatiotemporally context-aware intelligent identification mechanism for front initial seed points, and a dynamic optimization strategy for front evolution paths based on a fusion of physical constraints and data-driven approaches. This enables fully automatic and robust tracking of ocean temperature fronts, simultaneously extracting key feature parameters such as their location, intensity, direction, curvature, and lifespan. The specific implementation of this method will be described in detail below, following the S-step sequence explicitly specified in the invention description.

[0022] The method first performs step S1: acquiring continuous time-series sea surface temperature remote sensing data covering the target sea area. This step specifically includes receiving raw brightness temperature data collected by infrared radiometers carried by polar-orbiting or geostationary oceanographic satellites. The raw brightness temperature data needs to undergo atmospheric correction processing to eliminate interference from atmospheric components such as water vapor, aerosols, and carbon dioxide on infrared radiation transmission; subsequently, cloud pollution removal is performed, using a multi-channel thresholding method combined with spatial texture analysis to identify and mask pixels obscured by clouds; finally, orbital stitching processing is performed to fuse transit data from different times on the same day into a single-day seamless coverage map. The processed data is resampled to a unified latitude and longitude grid system with a temporal resolution of no less than once per day and a spatial resolution of no less than 1 / 4 degree, ensuring sufficient spatiotemporal density and consistency for subsequent analysis.

[0023] After completing step S1, step S2 is executed: preprocessing the sea surface temperature remote sensing data to generate a standardized spatiotemporally consistent sea surface temperature field. This preprocessing process includes three sub-operations. The first sub-operation is outlier detection and repair: a sliding window median filter is applied to the time series of each grid point, with a window length of 7 days, and the absolute deviation between the current point and the median of its neighborhood is calculated; if the deviation exceeds 3 times the interquartile range, it is determined to be an outlier and replaced by the neighborhood median. The second sub-operation is spatiotemporal interpolation of missing regions: for data gaps caused by residual cloud pollution or sensor malfunction, anisotropic diffusion equations are used for smoothing and filling. The diffusion coefficient tensor is determined by the direction of the local temperature gradient, with a high diffusivity set along the isotherm direction and a low diffusivity set in the vertical direction, thereby achieving internal smoothing while maintaining the sharpness of the frontal structure. The third sub-operation is mean-variance standardization: For each grid point in the entire time series, its historical mean and standard deviation are calculated, and the original temperature values ​​are converted into standardized variables with zero mean and unit variance. This eliminates the interference of background temperature differences in different sea areas on gradient calculation, making subsequent multi-scale analysis comparable.

[0024] Next, step S3 is performed: based on the standardized sea surface temperature field, a multi-scale temperature gradient tensor field is constructed. This step first uses Gaussian kernel functions with scale parameters of 0.5 degrees, 1 degree, and 1.5 degrees to perform convolution smoothing in the standardized field, generating three smoothed temperature fields at different scales. The standard deviation of the Gaussian kernel corresponds to the aforementioned scale parameters, used to simulate the response of ocean fronts of different widths. Then, the first-order partial derivatives in the east-west (meridian) and north-south (zonal) directions are calculated for each smoothed temperature field, and numerical differentiation is achieved using a central difference scheme to obtain two orthogonal temperature gradient components. The two gradient components at the same scale are combined into a two-dimensional gradient vector. And calculate its outer product. This forms a second-order temperature gradient tensor at that scale. Finally, the gradient tensors from the three scales are weighted and fused according to scale, with the weights... The signal-to-noise ratio (SNR) is determined by the local structure at each scale. The SNR is defined as the ratio of the mean to the standard deviation of the local gradient magnitudes. A higher SNR indicates a stronger representation of the true frontal structure at that scale, and is therefore assigned a higher weight. The fused multi-scale gradient tensor field is denoted as: ; Among them, the weights satisfy the normalization condition. This tensor field not only preserves the local directional information of the front, but also enhances its adaptability to fronts of different intensities and widths through multi-scale fusion.

[0025] Then, step S4 is executed: using the multi-scale temperature gradient tensor field, the initial seed point set of the temperature front is automatically identified through a joint criterion of local extremum response and global structural saliency. This identification process first calculates the gradient magnitude of each grid point. Consistency index with gradient direction The directional consistency index is defined as the reciprocal of the standard deviation of the angles of all gradient directions within a 5×5 neighborhood centered at the current point, i.e. ,in For the first The gradient direction angles of the neighborhood points. High consistency means that the gradient directions within the neighborhood are highly parallel, conforming to the geometric characteristics of the linear structure of the front. Next, a gradient magnitude threshold is set. and direction consistency threshold Both are automatically determined from their respective global distributions using the non-parametric Otsu's method. Otsu's method divides the histograms of gradient magnitudes or consistency values ​​into two classes, maximizing the inter-class variance and thus obtaining the optimal segmentation threshold. Simultaneously satisfying... and The grid points are marked as candidate seed points. To avoid dense repetition, non-maximum suppression is performed on the candidate point set: within the 3×3 neighborhood of each candidate point, only the point with the largest gradient magnitude is retained, and the rest are discarded, ultimately forming a sparse and representative initial seed point set.

[0026] After obtaining the initial seed points, step S5 is executed: starting from the set of initial seed points, bidirectional chain growth is performed along the main direction of the temperature gradient to generate preliminary frontal segments. For each initial seed point, the gradient direction at its location is used to... The initial growth direction is used as the growth direction. The growth process is divided into two independent branches: forward and reverse. In each growth step, the algorithm searches for the next node in the 8-neighborhood of the current node that satisfies the following two conditions: first, the gradient magnitude of the neighboring point is a local maximum; second, the absolute value of the angle between its gradient direction and the current growth direction is less than 45 degrees. The growth step size is fixed at one grid cell, that is, it moves to the adjacent cell each time. If a neighboring point that simultaneously satisfies the maximum magnitude and direction constraints cannot be found in three consecutive steps, or if the new node already belongs to another frontal segment, the growth in that direction is terminated. The nodes obtained from the forward and reverse growth are connected sequentially to form an open or closed polyline, which is the initial frontal segment. This chain growth mechanism ensures that the frontal trajectory extends along the path with the strongest temperature gradient and the most consistent direction, effectively capturing the continuous geometric structure of the front.

[0027] After the initial frontal line segment is generated, step S6 is executed: Oceanic dynamics prior constraints and spatiotemporal continuity regularization terms are applied to the initial frontal line segment, and the optimal frontal trajectory is solved using an iterative optimization algorithm. This optimization process constructs a composite objective function. Its expression is: ; in, Indicates the first frontal trajectory The latitude and longitude coordinates of each node, The total number of nodes. express The complete frontal trajectory at any given moment. The first term is for data fidelity, requiring maximizing the sum of temperature gradient magnitudes at all points on the trajectory, with a coefficient... Control its weights. The second term is the dynamic smoothing term. Indicates at node The penalty for the deviation between the frontal trend and the background geostrophic current direction is applied. The background geostrophic current is obtained by inversion from contemporaneous sea level height anomaly remote sensing products: the spatial gradient of the sea level height anomaly is calculated, and then applied according to the geostrophic balance equation. (in For Coriolis parameters, It is the acceleration due to gravity. This is due to an anomaly in sea level. The geostrophic vector was obtained using the reference depth. Its direction is the reference for the permissible frontal direction. If the direction of the frontal tangent is... If the directional angle exceeds 30 degrees, a quadratic penalty will be applied. Otherwise, it is 0. The third term is the spatiotemporal continuity term. It measures whether the maximum displacement of the frontal position at adjacent moments exceeds the limit. Specifically, it calculates... and The spherical distance between all corresponding node pairs is used as the reference distance. If any distance exceeds the displacement corresponding to the maximum allowable propagation speed for the day, a penalty is applied. The maximum propagation speed is dynamically set according to latitude: 80 km / day in the equatorial region, 50 km / day in the mid-latitude region, and 30 km / day in the high-latitude region. (Coefficient) The strength of dynamic and spatiotemporal constraints are controlled separately. The objective function is minimized using a sequential quadratic programming algorithm, with the latitude and longitude coordinates of all nodes as variables. Iteration continues until the change in the objective function value is less than [a certain value]. It can reach the upper limit of 100 iterations and output the optimal frontal trajectory after convergence.

[0028] Finally, step S7 is executed: Based on the optimal frontal trajectory, the geometric and physical characteristic parameters of the temperature front are calculated and output. The positional parameters are directly represented by the latitude and longitude coordinate sequence of all nodes on the trajectory. The intensity parameter is calculated as the arithmetic mean of the temperature gradient amplitude at each node on the trajectory, and its standard deviation is output simultaneously to quantify the non-uniformity of frontal intensity along the front. The direction parameter is determined using principal component analysis: all node coordinates are treated as a two-dimensional point set, and the principal eigenvector of its covariance matrix is ​​calculated; the direction angle of this vector is the overall direction of the front. The curvature parameter is calculated at each internal node using a 3-point circle fitting method: the current node and its preceding and following adjacent nodes are taken, and a circle passing through these three points is fitted; the reciprocal of its radius is the local curvature. The number of curvature sign-changing points is simultaneously counted to identify frontal inflection points or vortex embedding structures. The lifecycle parameter records the total number of days from the first detection of the front in the time series to the last successful tracking. In addition, frontal type identification is performed: along the direction of the frontal trajectory, the average temperature within a 5-kilometer buffer zone on both the left and right sides is compared; if the average temperature on the right side is higher than that on the left side, it is determined to be a cold front; if the average temperature on the left side is higher than that on the right side, it is a warm front; if the absolute value of the temperature difference between the two sides is less than 0.5 degrees Celsius, it is classified as a weak front or a mixed front.

[0029] In the above implementation process, several steps possess adaptive adjustment capabilities. For example, in step S3, the scale parameter of the Gaussian kernel can be dynamically adjusted according to the typical mesoscale eddy scale of the target sea area. Specifically, the power spectral density of historical sea surface temperature data for this sea area is analyzed, and the wavelength corresponding to the peak energy spectrum is selected as the baseline scale. The other two scales are set to 50% and 150% of the baseline scale, respectively, thereby achieving adaptation to the regional ocean dynamic characteristics. In step S6, the background geostrophic data is strictly spatiotemporally aligned with the sea surface temperature data to ensure the physical consistency of the dynamic constraints. The propagation velocity threshold in the spatiotemporal continuity term also varies with latitude, reflecting the influence of the Earth's rotation on the eddy propagation velocity.

[0030] The entire methodology is fully automated, requiring no manual intervention to set thresholds or select seed points. The construction of multi-scale gradient tensor fields overcomes the limitations of single-scale representation of fronts of varying widths; the joint criterion of local extrema and global significance eliminates dependence on empirical thresholds; and the optimization framework integrating ocean dynamics and spatiotemporal continuity ensures that the tracking results are both faithful to observational data and conform to physical laws. The final output feature parameter system is structurally complete, rigorously defined, and has clear physical meaning, allowing direct input into operational marine environmental systems for fisheries resource hotspot prediction, research on ocean frontogenesis and dissipation processes, or numerical model assimilation. This significantly outperforms existing technologies that suffer from missed detections, fragmentation, or feature distortion due to parameter rigidity.

Claims

1. A method for automatic tracking and feature parameter extraction of ocean temperature fronts, characterized in that, include: Acquire continuous time-series remote sensing data of sea surface temperature covering the target sea area; The sea surface temperature remote sensing data is preprocessed to generate a standardized spatiotemporally consistent sea surface temperature field. Based on the standardized sea surface temperature field, a multi-scale temperature gradient tensor field is constructed. Using the multi-scale temperature gradient tensor field, the initial seed point set of temperature fronts is automatically identified by a joint criterion of local extreme response and global structural saliency. Starting from the initial set of seed points, bidirectional chain growth is carried out along the main direction of the temperature gradient to generate preliminary frontal segments; Ocean dynamics prior constraints and spatiotemporal continuity regularization terms are applied to the preliminary frontal line segment, and the optimal frontal trajectory is solved by an iterative optimization algorithm. Based on the optimal frontal trajectory, the geometric and physical characteristic parameters of the temperature front are calculated and output.

2. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 1, characterized in that, The acquisition of continuous time-series sea surface temperature remote sensing data covering the target sea area includes: Receive raw brightness temperature data from infrared radiometers carried by polar-orbiting or geostationary ocean satellites; The raw brightness temperature data is subjected to atmospheric correction, cloud pollution removal, and orbital stitching processing. The processed data will be resampled to a uniform latitude and longitude grid, with a time resolution of no less than once per day and a spatial resolution of no less than 1 / 4 degree.

3. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 2, characterized in that, The preprocessing of sea surface temperature remote sensing data to generate a standardized spatiotemporally consistent sea surface temperature field includes: Outlier detection and repair were performed on the resampled sea surface temperature data. A sliding window midpoint filter was used to identify and replace data points that deviated from more than 3 times the interquartile range. Spatiotemporal interpolation is performed on the missing regions, and anisotropic diffusion equations are used for smooth filling. The diffusion coefficient is adaptively adjusted according to the direction of the local temperature gradient. The entire time series is standardized by mean-variance so that the mean of the time series at each grid point is 0 and the variance is 1.

4. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 3, characterized in that, The construction of the multi-scale temperature gradient tensor field includes: On the standardized sea surface temperature field, Gaussian kernel functions with scale parameters of 0.5 degrees, 1 degree, and 1.5 degrees were used for convolution smoothing to generate three smooth temperature fields with different scales. For each smooth temperature field, its first-order partial derivative is calculated to obtain the temperature gradient components in the east-west and north-south directions. Two gradient components at the same scale are combined into a two-dimensional gradient vector, and their outer product is calculated to form a second-order temperature gradient tensor at that scale. The gradient tensors of the three scales are fused by scale weighting, with the weights determined by the signal-to-noise ratio of the local structure at each scale. Scales with higher signal-to-noise ratios are given higher weights.

5. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 4, characterized in that, The automatic identification of the initial seed point set of temperature fronts using a joint criterion of local extreme response and global structural saliency includes: In the fused multi-scale gradient tensor field, the gradient magnitude and gradient direction consistency index of each grid point are calculated; The gradient magnitude is defined as the Euclidean norm of the gradient vector, and the gradient direction consistency index is defined as the reciprocal of the standard deviation of the gradient direction in the neighborhood. Gradient magnitude threshold and direction consistency threshold are set, both of which are automatically determined from the global distribution using the non-parametric Otsu method; Grid points that simultaneously meet two threshold conditions are marked as candidate seed points; Non-maximum suppression is applied to the candidate seed points, retaining the points with the largest local gradient magnitude to form the final set of initial seed points.

6. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 5, characterized in that, The process of generating preliminary frontal segments by starting with an initial set of seed points and proceeding along the principal direction of the temperature gradient in a bidirectional chain includes: For each initial seed point, the gradient direction at its location is used as the initial growth direction; During the growth process, at each step, the grid point with the largest gradient magnitude and whose direction makes an angle of less than 45 degrees with the current growth direction in the neighborhood is searched forward as the next node. The growth step size is fixed at one grid cell; If a node that meets the conditions cannot be found in three consecutive steps, or if an existing frontal segment is encountered, growth in that direction is stopped. Perform growth in both directions for each seed point, connecting all nodes to form a closed or open initial frontal segment.

7. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 6, characterized in that, The process of applying ocean dynamics prior constraints and spatiotemporal continuity regularization terms to the initial frontal line segment, and solving for the optimal frontal trajectory through an iterative optimization algorithm, includes: An objective function is constructed, consisting of three terms: the first term is a data fidelity term, which measures the sum of the temperature gradient amplitudes at each point on the optimized frontal trajectory and aims to maximize it; the second term is a dynamic smoothing term, which introduces geostrophic constraints, requiring that the deviation between the frontal direction and the background geostrophic direction does not exceed 30 degrees, with a heavier penalty for larger deviations; the third term is a spatiotemporal continuity term, requiring that the rate of change of the frontal position between adjacent moments does not exceed the typical propagation speed of a mesoscale eddy in the ocean, i.e., no more than 50 kilometers per day. The objective function is minimized using a sequential quadratic programming algorithm, with the variables being the latitude and longitude coordinates of all nodes on the frontal trajectory; Iterate until the change in the objective function value is less than the preset convergence threshold, or until the maximum number of iterations is reached.

8. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 7, characterized in that, The background geostrophic data is derived from remote sensing products of sea surface height anomalies from the same period. It is obtained by calculating the gradient of sea surface height anomalies and applying the geostrophic balance equation for inversion, and the spatial resolution is consistent with the sea surface temperature field.

9. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 7, characterized in that, The typical propagation speed of the ocean mesoscale eddies is adjusted according to latitude, taking 80 km / day in the equatorial region, 50 km / day in the mid-latitude region, and 30 km / day in the high-latitude region.

10. The method for automatic tracking and feature parameter extraction of ocean temperature fronts according to claim 7, characterized in that, The calculation and output of geometric and physical characteristic parameters of the temperature front based on the optimal frontal trajectory includes: The location parameters are the latitude and longitude coordinates of all nodes on the frontal trajectory; The intensity parameter is the average value of the temperature gradient amplitude at each node on the trajectory; The directional parameter is the principal axis direction of the entire trajectory, which is calculated using principal component analysis. The curvature parameter is the reciprocal of the local radius of curvature at each node on the trajectory, and the radius of curvature is calculated by the 3-point circle fitting method; The lifecycle parameter is the time span from the first detection of the front to the last tracking of the front, in days.

Images