Marine ship association identification method based on AIS and satellite-borne SAR data and electronic equipment

Abstract

The invention relates to the technical field of remote sensing data ocean application, and provides an ocean ship association identification method based on AIS and satellite-borne SAR data and electronic equipment, and the method constructs a full-process technical route from data acquisition, preprocessing, ship target detection to trace point association. The method runs through the whole process from SAR data and AIS trajectory preprocessing, intelligent identification of ship targets in SAR images and optimal matching of plots of the SAR data and the AIS trajectories, and a set of uniform, cooperative and efficient technical links is formed. According to the SAR-AIS fusion monitoring method, accurate alignment of AIS data and SAR imaging time is realized by introducing cubic spline interpolation, a ship target in an SAR image is automatically extracted in combination with a deep learning model, finally, global optimal association between trace points is completed by means of a Munkres algorithm, links from data preprocessing to result output are seamlessly connected, and the precision and efficiency of SAR-AIS fusion monitoring are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing data ocean applications, and in particular, to a method and an electronic device for associating and identifying ocean vessels based on AIS and spaceborne SAR data. Background Art

[0002] Spaceborne SAR (Synthetic Aperture Radar, SAR, spaceborne synthetic aperture radar) is an active remote sensing device that uses radar imaging technology to monitor the sea surface, capable of obtaining high-resolution images and identifying the positions and contours of vessels. And AIS (Automatic Identification System, AIS, vessel automatic identification system) is an automatic identification signal automatically sent by vessels, containing information such as position and speed, but depending on whether the vessel's equipment is turned on and the signal coverage range. The relationship between the two is as follows: (1) Vessel Detection and Identity Verification Wide-area monitoring by SAR: Spaceborne SAR can scan the sea surface over a large range and detect vessels without AIS turned on (such as illegal fishing vessels and smuggling vessels), and these vessels may evade supervision by turning off AIS.

[0003] Identity matching by AIS: Through the vessel identity and trajectory data provided by AIS, the targets detected by SAR can be quickly verified whether they are legal vessels, reducing false positives (such as distinguishing merchant ships from icebergs and sea clutter).

[0004] (2) Identification of Non-cooperative Targets (Dark Targets) Discovering "stealth" vessels: Some vessels deliberately turn off AIS or forge information (such as illegal fishing and pirate activities), and SAR can independently discover these targets through radar echoes, filling the regulatory loopholes of AIS.

[0005] Data fusion to improve accuracy: By superimposing the SAR image and AIS data, it can be verified whether the reported trajectory of the vessel is true (for example: AIS shows the vessel at point A, but SAR detects it at point B).

[0006] In current maritime safety monitoring and maritime management work, both Synthetic Aperture Radar (SAR) and Automatic Identification System (AIS) play irreplaceable roles. Due to its all-weather and all-time observation capabilities, SAR technology can capture detailed images of the sea surface and ships through electromagnetic wave imaging under complex meteorological conditions and harsh sea states. At the same time, the AIS system relies on the position information, ship identity, and dynamic data transmitted by ships themselves through radio frequency bands to provide real-time data support for maritime traffic regulation, collision avoidance warning, and ship status monitoring. In the existing technology, the industry mostly uses separate SAR or AIS data for target detection and tracking, but single data sources have their respective limitations: Although SAR images have high spatial resolution, they are greatly affected by data processing algorithms in target recognition, classification, and behavior analysis, while AIS data is subject to signal noise, data loss, or time errors and has blind spots at sea.

[0007] To make up for the deficiencies of single data sources, many studies have attempted to jointly apply SAR and AIS data, leveraging their complementary advantages in spatio-temporal information to achieve the sharing and cross-calibration of target information. Some existing systems have adopted steps such as data preprocessing, target detection, and tracklet matching to initially achieve data fusion. For example, the time information is synchronized by simply interpolating the AIS data, or ship targets in SAR images are detected through traditional image processing algorithms, and a heuristic matching algorithm is used to complete the initial association. However, such methods often fail to achieve the expected accuracy and real-time performance in practical applications and do not fully solve problems such as the rapid movement of targets, large data noise, and matching confusion in a multi-target environment in the dynamic maritime environment. At the same time, due to the lack of systematic integration among various methods, the entire data fusion processing flow is scattered, and the cooperation degree between modules is relatively low, making it difficult to meet the application requirements of large-scale maritime monitoring and intelligent warning.

[0008] Currently, in the process of the fusion application of SAR and AIS data, the lack of a complete, reliable, and accurate data fusion full-process system limits the ability of the maritime monitoring system to quickly respond and accurately judge multi-ship targets in harsh environments. There are mainly the following several technical problems: (1) There is often a problem that the timestamp in the AIS data is inconsistent with the actual SAR imaging time during the acquisition and transmission process. Due to the non-linear and variable nature of the operating trajectories of ships at sea, using traditional linear or simple interpolation methods for time correction often fails to fully reflect the actual motion state of ships, thus introducing additional positioning errors; in addition, due to the influence of the marine radio signal environment, some AIS data may also be interrupted or delayed, further affecting the accuracy of data spatial matching.

[0009] (2) Ship target detection in SAR images often relies on traditional detection algorithms, such as the constant false alarm detection method based on statistical characteristics. This type of method is prone to false alarms or missed detections under complex sea conditions, and has limited ability to suppress background noise, resulting in detection accuracy and reliability that are difficult to meet actual needs.

[0010] (3) Existing data association methods mostly use local heuristic algorithms or nearest neighbor matching strategies. This method cannot guarantee global optimization in a multi-target environment and there is a risk of target mismatching or repeated matching, which in turn affects subsequent target tracking and behavior analysis. Summary of the Invention

[0011] The purpose of the present invention is to overcome the above technical deficiencies and provide a method and electronic equipment for ocean vessel association identification based on AIS and spaceborne SAR data, so as to solve the problem of lack of complete, reliable and accurate SAR and AIS data fusion method in related technologies.

[0012] In order to achieve the above technical objectives, the present invention adopts the following technical solutions: According to a first aspect of the present invention, a method for associating and identifying ocean vessels based on AIS and spaceborne SAR data is provided, comprising: Preprocess SAR data of different levels to obtain standardized SAR products in which each pixel carries latitude, longitude and elevation information with a preset accuracy not less than that of the preset level. Obtain the original AIS data table containing timestamp, longitude, and latitude information, and use the Vondrak algorithm to remove abnormal data points in the original AIS data table; The cleaned AIS data points are converted into AIS point vector files, and the discontinuous tracks of different ships are extracted. For each discontinuous track of a ship, the longitude and latitude coordinates at the time of SAR imaging are estimated based on the cubic spline interpolation method to form an AIS measurement set. Identify ship targets in SAR images and calculate the latitude and longitude coordinates of each ship target to form a SAR measurement set; The SAR measurement set and the AIS measurement set are paired to obtain a set of AIS traces and SAR traces within a preset association range, and a cost matrix of the SAR traces and AIS traces in the set is calculated. The cost matrix is optimally allocated based on the Munkres algorithm to obtain a trace association result.

[0013] Preferably, the removing of abnormal data points in the original AIS data table by using the Vondrak algorithm includes: Sort the raw data points in the AIS data table by timestamp to ensure time continuity; Use polynomial or spline functions to fit the original data points to obtain an initial smooth curve; Calculate the deviation between the original data points and the current smoothed curve. When calculating for the first time, assign the current smoothed curve as the initial smoothed curve. Optimize the coefficients of the current smoothed curve according to the deviation to minimize the energy function Q. Take the current smoothed curve corresponding to the minimum value of the energy function Q as the reference AIS trajectory sequence. Calculate the deviation between the original data points and the reference AIS trajectory sequence. If the deviation exceeds the set threshold, mark it as an abnormal data point and delete it.

[0014] Preferably, converting the cleaned AIS data points into an AIS position vector file and extracting the discontinuous trajectories of different vessels includes: Import the cleaned AIS data points into the Geographic Information System (GIS) ArcGIS. According to the timestamp, longitude, and latitude information of each AIS data point, map the AIS data points onto the map to directly observe the spatio-temporal distribution of vessel movements. Through vectorization in ArcGIS, classify the characteristics of the cleaned data according to the relative position relationship of the vessels and the vessel type attributes, so as to aggregate the scattered points into independent trajectories of different vessels. The vessel type attributes at least include: MMSI code, vessel speed, and course.

[0015] Preferably, for the discontinuous trajectory of each vessel, estimating the longitude and latitude coordinate information at the SAR imaging moment based on the cubic spline interpolation method to form an AIS measurement set, including: Estimate the longitude coordinate information and latitude coordinate information at the SAR imaging moment based on the cubic spline interpolation method respectively. Among them, the methods for estimating the longitude coordinate information and latitude coordinate information are the same, including: Assume that the longitude of the vessel is a function of time and satisfies the cubic spline function (1), then the velocity component v of the vessel in the longitude direction y is the first derivative of, and the acceleration component a of the vessel in the longitude direction y is the second derivative of; According to the boundary conditions, at and moments, it should satisfy: (2) Substitute the boundary values , , , Substituting into Equation (2), the values of the coefficients a, b, c, and d of the cubic spline function can be obtained. When calculating the longitude value at any moment within the time period [t1, t2], substitute the corresponding moment t into Equation (1).

[0016] Preferably, the identification of vessel targets in the SAR image includes: Collect a large number of SAR images with vessel markings to train the Faster R-CNN model, so that the trained model can learn the features of vessels, including: shape, texture, and size; Input the SAR image to be detected into the trained model. The convolutional neural network in the model divides the entire SAR image to be detected into countless small blocks, analyzes the details of each small block, and determines which small blocks have suspected vessel contours; The region proposal network in the model generates candidate boxes that may contain the target vessel in the small blocks with suspected vessel contours according to the learned vessel features; The fully connected layer in the model determines whether there is really a vessel in these candidate boxes. If so, it adjusts the position and size of the candidate boxes to make the candidate boxes become SAR bounding boxes that fit the actual contour of the vessel better; The model outputs all adjusted SAR bounding boxes and gives the coordinates of each SAR bounding box.

[0017] Preferably, the calculation of the longitude and latitude coordinate information of each vessel target to form a SAR measurement set includes: For each identified vessel target, define the center pixel point coordinates of its SAR bounding box as the center position of the vessel target in the image space; Use the RPC model or geometric correction parameters of the SAR image to convert the center pixel point coordinates into geographic coordinates, and finally form a SAR measurement set.

[0018] Preferably, the pairing of the SAR measurement set and the AIS measurement set to obtain a set of AIS traces and SAR traces within a preset association range includes: The data in the SAR measurement set includes: the SAR longitude and latitude coordinate information of the detected vessel target and the SAR bounding box of the vessel target; among them, each SAR bounding box corresponds to a SAR longitude and latitude coordinate information, and a SAR longitude and latitude coordinate information represents a SAR trace; The data in the AIS measurement set includes: the movement trajectory of each vessel, and the movement trajectory is represented by a series of timestamps and the AIS longitude and latitude coordinate information corresponding to the timestamps; among them, each AIS longitude and latitude coordinate information corresponds to an AIS trace; For each SAR bounding box, find all AIS track points within a preset range near it, and obtain a set of AIS and SAR traces within the preset association range.

[0019] Preferably, calculate the cost matrix of SAR and AIS traces in the set, and perform optimal allocation on the cost matrix based on the Munkres algorithm to obtain the trace association result, including: Assume there are n SAR traces in the SAR measurement set and m AIS traces in the AIS measurement set. The cost matrix c is expressed as follows:

[0020] Then the cost coefficient of the i-th SAR trace and the j-th AIS trace is , and its value is , where represents the longitude and latitude of AIS, represents the longitude and latitude of SAR; Perform global minimum distance trace matching according to the cost matrix. The minimum value , where is a binary function. If it is , it means the i-th SAR trace is associated with the j-th AIS trace; if it is , it means the i-th SAR trace is not associated with the j-th AIS trace; Judge whether the Euclidean distance of the longitudes and latitudes of the associated trace pairs is within the set judgment threshold. If it is within the judgment threshold, the trace association is successful, and the trace association result is saved; otherwise, the association fails, and the trace association result is deleted to complete the trace matching.

[0021] Preferably, the method further includes: Generate a SAR-AIS point position attribute information association table, including: the correspondence between AIS traces and SAR traces, and the attribute information of each pair of associated traces, to ensure the traceability and analyzability of data; And / or, Generate a SAR-AIS point position spatial distribution map, and draw an association map of AIS traces and SAR traces through visualization means to show the correspondence between ship targets in SAR images and AIS ship trajectories; And / or, Generate a SAR-AIS point position spatial association map, show the spatial distribution of ship targets in SAR images and the corresponding AIS traces, and highlight the correlation between the two through color annotation and connection lines.

[0022] According to the second aspect of the present invention, an electronic device is provided, including: A processor and a memory; The memory is used to store a program, and the processor is used to run the program to implement the above method.

[0023] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: It can be understood that for the technical solution provided by the present invention, firstly, in the data preprocessing stage, aiming at the time and space matching problems of AIS data, a cubic spline interpolation method is adopted. By performing high-order interpolation processing on AIS data, the longitude and latitude positions of ships at the SAR imaging moment are accurately predicted, thereby effectively making up for the deficiencies of traditional interpolation methods in non-linear motion capture.

[0024] Secondly, for SAR image target detection, the present invention introduces a target recognition technology based on deep learning. By extracting the ship features in the SAR image, ship targets are accurately recognized.

[0025] Furthermore, in the data association stage, through the matching of SAR traces and AIS traces, a global optimal allocation strategy based on the Munkres algorithm is adopted. By constructing a cost matrix, the matching relationship between the two data sources is abstracted into an optimization problem, and the Munkres algorithm (i.e., the Hungarian algorithm) is used to perform global optimal allocation on each possible matching scheme.

[0026] The present invention constructs a full-process technical route from data acquisition, preprocessing, ship target detection to trace association. This method runs through the whole process from the preprocessing of SAR data and AIS trajectories, the intelligent recognition of ship targets in SAR images, to the optimal matching of their traces, forming a unified, coordinated and efficient technical link. By introducing cubic spline interpolation to achieve the accurate alignment of AIS data and the SAR imaging moment, combined with a deep learning model to automatically extract ship targets in SAR images, and finally using the Munkres algorithm to complete the global optimal association between traces, all links from data preprocessing to result output are seamlessly connected, significantly improving the accuracy and efficiency of SAR–AIS fusion monitoring, and promoting the development of maritime target monitoring technology towards a process-oriented and intelligent direction. Description of the Drawings

[0027] Figure 1 is a flowchart of a method for associating and recognizing marine vessels based on AIS and spaceborne SAR data shown according to an exemplary embodiment; Figure 2 is a schematic diagram of the recognition result of a SAR ship target shown according to an exemplary embodiment; Figure 3 is a schematic diagram of the output result of a method for associating and recognizing marine vessels based on AIS and spaceborne SAR data shown according to an exemplary embodiment; Figure 4The figure is a schematic diagram showing the output results of a method for ocean vessel association and identification based on AIS and spaceborne SAR data according to an exemplary embodiment. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0029] Example 1 Figure 1 This is a flow chart showing a method for associating and identifying ocean vessels based on AIS and spaceborne SAR data according to an exemplary embodiment. Figure 1 , the method comprising: Step S11: pre-processing SAR data of different levels to obtain standardized SAR products in which each pixel carries latitude, longitude, and elevation information with a preset accuracy or higher; Step S12: obtaining an original AIS data table containing timestamp, longitude, and latitude information, and using the Vondrak algorithm to remove abnormal data points in the original AIS data table; Step S13: Convert the cleaned AIS data points into an AIS point vector file, extract the discontinuous tracks of different ships, and estimate the longitude and latitude coordinate information of each ship's discontinuous track at the time of SAR imaging based on the cubic spline interpolation method to form an AIS measurement set; Step S14: Identify the ship targets in the SAR image and calculate the latitude and longitude coordinate information of each ship target to form a SAR measurement set; Step S15: Pair the SAR measurement set and the AIS measurement set to obtain a set of AIS traces and SAR traces within a preset association range, calculate the cost matrix of the SAR traces and AIS traces in the set, and optimally allocate the cost matrix based on the Munkres algorithm to obtain a trace association result.

[0030] It should be noted that, in practice, the technical solution provided in this embodiment is loaded and executed in electronic devices. Such electronic devices include, but are not limited to, tablet computers, computers, and smartphones. This technical solution not only achieves dual improvements in accuracy and efficiency in every key step of data processing and matching, but also has excellent scalability and application prospects, and can be widely used in various fields such as maritime monitoring, ship management, and emergency response.

[0031] In step S11, SAR data at different levels refer to different grades of synthetic aperture radar (SAR) data products divided according to the processing degree and information content. These levels usually include SLC (single look complex), GRD (ground range), and GEC (geocoded), etc. There are significant differences in geometric accuracy, radiometric correction degree, and spatial resolution among data at each level. Therefore, targeted processing is required in the data processing flow to ensure that the final generated SAR products have high-precision geographic coordinate information.

[0032] SLC (single look complex) data: This is the original format of SAR data, containing the most complete radar echo information, but complex processing is required to be used. SLC data usually has a high resolution, but lacks radiometric correction and geocoding. Therefore, further processing is needed to convert it into available geographic information.

[0033] GRD (ground range) data: GRD data is SAR data that has been radiometrically corrected and geocoded, usually expressed in units of ground range. This data has undergone preliminary processing, has good radiometric consistency and georeference information, but may still require further geometric correction to improve accuracy.

[0034] GEC (geocoded) data: GEC data is a higher-level SAR data product, which usually has completed radiometric correction, geometric correction, and geocoding, and can be directly used in geographic information systems (GIS) and other applications. The spatial resolution of GEC data may be low, but it has high geometric accuracy and georeference information.

[0035] In specific practices, corresponding data processing flows need to be formulated for SAR data at different levels. For example: SLC data: It is necessary to perform radiometric calibration, speckle noise suppression, and geocoding, etc., to convert it into GRD or GEC data.

[0036] GRD data: Further geometric correction and geocoding are required to improve its spatial accuracy.

[0037] GEC data: Only simple post-processing, such as data cropping or format conversion, is required to generate a standardized SAR product.

[0038] Through these processing steps, the ultimate goal is to generate a standardized SAR product with high-precision geographic coordinate information (each pixel carries high-precision longitude, latitude coordinates, and elevation information) for effective fusion and application with AIS data (which also has attributes such as longitude, latitude, speed, and heading).

[0039] In the prior art, when fusing SAR and AIS data, single links are often optimized in isolation (such as interpolation, detection, or association), while the systematic coupling of the entire process is ignored. For example, the AIS interpolation error will directly lead to an increase in the spatial search radius for SAR-AIS association, thereby increasing the computational burden of the algorithm; the missed detection of SAR ship target recognition may cause the breakage of the association chain. This fragmented technical route leads to problems such as the progressive transmission of accuracy loss and low resource utilization rate in practical applications. In addition, existing achievements mostly rely on the offline batch processing mode, which is difficult to meet the timeliness requirements of regional real-time monitoring tasks.

[0040] The deficiencies of the prior art indicate that there is an urgent need for a full-chain technical system that deeply integrates spatio-temporal synchronization, intelligent recognition, and global optimization of association to achieve the coordinated improvement of the accuracy and efficiency of marine target monitoring.

[0041] It can be understood that for the technical solution provided in this embodiment, first, in the data preprocessing stage, for the time and space matching problem of AIS data, the cubic spline interpolation method is adopted. By performing high-order interpolation processing on AIS data, the longitude and latitude positions of ships at the SAR imaging moment are accurately predicted, thereby effectively making up for the deficiencies of traditional interpolation methods in non-linear motion capture.

[0042] Secondly, for SAR image target detection, the present invention introduces a target recognition technology based on deep learning. By extracting the ship features in the SAR image, ship targets are accurately recognized.

[0043] Furthermore, in the data association stage, through the matching of SAR traces and AIS traces, a global optimal allocation strategy based on the Munkres algorithm is adopted. By constructing a cost matrix, the matching relationship between the two data sources is abstracted into an optimization problem, and the Munkres algorithm (i.e., the Hungarian algorithm) is used to perform global optimal allocation on each possible matching scheme.

[0044] This embodiment constructs a full-chain technical route from data acquisition, preprocessing, ship target detection to trace association. This method runs through the whole process from the preprocessing of SAR data and AIS trajectories, the intelligent recognition of ship targets in SAR images, to the optimal matching of their traces, forming a unified, coordinated and efficient technical link. By introducing cubic spline interpolation to achieve the accurate alignment of AIS data with the SAR imaging moment, combined with the deep learning model to automatically extract ship targets in SAR images, and finally using the Munkres algorithm to complete the global optimal association between traces, all links from data preprocessing to result output are seamlessly connected, significantly improving the accuracy and efficiency of SAR-AIS fusion monitoring, and promoting the development of maritime target monitoring technology towards a process-oriented and intelligent direction.

[0045] In specific practice, in step S12, the Vondrak algorithm is used to eliminate abnormal data points in the original AIS data table, including: Sort the original data points in the AIS data table according to the time stamp to ensure continuous time; Use a polynomial or spline function to fit the original data points to obtain an initial smoothed curve; Calculate the deviation between the original data points and the current smoothed curve. When calculating for the first time, assign the current smoothed curve to the initial smoothed curve; According to the deviation, optimize the coefficients of the current smoothed curve to minimize the energy function Q; Take the current smoothed curve corresponding to the minimum of the energy function Q as the reference AIS trajectory sequence; Calculate the deviation between the original data points and the reference AIS trajectory sequence. If the deviation exceeds the set threshold, mark it as an abnormal data point and delete it.

[0046] It should be noted that the energy function Q of the Vondrak algorithm belongs to the prior art, and the specific formula of the energy function Q will not be listed in this embodiment.

[0047] It can be understood that the detailed steps of preprocessing based on AIS time-latitude and longitude data using the Vondrak data smoothing method are aimed at deleting incorrect data and generating a coherent ship trajectory sequence. This method removes noise through mathematical optimization while retaining the true motion trend.

[0048] In specific practice, in step S13, the cleaned AIS data points are converted into an AIS point vector file to extract discontinuous trajectories of different ships, including: Step S131: Import the cleaned AIS data points into the geographic information system ArcGIS. According to the time stamp, longitude, and latitude information of each AIS data point, map the AIS data points onto the map to directly observe the spatio-temporal distribution of ship movements; Step S132: Through ArcGIS vectorization, classify the cleaned data according to the relative position relationship of the ships and the ship type attributes, so as to aggregate the scattered points into independent trajectories of different ships. The ship type attributes at least include: MMSI code, ship speed, and heading.

[0049] It can be understood that importing the cleaned AIS data points into the geographic information system (ArcGIS) is to utilize its powerful spatial analysis, visualization, and automated processing capabilities to achieve the following core objectives: 1. Data visualization and interactive analysis Intuitive display of trajectories: Map AIS point data (longitude, latitude + time) onto a map to directly observe the spatio-temporal distribution of ship movements. For example, identify areas where ship trajectories are unusually dense (possibly intersection points of shipping lanes) or suddenly interrupted (possibly signal loss areas).

[0050] Interactive exploration: View attributes (such as ship speed, heading, MMSI code) by clicking on point features to quickly locate problem data (such as abnormal points with excessive speed).

[0051] 2. Spatial analysis and trajectory modeling Verification of kinematic constraints: Use ArcGIS spatial analysis tools (such as Near, Buffer) to automatically detect trajectory segments that do not conform to physical laws: Abnormal speed: Calculate the distance / time ratio between adjacent points to filter out points with excessive speed or stagnation (such as a cargo ship with a speed exceeding 20 knots).

[0052] Sudden change in heading: Detect sharp turns through the change in direction angle (such as a sudden change in heading from 0° to 180°).

[0053] Trajectory interpolation and completion: Perform spatial interpolation (such as cubic spline interpolation) on missing data (such as during signal loss) to generate continuous trajectories.

[0054] 3. Automated classification and feature extraction Grouping based on attributes: Automatically classify the trajectories of different ships according to attributes such as MMSI code, ship type (inferred from ship speed / heading), etc. For example, filter out the trajectories of oil tankers with all MMSI codes starting with "123".

[0055] It can be understood that even after removing abnormal points, there may still be time discontinuities in AIS data (such as ships briefly turning off AIS devices, signal occlusion, etc.), resulting in scattered trajectory segments of the same ship. In step S132, the cleaned data is classified according to features such as relative position relationships (such as the reasonableness of the spacing between adjacent points) and ship type attributes (such as MMSI code, speed, heading consistency) through ArcGIS vectorization, with the aim of aggregating scattered points into independent trajectories of different ships. For example, two ships may be mis-associated due to spatial proximity and need to be separated through classification.

[0056] In specific practice, for the discontinuous trajectories of each ship in step S13, the longitude and latitude coordinate information at the SAR imaging moment is estimated based on the cubic spline interpolation method to form an AIS measurement set, including: Estimate the longitude coordinate information and latitude coordinate information at the SAR imaging moment respectively based on the cubic spline interpolation method. Among them, the methods for estimating longitude coordinate information and latitude coordinate information are the same, including: Assume the longitude of the ship is a function of time and satisfies the cubic spline function within the time period [t1, t2]. (1), then the velocity component v of the ship in the longitude direction y is the first derivative of, and the acceleration component a of the ship in the longitude direction y is the second derivative of; According to the boundary conditions, at and the following should be satisfied at the moment: (2) Substitute the boundary values , , , into formula (2), and the numerical values of the coefficients a, b, c, and d of the cubic spline function can be obtained. When the longitude value at any moment within the time period [t1, t2] needs to be calculated, substitute the corresponding moment t into formula (1).

[0057] It can be understood that after classification in step S132, there may still be time breaks in the trajectory of each ship (such as data loss before and after the SAR imaging moment). In step S133, cubic spline interpolation estimates the continuous longitude and latitude of each ship at the SAR imaging moment by constructing a smooth function (piecewise fitting of cubic polynomials that satisfy the boundary conditions), ensuring the integrity of the trajectory in the spatio-temporal dimension.

[0058] There is a time difference between the SAR imaging moment and the AIS sampling moment (for example, SAR is instantaneous imaging and AIS is updated periodically). Cubic spline interpolation can more accurately restore the dynamic position of the ship at the moment of SAR imaging, reducing the positioning error caused by simple linear interpolation. Compared with piecewise linear interpolation, cubic spline interpolation avoids trajectory mutations by constraining the continuity of the second derivative, which is more in line with the physical laws of ship navigation (such as smooth transition when turning).

[0059] In specific practice, identifying ship targets in the SAR image in step S14 includes: Collect a large number of SAR images with ship markings to train the Faster R-CNN model, so that the trained model can learn the characteristics of ships, including: shape, texture, size; Input the SAR image to be detected into the trained model. The convolutional neural network in the model divides the entire SAR image to be detected into countless small blocks, analyzes the details of each small block, and determines which small blocks have suspected ship contours; The region proposal network in the model generates candidate boxes that may contain the target ship in the small blocks with suspected ship contours according to the learned ship characteristics; The fully connected layer in the model determines whether there are really ships in these candidate boxes. If so, it adjusts the position and size of the candidate boxes to make the candidate boxes become SAR bounding boxes that fit the actual contour of the ships better. The model outputs all the adjusted SAR bounding boxes and gives the coordinates of each SAR bounding box.

[0060] For the sake of easy understanding, the basic principle of using the Faster R-CNN model for ship target detection is now explained in popular language: Suppose there is a sea area in the SAR image, in which there are several cargo ships and a pile of sea wave noises: The model first "scans" the whole image and finds several suspicious areas (such as there are regular shapes around a certain bright spot).

[0061] Then it checks these areas one by one and excludes the parts that look like sea waves or noises.

[0062] Finally, candidate boxes are drawn at the correct positions, marking the positions and sizes of each ship.

[0063] The advantages of using the Faster R-CNN model for ship target detection are as follows: 1. Strong anti-noise ability: There will be speckle noises in SAR images, but the model can distinguish noises and real targets after training.

[0064] 2. Adapt to different scales: Ships may be very small (fishing boats) or very large (cargo ships), and the model can automatically adjust the detection range.

[0065] 3. Fast speed: Faster R-CNN is faster than the early detection models and is suitable for real-time or batch processing of satellite data.

[0066] In specific practice, in step S14, the longitude and latitude coordinate information of each ship target is calculated to form a SAR measurement set, including: For each identified ship target, the center pixel point coordinates of its SAR bounding box are defined as the center position of the ship target in the image space; Using the RPC model or geometric correction parameters of the SAR image, the center pixel point coordinates are converted into geographic coordinates, and finally a SAR measurement set is formed.

[0067] In specific practice, in step S15, the SAR measurement set and the AIS measurement set are paired to obtain a set of AIS traces and SAR traces within the preset association range, including: The data in the SAR measurement set includes: the SAR longitude and latitude coordinate information of the detected ship target and the SAR bounding box of the ship target; among them, each SAR bounding box corresponds to a SAR longitude and latitude coordinate information, and a SAR longitude and latitude coordinate information represents a SAR trace. The data in the AIS measurement set includes: the movement trajectories of each vessel, where the movement trajectories are represented by a series of timestamps and the AIS longitude and latitude coordinate information corresponding to the timestamps; among them, each AIS longitude and latitude coordinate information corresponds to an AIS trace point; For each SAR bounding box, find all AIS trajectory points within a preset range near it, and obtain a set of AIS trace points and SAR trace points within the preset association range.

[0068] In specific practice, in step S15, calculate the cost matrix of the SAR trace points and AIS trace points in the set, and perform optimal assignment on the cost matrix based on the Munkres algorithm to obtain the trace point association result, including: Assume that there are n SAR trace points in the SAR measurement set and m AIS trace points in the AIS measurement set. The cost matrix c is expressed as follows:

[0069] Then the cost coefficient of the i-th SAR trace point and the j-th AIS trace point is , and its value is , where represents the longitude and latitude of AIS, represents the longitude and latitude of SAR; Perform global minimum distance trace point matching according to the cost matrix. The minimum value , where is a binary function. If it is , it means that the i-th SAR trace point is associated with the j-th AIS trace point; if it is , it means that the i-th SAR trace point is not associated with the j-th AIS trace point; Judge whether the Euclidean distance of the longitudes and latitudes of the associated trace point pairs is within the set judgment threshold. If it is within the judgment threshold, the trace point association is successful, and the trace point association result is saved; otherwise, the association fails, and the trace point association result is deleted to complete the trace point matching.

[0070] Preferably, the method further includes: Generate a SAR-AIS point position attribute information association table, including: the correspondence between AIS trace points and SAR trace points, and the attribute information of each pair of associated trace points, to ensure the traceability and analyzability of data; and / or, Generate a SAR-AIS point position spatial distribution map, and draw an association map of AIS trace points and SAR trace points through visualization means to display the correspondence between vessel targets in the SAR image and AIS vessel trajectories; and / or, Generate a spatial association map of SAR-AIS points, display the spatial distribution of ship targets in the SAR image and the corresponding AIS traces, and highlight the correlation between the two through color annotation and connecting lines.

[0071] It can be seen that in this embodiment, by organically combining cubic spline interpolation, deep learning target recognition, and the Munkres global optimal assignment algorithm, an efficient and accurate SAR–AIS full-process data association method is formed. Compared with the prior art, the present invention has the following significant advantages: (1) Improve data synchronization accuracy: Use the cubic spline interpolation method to perform time synchronization processing on AIS data, so that the AIS data is highly matched with the SAR imaging time, accurately predict the position of the ship at the imaging moment, and significantly improve the spatio-temporal accuracy of data fusion.

[0072] (2) Enhance the accuracy of target detection: Introduce deep learning technology and use a convolutional neural network to automatically detect ship targets in the SAR image. This method shows high detection accuracy and robustness in complex sea conditions and backgrounds, and overcomes the problems of high false alarm rate and serious missed detection in traditional methods.

[0073] (3) Achieve global optimal target association: Use the Munkres algorithm to perform global optimal matching on SAR detection targets and AIS reported targets, ensure optimal association in a multi-target environment, avoid possible incorrect matching or omission in traditional methods, and improve the reliability of target tracking and monitoring.

[0074] (4) Construct a complete technical process: Integrate the three links of data preprocessing, target detection, and target association into one, form a complete technical route from data acquisition to result output, improve the automation degree and processing efficiency of the system, and meet the needs of real-time monitoring of large-scale sea areas.

[0075] In order to verify the effectiveness and feasibility of the technical solution provided in this embodiment, the following is an example based on the SAR data of a certain sea area taken by the SIXIANG AS01 satellite on January 19, 2025 and the corresponding AIS data to illustrate a SAR–AIS full-process data association method based on deep learning and the Munkres algorithm.

[0076] First, based on the SAR data imaging satellite, namely AS-01 and the SAR data image name "AS01_SAR_SS_HH_JH_008359_20250119141947_001_019_L2_00001399", it is determined that the current SAR data is in strip mode, L2-level (GEC) product, and corresponding preprocessing is carried out for the GEC product.

[0077] Data within 2 hours before and after the imaging time within the SAR area is acquired. After processing, vectorizing, and classifying using the Vondrak data smoothing method, 11 discontinuous AIS trajectories are retained. The cubic spline interpolation method is used to obtain the trajectory at any time, and the longitude and latitude of the AIS position at the SAR imaging time are calculated and vectorized.

[0078] The Faster R-CNN model is used to extract ship targets from the processed AS-01 SAR image data. Combining with the geographic information of the image, the longitude and latitude coordinate information of the ship targets is obtained. Finally, based on Arcgis, the coordinate information is vectorized and overlaid on the SAR image for display. The processing results are as Figure 2 shown. The ship position information on the SAR image extracted based on deep learning is shown in Table 1.

[0079] Table 1 Ship position information on the SAR image extracted based on deep learning

[0080] The processed and prepared AIS and SAR measurement sets are used as the input of the Munkres algorithm. The threshold is set to 1 km, and the points within the threshold are output as successfully associated points. The point track association result is saved. Thus, the entire process is completed. The final results output by a SAR–AIS full-process data association method based on deep learning and the Munkres algorithm are as Figure 3 and Figure 4 shown.

[0081] Embodiment 2 An electronic device according to an exemplary embodiment includes: A processor and a memory; The memory is used to store a program, and the processor is used to run the program to implement the above method.

[0082] It can be understood that for the technical solution provided in this embodiment, first, in the data preprocessing stage, for the time and space matching problem of AIS data, the cubic spline interpolation method is adopted. By performing high-order interpolation processing on the AIS data, the longitude and latitude positions of the ship at the SAR imaging time are accurately predicted, thus effectively making up for the deficiency of the traditional interpolation method in non-linear motion capture.

[0083] Secondly, for SAR image target detection, the present invention introduces a target recognition technology based on deep learning. By extracting the ship features in the SAR image, the ship targets are accurately recognized.

[0084] Furthermore, in the data association stage, through the matching of SAR traces and AIS traces, a global optimal allocation strategy based on the Munkres algorithm is adopted. By constructing a cost matrix, the matching relationship between the two data sources is abstracted into an optimization problem, and the Munkres algorithm (i.e., the Hungarian algorithm) is used to perform global optimal allocation for each possible matching scheme.

[0085] In this embodiment, a full-process technical route from data acquisition, preprocessing, ship target detection to trace association is constructed. This method runs through the whole process from the preprocessing of SAR data and AIS trajectories, the intelligent recognition of ship targets in SAR images, to the optimal matching of their traces, forming a unified, collaborative and efficient technical link. By introducing cubic spline interpolation, the AIS data is accurately aligned with the SAR imaging time. Combining with a deep learning model, ship targets in SAR images are automatically extracted. Finally, with the help of the Munkres algorithm, the global optimal association between traces is completed. Each link from data preprocessing to result output is seamlessly connected, significantly improving the accuracy and efficiency of SAR–AIS fusion monitoring and promoting the development of maritime target monitoring technology towards a process-oriented and intelligent direction.

[0086] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0087] If the integrated unit in the above embodiments is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in the above computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing one or more computer devices (which can be personal computers, servers or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.

[0088] In the above embodiments of the present application, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0089] In several embodiments provided by the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0090] The unit described as a separation component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0091] In addition, each functional unit in various embodiments of the present application may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0092] The above is only the preferred embodiment of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. An ocean vessel association and recognition method based on AIS and spaceborne SAR data, characterized in that include: Preprocess SAR data of different levels to obtain standardized SAR products in which each pixel carries latitude, longitude and elevation information with a preset accuracy not less than that of the preset level. Obtain the original AIS data table containing timestamp, longitude, and latitude information, and use the Vondrak algorithm to remove abnormal data points in the original AIS data table; The cleaned AIS data points are converted into AIS point vector files, and the discontinuous tracks of different ships are extracted. For each discontinuous track of a ship, the longitude and latitude coordinates at the time of SAR imaging are estimated based on the cubic spline interpolation method to form an AIS measurement set. Identify ship targets in SAR images and calculate the latitude and longitude coordinates of each ship target to form a SAR measurement set; The SAR measurement set and the AIS measurement set are paired to obtain a set of AIS traces and SAR traces within a preset association range, and a cost matrix of the SAR traces and AIS traces in the set is calculated. The cost matrix is optimally allocated based on the Munkres algorithm to obtain a trace association result.

2. The method according to claim 1, wherein The method of using the Vondrak algorithm to remove abnormal data points in the original AIS data table includes: Sort the raw data points in the AIS data table by timestamp to ensure time continuity; Use polynomial or spline functions to fit the original data points to obtain an initial smooth curve; Calculating the deviation between the original data point and the current smooth curve, and assigning the current smooth curve to the initial smooth curve during the first calculation; According to the deviation, the coefficients of the current smooth curve are optimized so as to minimize the energy function Q; The current smooth curve corresponding to the minimization of the energy function Q is used as the benchmark AIS trajectory sequence; The deviation between the original data point and the benchmark AIS trajectory sequence is calculated. If the deviation exceeds the set threshold, it is marked as an abnormal data point and deleted.

3. The method according to claim 1, wherein The cleaned AIS data points are converted into AIS point vector files to extract discontinuous tracks of different ships, including: Import the cleaned AIS data points into the geographic information system ArcGIS, and map the AIS data points onto a map based on the timestamp, longitude, and latitude information of each AIS data point to directly observe the spatiotemporal distribution of ship movements; The cleaned data is vectorized using ArcGIS to perform feature classification according to the relative position relationship and ship type attributes of the ships, so that the scattered points can be aggregated into independent tracks of different ships. The ship type attributes include at least: MMSI code, ship speed, and heading.

4. The method according to claim 3, characterized in that, For each vessel's discontinuous trajectory, the longitude and latitude coordinate information at the time of SAR imaging is estimated based on the cubic spline interpolation method to form an AIS measurement set, including: The longitude coordinate information and the latitude coordinate information at the time of SAR imaging are estimated based on the cubic spline interpolation method, wherein the method for estimating the longitude coordinate information and the latitude coordinate information is the same, including: Assume the longitude of the vessel is a function of time and satisfies a cubic spline function in the time interval [t1, t2]. Then the velocity component v y in the longitude direction of the vessel is the first derivative of, and the acceleration component a y in the longitude direction of the vessel is the second derivative of; According to the boundary conditions, at and the moment, the following should be satisfied: (2) Bring the boundary values , , , into formula (2), and the values of the coefficients a, b, c, and d of the cubic spline function can be obtained. When calculating the longitude value at any moment within the time period [t1, t2], substitute the corresponding moment t into formula (1).

5. The method according to claim 1, characterized in that, The identifying of the ship target in the SAR image includes: Collect a large number of SAR images with ship logos to train the Faster R-CNN model, so that the trained model can learn the characteristics of ships, including: shape, texture, and size; Input the SAR image to be detected into the trained model. The convolutional neural network in the model divides the entire SAR image to be detected into countless small blocks, analyzes the details of each small block, and determines which small blocks have suspected ship contours; The region proposal network in the model generates candidate boxes that may contain the target ship in the small blocks with suspected ship contours based on the learned ship features; The fully connected layer in the model determines whether there is really a ship in these candidate boxes. If so, it adjusts the position and size of the candidate boxes to make the candidate boxes become SAR bounding boxes that fit the actual contour of the ship better; The model outputs all the adjusted SAR bounding boxes and gives the coordinates of each SAR bounding box.

6. The method according to claim 5, wherein Calculating the longitude and latitude coordinate information of each ship target to form a SAR measurement set, including: For each identified ship target, define the center pixel point coordinate of its SAR bounding box as the center position of the ship target in the image space; Use the RPC model or geometric correction parameters of the SAR image to convert the center pixel point coordinates into geographic coordinates, and finally form a SAR measurement set.

7. The method according to claim 6, wherein Pairing the SAR measurement set and the AIS measurement set to obtain a set of AIS traces and SAR traces within a preset association range, including: The data in the SAR measurement set includes: the SAR longitude and latitude coordinate information of the detected ship target and the SAR bounding box of the ship target; among them, each SAR bounding box corresponds to a SAR longitude and latitude coordinate information, and a SAR longitude and latitude coordinate information represents a SAR trace; The data in the AIS measurement set includes: the movement trajectory of each ship, and the movement trajectory is represented by a string of timestamps and the AIS longitude and latitude coordinate information corresponding to the timestamps; among them, each AIS longitude and latitude coordinate information corresponds to an AIS trace; For each SAR bounding box, find all AIS trajectory points within a preset range near it to obtain a set of AIS traces and SAR traces within a preset association range.

8. The method according to claim 7, characterized in that, Calculating the cost matrix of the SAR traces and AIS traces in the set, and performing optimal assignment on the cost matrix based on the Munkres algorithm to obtain the trace association result, including: Assume that there are n SAR traces in the SAR measurement set and m AIS traces in the AIS measurement set. The cost matrix c is expressed as follows: Then the cost coefficient between the i-th SAR trace and the j-th AIS trace is , and its value is . In the formula represents the longitude and latitude of AIS, represents the longitude and latitude of SAR; Perform global minimum distance point track matching according to the cost matrix, and the minimum value , where is a binary function. If , it means that the i-th SAR point track is associated with the j-th AIS point track; if , it means that the i-th SAR point track is not associated with the j-th AIS point track; Judge whether the Euclidean distance of the longitudes and latitudes of the associated trace pairs is within the set judgment threshold. If it is within the judgment threshold, the trace association is successful, and the trace association result is saved; otherwise, the association fails, and the trace association result is deleted to complete the trace matching.

9. The method according to any one of claims 1 to 8, characterized in that, It also includes: Generating a SAR-AIS point position attribute information association table, including: the correspondence between AIS traces and SAR traces, and the attribute information of each pair of associated traces, to ensure the traceability and analyzability of the data; and / or Generating a SAR-AIS point position spatial distribution map, and drawing an association map of AIS traces and SAR traces through visualization means to show the correspondence between the ship targets in the SAR image and the AIS ship trajectories; and / or Generate a spatial association map of SAR-AIS points, display the spatial distribution of ship targets in the SAR image and the corresponding AIS traces, and highlight the correlation between the two through color annotation and connecting lines.

10. An electronic device, characterized in that, Including: A processor and a memory; The memory is used to store a program, and the processor is used to run the program to implement the method according to any one of claims 1-9.

Images