AIS Matcher:A Precise Matching Method of SAR Imagery and AIS Vessel Data for Maritime Object Identification

Abstract

This study proposes AIS Matcher, a precise matching methodology that integrates Synthetic Aperture Radar (SAR) imagery and Automatic Identification System (AIS) vessel data for enhanced maritime object identification.In conventional maritime surveillance workflows, SAR-based vessel detection and AIS-based positional information are typically processed independently, making cross-validation difficult and limiting the operational effectiveness of fused data utilization. Focusing on the complementary characteristics of the two sensors, this research presents an algorithm that matches vessel candidate regions (bounding boxes) extracted from SAR imagery with AIS trajectories through a multi-dimensional feature framework that includes spatial proximity (distance), temporal synchronization, geometric attributes(length and width), and heading direction. Unlike traditional approaches that rely solely on spatial proximity, the proposed method employs a weighted score-based matching mechanism that integrates distance, geometric similarity, and heading consistency, enabling stable candidate discrimination across diverse maritime environments.In real-world scenarios, multiple AIS signals may fall within the same SAR detection region, or some vessels may not transmit AIS at all,resulting in challenges such as multi-AIS ambiguity and missing matches. To address these issues, this study incorporates an intuitive verification interface that allows analysts to manually review uncertain cases, enhancing transparency in the matching process and enabling efficient post-processing even for large-scale datasets. Experiments conducted using commercial SAR imagery and real AIS logs collected across various sea regions demonstrate that the proposed method delivers consistent matching performance, with confirmed applicability even in high-density vessel traffic areas. The outcomes of this research offer significant potential for applications such as non-cooperative vessel tracking, illegal fishing monitoring, maritime safety management, and national security surveillance. By refining maritime situational awareness through SAR-AIS sensor fusion, AIS Matcher is expected to serve as a key component in next-generation intelligent maritime surveillance platforms.

1. Introduction

Illegal fishing, smuggling, maritime accidents, and deliberate AIS signal suppression pose significant threats to national security and the maritime economy (Zhao et al., 2014). In particular, optical satellite–based surveillance becomes highly constrained during nighttime, adverse weather, or in areas with poor communication coverage (Mourampetzis, 2025), highlighting the need for alternative sensors to ensure reliable monitoring capabilities. Synthetic Aperture Radar (SAR) provides high-resolution imagery regardless of illumination or weather conditions; however, SAR imagery alone cannot directly reveal a vessel’s identity, type, or intent (Corbane et al., 2010; Kanjir et al., 2018). Conversely, the Automatic Identification System (AIS) continuously broadcasts rich navigational and vessel metadata—such as Maritime Mobile Service Identity (MMSI), speed, heading, length, and width—yet the reliability of AIS data can be compromised by intentional shutdown, spoofing, or sensor errors (Galdelli et al., 2021).

Given these complementary strengths and weaknesses, integrating SAR and AIS has emerged as a key requirement for building more trustworthy maritime surveillance systems (Achiri et al., 2018). Traditional SAR–AIS matching approaches rely primarily on spatial proximity (Rodger and Guida, 2021), which often results in high false-match rates in complex environments such as ports or high-density shipping areas. Recent studies have introduced fusion techniques that incorporate multidimensional features—including vessel size, orientation, and temporal alignment—but most existing tools operate as stand-alone analytical modules, limiting analysts’ ability to visually verify and refine results in operational workflows (Chaturvedi et al., 2012; Graziano et al., 2019). To address these limitations, this study introduces AIS Matcher, a Quantum Geographic Information System (QGIS)-based integrated matching system designed to precisely associate SAR-detected vessel bounding boxes with AIS trajectories using a multi-feature matching strategy. Fig. 1 illustrates the overall system architecture of the proposed AIS Matcher.

Fig. 1. Overall workflow of the proposed AIS Matcher for maritime object identification.

The AIS Matcher precisely associates SAR-detected vessel bounding boxes with AIS trajectories by leveraging multidimensional features such as spatial distance, temporal alignment, vessel dimensions, and heading information. To overcome the limitations of traditional proximity-only matching approaches (Rodger and Guida, 2021), this study designs a weighted score-based algorithm that integrates distance, shape similarity, and orientation consistency. In addition, to address practical challenges frequently encountered in real maritime environments—such as multi-AIS ambiguity, missing AIS signals, and multiple SAR detections—the system provides an intuitive verification interface that enables analysts to visually inspect and refine candidate matches.

This paper provides a comprehensive description of the system architecture, data preprocessing pipeline, matching algorithm, user interface, and real-world case studies, and discusses the potential of the proposed approach for enhancing multi-sensor maritime surveillance frameworks.

2. Materials and Methods

2.1. System Overview and Data Preprocessing

In this study, the AIS Matcher was evaluated using a combined dataset of high-resolution SAR imagery and operational AIS logs. The SAR dataset consists of 97 scenes acquired from the Umbra commercial SAR constellation between January and December 2024, with ground resolutions ranging from 0.25 m to 1.0 m, enabling reliable detection of small and medium-sized vessels in both coastal and open-sea environments. Vessel bounding boxes were produced through a dedicated detection-and-labeling workflow developed in collaboration with SAR vessel-image analysis experts, who applied strict class-specific annotation guidelines to ensure the geometric accuracy of vessel extents. This expert-driven labeling process significantly reduces uncertainties typically associated with fully automated detection and provides high-fidelity training and evaluation data for the matching system. The AIS dataset was sourced from Seavantage, a commercial AIS provider aggregating terrestrial and satellite AIS streams.

The raw AIS data include MMSI identifiers, timestamps, geographic coordinates, speed over ground (SOG), course over ground (COG)/Heading, and vessel dimensions. To ensure temporal alignment with SAR acquisitions, a ±5-minute temporal window around each SAR scene timestamp was applied to construct AIS trajectory segments, compensating for AIS reporting delays or vessel motion between messages. After filtering corrupted or inconsistent messages and interpolating irregular sampling intervals to a one-minute resolution, the AIS data were stored as both point features and LineString trajectories for use in the multi-stage matching algorithm. This dataset configuration—bringing together high-resolution SAR imagery, expert-validated vessel bounding boxes, and temporally aligned AIS tracks—establishes a robust foundation for evaluating the AIS Matcher under realistic maritime surveillance conditions.

To compensate for temporal discrepancies between SAR acquisition and AIS reporting intervals, a ±5-minute temporal window was applied to construct interpolated AIS trajectories. The interpolation is performed at a 1-minute resolution, producing movement patterns that directly reflect the vessel’s actual reported positions: rapidly moving vessels form elongated linear segments, slow-moving vessels generate shorter segments, and anchored vessels yield near-static point clusters. This observation-driven temporal alignment minimizes positional mismatch at the SAR acquisition time and reduces the need for Kalman filter–based prediction, which relies on assumptions that can amplify uncertainty when AIS speed and heading values fluctuate.

The AIS Matcher is designed as a three-tier architecture composed of a data layer, an algorithm layer, and a user interface (UI) layer. The data layer loads AIS and SAR inputs and performs preprocessing following a standardized schema. The algorithm layer conducts precise matching using distance, vessel dimensions, heading, and speed information, and subsequently computes reliability indicators. Finally, the UI layer is implemented as a QGIS plugin, enabling users to visually inspect and validate the matching results. Such a hierarchical and modular design enhances maintainability and scalability in maritime surveillance systems. AIS data are provided in Comma-Separated Values (CSV) format and processed through the ais_preprocessor module, which validates key fields including MMSI, timestamp, geographic position (lat/lon), SOG, COG/heading, and vessel dimensions (length and width). Because AIS streams frequently contain observation noise, missing values, and erroneous vessel dimension information, rigorous preprocessingis essential for reliable analysis (He et al., 2021). Irregularly sampled AIS data are grouped by MMSI and linearly interpolated at one-minute intervals, after which the results are stored as LineString trajectories and Point features, distinguished by the feature_type attribute.

SAR bounding-box detections are provided in Shapefile format. In the shape_preprocess module, all geometries are transformed into the WGS84 coordinate reference system (EPSG:4326), and major/minor axis lengths are computed using pyproj.Geod. Because SAR imagery is subject to geometric distortion and speckle noise, vessel dimension estimation may contain errors (Moreira et al., 2013). Therefore, topological inconsistencies are corrected using the shapely.buffer(0)technique. During coordinate conversion, the average longitude of the study area is used to automatically infer the appropriate UTM zone, and region-specific inconsistencies are minimized by applying recommended EPSG codes (Moreira et al., 2013). As summarized in Table 1, all datasets are managed following a standardized directory structure.

Table 1. Standardized data directory structure

2.2. A Precision Matching Algorithm

The core of the AIS Matcher is a score-based matching algorithm that integrates distance, vessel dimensions, heading, and speed information. First, AIS points corresponding to the same acquisition time (at one-minute resolution) are filtered, and only those within a predefined distance threshold dist_m from the center of a SAR bounding box are considered as candidates. Distance-based candidate filtering is one of the primary criteria commonly used in SAR–AIS data fusion studies (Zhao et al., 2014).

The dimensional discrepancy between a vessel and a bounding box is computed as follows (Eq. 1):

\(\begin{align}\Delta L=\frac{\mid \text { Lbbox }- \text { LAIS } \mid}{\text { LAIS }}, \quad \Delta W=\frac{\mid \text { Wbbox }- \text { WAIS } \mid}{\text { WAIS }}\end{align}\)       (1)

The heading difference is calculated by comparing the orientation of the rotated minimum-area rectangle (minAreaRect) with the AIS COG or Heading. Because AIS headings can become unstable during low-speed or anchored situations, and SAR-based orientation estimation is also subject to various image characteristics and noise (Alexandre et al., 2024), the heading error Δθ is normalized within the range of 0–90 degrees.

The final matching score is defined as follows (Eq. 2):

Score = 10 x [wpos(dist/dist_m) + wlen |ΔL| + wwid |ΔW| + whead(Δθ/90)]       (2)

where dist is the geodesic distance between the AIS point and the bounding-box center; dist_m is the maximum allowed matching distance; wpos, wlen, wwid, and wheadare user-defined weighting factors controlling the influence of positional, length, width, and heading terms; ΔL and ΔW represent normalized discrepancies in vessel length and width; Δθ is the normalized heading difference between SAR-derived orientation and AIS heading/COG. where dist is the actual geodesic distance (in meters) between the bounding-box center and the AIS point, and dist_m is the distance threshold. The term dist/dist_m represents a normalized distance ratio in the range from 0 to 1. All weighting factors are user-defined, and lower scores indicate a higher likelihood of a correct match.

Step 1: Select AIS points within the ±5-minute temporal window around the SAR acquisition time.

Step 2: Compute geodesic distance from each AIS point to the bounding-box center.

Step 3: Remove candidates for which the distance exceeds dist_m.

Step 4: Compute ΔL, ΔW, and Δθ for remaining candidates.

Step 5: Apply the weighted scoring model to each candidate.

Step 6: Rank candidates by score, where lower scores indicate higher match likelihood.

A Reliability Index is additionally computed based on data quality and the score differences among candidate vessels. This reliability measure helps reduce ambiguity during the matching process by providing a clear separation between strong and weak candidates.

2.3. User Interface and Validation

As illustrated in Fig. 2, the UI of the AIS Matcher is implemented as a QGIS docking panel, allowing analysts to intuitively review matching candidates directly on the map display.

Fig. 2. AIS matcher user interface.

The main dialog (AIS Matcher Dialog) provides a step-by-step workflow that includes data loading, matching parameter configuration, candidate search, and result exporting. Several interactive features—such as SAR image transparency adjustment, automatic zooming upon candidate selection, candidate highlighting, MMSI-based candidate navigation, and explicit “No Match” assignment—were designed according to practical analyst requirements and are consistent with prior maritime surveillance UI studies (Graziano et al., 2019).

In the data-loading panel, users can select trajectory GeoJSON files, bounding-box Shapefiles, and SAR image files, with “zoom-to-layer” buttons enabling immediate navigation to corresponding map regions. SAR transparency can be adjusted using a slider. The matching parameter panel allows configuration of target time, maximum number of candidates per MMSI, maximum score thresholds, and the zoom ratio. Once the “Search Candidates” button is activated, the matching algorithm is executed based on the specified settings. The MMSI Candidate Selection panel displays the current MMSI under review, associated AIS information, and a ranked candidate list. Selecting any candidate automatically highlights the corresponding bounding box and AIS point on the map, while detailed attribute information is displayed in the interface. Users can move sequentially through vessels using the Previous/Next buttons or explicitly mark a vessel as having “No Matching AIS.”

The AIS Matcher Info Dock displays bounding-box information, AIS attributes, and comparative metrics in separate panels. Bounding-box attributes include ID, width, length, area, and orientation; AIS attributes include MMSI, vessel name, vessel type, width, length, heading, and speed. The comparison panel shows dimensional similarity, heading similarity, and the composite score. The AIS Layer Manager loads and manages trajectory layers, AIS point layers, bounding-box layers, and SAR imagery. Each layer is automatically styled: trajectories are color-coded by MMSI, AIS points are rendered as blue circular markers, and bounding boxes are shown with blue outlines and semi-transparent fills.

When a candidate is selected, the corresponding bounding box is highlighted in red with a label, and the associated AIS point is emphasized in red. The automatic zoom function navigates the map to the selected candidate according to the user-defined scale. Users may assign a matching result for each MMSI or designate it as “No Match.” The output is written to matched.csv and unmatched.csv, and vessels detected in SAR imagery but not associated with AIS (“dark ships”) are automatically exported as a GeoJSON layer for reporting. An example of a successfully matched vessel is provided in Fig. 3, and representative AIS parameters stored in the matched output file are listed in Table 2.

Fig. 3. Example of a successfully matched vessel using the AIS Matcher.

Table 2. Representative attribute fields contained in the AIS data

2.4. Secondary Matching Algorithm

The primary matching algorithm considers multiple attributes—distance, length, width, heading, and speed—to achieve high-precision matching. However, in complex maritime environments, errors in any single attribute may cause a true match to be incorrectly discarded. SAR-derived size and orientation estimates can be affected by speckle noise, resolution limitations, or interpretation ambiguity (Alexandre et al., 2024), while AIS-reported vessel length/width and heading values often become unreliable when a vessel is anchored or moving slowly. As aresult, these single-attribute errors can lead to cases where valid candidates fail to appear in the initial candidate list, reducing recall. Because single-attribute errors—such as unstable AIS headings or SAR orientation noise—may eliminate true matches during the primary stage, a secondary distance-based stage is necessary to recover candidates erroneously excluded by the strict multidimensional criteria.

To address this inherent limitation, the AIS Matcher incorporates a secondary matching algorithm designed to recover potential candidates that were missed during the primary stage. The goal of this secondary matching process is to compensate for the structural strictness of the first stage, thereby improving the overall matching success rate and reducing unnecessary “dark ship” classifications. The primary matching adopts a forward AIS-centric strategy in which AIS points are matched to bounding boxes using all multidimensional attributes. While this approach achieves high precision, its strict criteria can inadvertently exclude true matches—particularly when SAR bounding boxes exhibit geometric distortion or when AIS metadata contains erroneous vessel dimensions or unstable headings. This necessitates a supplementary mechanism.

The secondary matching performs reverse matching, taking the bounding boxes as the starting point to identify unmatched AIS targets. All AIS records not assigned during the first stage, along with all bounding boxes, are reloaded and normalized to the WGS84 coordinate system. The average longitude of bounding box centers is used to automatically estimate the appropriate UTM zone. After converting bounding boxes into UTM coordinates, an R-tree spatial index is constructed to enable efficient neighborhood queries. For each AIS point, bounding boxes within a predefined radius are searched, and Euclidean distances are computed in UTM coordinates. Only candidates within the threshold are retained, and their ordering is determined by ascending distance.

Furthermore, because AIS trajectories are constructed using a ±5-minute temporal window centered on the SAR acquisition time, the matching process is minimally affected by AIS timestamp drift. The interpolation process renders characteristic movement patterns for each vessel—elongated paths for fast movers, compact paths for slow movers, and nearly static clusters for anchored vessels—allowing reliable candidate recovery without explicit motion prediction. Within this limited temporal window, potential displacement remains small relative to the SAR scene extent, ensuring that the distance-only secondary matching remains effective.

In this secondary stage, distance is intentionally used as the sole matching criterion. Applying the same multidimensional filters used in the first stage would cause the exact same exclusions—again filtering out valid but imperfect candidates. Additionally, distance is the most stable attribute in SAR–AIS fusion: it is less affected by SAR speckle, orientation ambiguity, or AIS metadata inaccuracies compared to length, width, or heading. Therefore, using distance alone enables effective recovery of missed true matches.

As shown in Fig. 4, the secondary matching algorithm successfully restores valid matches that were omitted by the primary algorithm. This complementary design significantly improves overall recall and reduces false dark ship classifications.

Fig. 4. Primary and secondary matching results for recovering missed valid ship–AIS matches. (a) Example of valid match candidates missed in primary matching. (b) Restored matching results using the secondary matching algorithm.

The secondary matching outputs a dictionary containing candidate bounding boxes for each AIS MMSI, where each candidate is accompanied by its corresponding distance value. Analysts can review these candidates through the dedicated Second Match Dialog, an interactive interface designed specifically for validating secondary matches. Through this interface, users can visually inspect the spatial relationship between AIS points and bounding boxes, assess the plausibility of each candidate, and make the final decision on whether a match should be confirmed.

By integrating this supplementary matching stage with user-assisted verification, the system effectively recovers vessels that were missed during the primary matching phase. Consequently, the overall matching success rate and reliability are significantly improved, while the number of bounding boxes incorrectly classified as dark ships is substantially reduced.

3. Results

The proposed AIS Matcher provides an integrated framework for the precise association of SAR imagery with AIS vessel data, supporting the entire workflow—from preprocessing to final result generation—within a unified QGIS environment. The score-based candidate search method, which combines distance, vessel dimensions, and heading information, substantially reduces the burden of manual verification frequently required in maritime surveillance tasks while enhancing the consistency and reliability of the matching process. The framework also demonstrated strong robustness against variability in the two heterogeneous sensor inputs. Even when AIS messages were collected at irregular intervals or when SAR-derived bounding boxes exhibited geometric uncertainty, the system maintained stable performance through systematic preprocessing and standardized feature extraction procedures.

The interactive verification interface further enabled analysts to visually compare and validate candidates, effectively resolving ambiguous cases that may arise during automated processing and lowering the likelihood of matching errors. The modular system architecture increases adaptability to diverse maritime environments and operational scenarios. Each stage—including data loading, preprocessing, feature computation, candidate scoring, user validation, and result export—is designed as an independent module, allowing flexible configuration based on regional characteristics, sensor combinations, or image resolution. Notably, in high-traffic maritime areas where simple distance-based matching is prone to error, the proposed multi-feature method improved the ability to distinguish vessels located in close proximity.

Some limitations arise from variations in input data quality. Inaccurate AIS vessel specifications or degraded SAR imagery may introduce errors in bounding-box extraction or candidate ranking. However, these limitations can be largely mitigated through user-assisted verification and do not fundamentally undermine the overall stability of the system. Overall, AIS Matcher proves to be a practical tool for precisely correlating SAR-based detections with AIS trajectories, showing strong applicability across operational maritime surveillance tasks.

Moreover, the multi-sensor fusion strategy proposed in this study has the potential to evolve into a general-purpose framework applicable to a wide range of domains, including port monitoring, environmental incident tracking, and disaster response. The semi-automated verification stage is deliberately included due to operational realities in East Asian maritime regions, where illegal fleets frequently reuse a single MMSI across multiple vessels, deactivate AIS, or engage in covert ship-to-ship transfers. Such behaviors cannot be reliably resolved through fully automated algorithms alone. Incorporating analyst-in-the-loop validation enables correct differentiation of vessels sharing identifiers and ensures accurate attribution in high-risk scenarios.

4. Conclusions

The development and application of AIS Matcher have revealed several significant advantages and insights for maritime surveillance. The score-based matching approach, which integrates multiple feature dimensions, offers clear benefits over simple distance-based matching by combining distance, vessel dimensions, and heading with adjustable weighting. This enables users to tune the relative importance of each feature, providing adaptability across diverse maritime environments and vessel types. The method demonstrated stable performance even in high-density traffic regions, confirming its practical utility in real operational settings.

The two-stage matching strategy—comprising a precise first-stage match followed by a complementary second-stage recovery step—proved particularly effective in maximizing the overall matching success rate. By applying strict criteria in the first stage and later recovering missed candidates using a distance-only approach, the system achieves both high accuracy and broad coverage. This staged methodology highlights an important design insight: that a stepwise approach can be more effective than attempting to address all complexities within a single algorithmic step.

The user-centered interactive verification environment was found to be essential in mitigating the limitations of fully automated matching systems. By incorporating analysts’ expertise directly into the final decision-making process, the system reduces the likelihood of errors and enhances overall reliability. The full integration of all functions within the QGIS environment further lowers the adoption barrier, allowing analysts to employ the tool without significantly altering existing workflows.

Additionally, the modular and loosely coupled architecture ensures long-term maintainability and extensibility. Because preprocessing, matching, and interface components operate independently, new matching strategies or deep learning–based modules can be introduced without requiring major structural changes. Beyond maritime surveillance, AIS Matcher contributes academically by presenting a general methodology for designing multi-sensor fusion systems. The demonstrated effectiveness of SAR–AIS fusion suggests that similar strategies can be extended to other sensor combinations, such as optical imagery with AIS or radar with AIS. These findings have the potential to support advancements in maritime monitoring systems, enhance situational awareness, strengthen illegal fishing surveillance, improve maritime safety management, and contribute to national security operations.

Author Contributions

Conceptualization, Data curation, Formal analysis: Lee JH, Park YJ; Investigation: Lee JH, Park YJ, Kim JW; Methodology: Lee JH, Park YJ, Lee SC; Project administration: All authors; Resources: Lee JH, Kim JW, Lee SC; Software: Lee JH, Park YJ; Supervision: All authors; Validation: Lee JH, Lee SC; Visualization: Lee JH, Park YJ, Lee SC; Writing–original draft: Lee JH; Writing–review & editing: Lee JH, Kim KS, Lee SC.

Conflicts of Interest

No potential conflict of interest relevant to this article was reported.

Funding

This work was supported by the Ministry of SMEs and Startups (MSS) of the Republic of Korea under the Startup Growth Technology Development Program (DeepTech-TIPS)funded by the Korea Technology and Information Promotion Agency for SMEs (TIPA) (No. RS-2023-00272709), and by the Ministry of Trade, Industry and Energy (MOTIE) under the Materials and Components Technology Development Program (Convergence of Heterogeneous Technologies)funded by the Korea Evaluation Institute of Industrial Technology (KEIT)(No.RS-2024-00419532).

Data Availability Statement

The datasets used in this study were obtained under commercial licenses and internal company agreements and therefore cannot be shared publicly. Access requests may be considered by the corresponding author on a case-by-case basis.