Matching Method for Ship Identification Using Satellite-Based Radio Frequency Sensing Data

Abstract

Vessels can operate with their Automatic Identification System (AIS) turned off, prompting the development of strategies to identify them. Among these, utilizing satellites to collect radio frequency (RF) data in the absence of AIS has emerged as the most effective and practical approach. The purpose of this study is to develop a matching algorithm for RF with AIS data and find the RF's applicability to classify a suspected ship. Thus, a matching procedure utilizing three RF datasets and AIS data was employed to identify ships in the Yellow Sea and the Korea Strait. The matching procedure was conducted based on the proximity to AIS points, ensuring accuracy through various distance-based sections, including 2 km, 3 km, and 6 km from the AIS-based estimated points. Within the RF coverage, the matching results from the first RF dataset and AIS data identified a total of 798 ships, with an overall matching rate of 78%. In the cases of the second and third RF datasets, 803 and 825 ships were matched, resulting in an overall matching rate of 84.3% and 74.5%, respectively. The observed results were partially influenced by differences in RF and AIS coverage. Within the overlapped region of RF and AIS data, the matching rate ranged from 80.2% to 98.7%, with an average of 89.3%, with no duplicate matches to the same ship.

1. Introduction

Transportation of goods via sea is now considered one of the crucial pillars that helps to maintain today’s highly connected global economy. Furthermore, it is expected that transportation volumes will double by 2050, with more than 70% of goods transported via sea routes (Wolsing et al., 2022; International Transport Forum, 2021). Along with the expansion of maritime trades and transportation, marine surveillance has become one of the most significant tasks that provide emphasis on preventing maritime accidents and securing safe sailing (Wolsing et al., 2022).

To ensure navigation safety, the International Maritime Organization (IMO) developed the Automatic Identification System (AIS) in the 1990s as a complementary system to high-frequency radar technology (International Maritime Organization, 1998). AIS is one of the most utilized vessel tracking systems which aids in onboard broadcasting and transmits the ship’s information such as identity, position, course, destination, etc. in real time to the on-shore AIS receiving station situated at land by using a Very High Frequency (VHF) radio wave (Hong et al., 2018). This positional data aids neighboring ships in avoiding collisions, while onshore Vessel Traffic Services (VTSs) utilize AIS for traffic management and guidance. Typically, land-based AIS systems have a coverage range of around 20 nautical miles. However, this range can vary significantly depending on factors such as VHF propagation conditions and sensor height (Vespe et al., 2008; Yang et al., 2012), and thus the observation area has become limited to harbor or coastal region.

To broaden sea lane monitoring coverage, specialized low Earth orbit satellites are launched. These satellites can receive AIS messages, denoted as Satellite-based AIS (S-AIS), across a wider coastal area using terrestrial standard AIS receivers, offering a solution to this limitation (Hong et al., 2018). According to Høye et al. (2008), a satellite positioned at an altitude of 1,000 km can process messages from as many as 900 vessels, achieving a detection probability of over 99% within a 6-second reporting interval during its pass. In case of monitoring, the most significant benefit of using S-AIS is that information on a vessel engaged in international navigation can be explored at ease (Hong et al., 2018). Thus S-AIS data enables global monitoring and extends tracking capabilities by addressing the gaps of terrestrial AIS in ocean observations (Cervera and Ginesi, 2008; Metcalfe et al., 2018).

Furthermore, satellite data, particularly synthetic aperture radar (SAR) data, holds significant importance in maritime traffic monitoring. SAR satellites provide distinct advantages for ocean surveillance, such as day-night imaging capabilities and the ability to penetrate cloud cover. These attributes make SAR data highly valuable for vessel detection and monitoring studies. To enhance SAR-based ship detection and monitoring, advanced techniques like windowing-based target detection methods and artificial intelligence (AI) technologies such as deep learning (DL), convolutional neural networks (CNNs), etc. are widely employed (Shin et al., 2024). Several experimental studies have utilized SAR data for ship detection, employing algorithms like Region-based CNN (RCNN), Single Shot Multi Box Detector (SSD), and You Only Look Once (YOLO), which are prevalent in object detection tasks (Nie et al., 2017; Zhao et al., 2018; Chang et al., 2019).

However, SAR data presents challenges including discontinuous monitoring and speckle noise, making it difficult to differentiate ships with low backscatter values from speckle noise of similar intensity that leads to misdetection (Jeong and Yang, 2016; Shin et al., 2024). Besides, vessels can operate with their AIS switched off (dark vessel) against IMO regulations, and therefore, several strategies have been employed to identify them among which, using satellites to collect radio frequency (RF) data in the absence of AIS is the most effective and useful approach (Dark Shipping Solutions, 2024).

RF data encompasses the observation of the electromagnetic radiation spectrum. When a radio wave oscillates within this spectrum, it emits a frequency that can be gauged within the range of 300 GHz to 9 kHz (Travis Turgeon, 2023). While AIS may be subject to manipulation, RF strategy offers a more reliable means of observing ship operations at sea, irrespective of weather conditions, time of day, and other factors. Furthermore, RF data provides geolocation accuracy of approximately 2 km (SatMagazine, 2023). Several studies for identifying and monitoring dark ships in the absence of valid AIS data were conducted by some of the prominent companies offering space-based RF services. One of the case studies was performed by Unseenlabs where a fishing fleet was monitored in the north Arabian Sea by utilizing the RF data collection (Dark Shipping Solutions, 2024).

The data collection process started with the identification of all RF signals in the northern Arabian Sea. After accumulating the data, analysis was conducted to determine how many of the ships emitting RF signals were also transmitting AIS data. Following the analysis, unseenlabs discovered that approximately 65% of the vessels identified through RF data were also broadcasting AIS data. This implies that roughly 35% of the vessels in the area were operating without AIS visibility (Dark Shipping Solutions, 2024). Thus, it depicted the significance of RF data for marine authorities to visualize the overall and correct circumstances of all the vessels operating in the sea.

The purpose of this study is to propose a matching methodology for RF data and AIS information and evaluate the feasibility of RF data for ship identification. Therefore, this study used the RF data to obtain the ship information which was then matched with the S-AIS information for identifying the vessels operating in the Yellow Sea and the western part of the Korean Strait. The findings of this study illustrate the matching rate between RF and S-AIS data, offering valuable insights for future ship detection studies utilizing multiple datasets.

2. Materials and Methods

2.1. Study Area and Data

The study area extended from 120°E to 128°E and 30°N to 36°N (Fig. 1), covering the Yellow Sea and west part of the Korea Strait region which serves as an economic link between Korea and other Asian countries, and plays a central role in the logistics network of the Asia-Pacific region. The study region, adjacent to major international maritime routes and featuring connections to significant ports such as Dalian, Qingdao, Shanghai, Hai Phong, and Port Klang, serves as a thoroughfare for numerous vessels navigating between Korea and other Asian destinations. Additionally, commercial fishing activity is also conducted in that region due to the availability of aquatic species such as blue crabs, croakers, anchovies, etc. which have high marketable values.

Fig. 1. Research area in the Yellow Sea with coverages of radio frequency (RF) data acquired on 25 September 2023 where red, green, and blue boxes represent the RF1, RF2, and RF3 datasets, respectively. RF1 consists of the ship information collected at 03:24:26 UTC within frequency 3036.7–3076.1 MHz and at 03:24:28 UTC within frequency 9375.7–9423.3 MHz. RF2 contains the ship information collected at 12:18:56 UTC within frequency 3024.0–3076.5 MHz and at 12:18:58 UTC within frequency 9378.4–9424.8 MHz. RF3 comprises the ship information collected at 13:54:42 UTC within frequency 9374.2–9420.9 MHz and at 13:54:44 UTC within frequency 3033.5–3077.0 MHz. UTC: Coordinated Universal Time.

Unseenlabs RF satellite data were collected on 25 September 2023 (Table 1). RF satellites are equipped with specialized receiving antennas and can detect signals over large areas. After analyzing signals, these satellites precisely geolocate all emitting sources in space-time, assigning each vessel at sea a unique signature for real-time monitoring of its location (SatMagazine, 2023). With seven satellites in its constellation, Unseenlabs achieves a revisit time of about six hours for a designated area, maintaining a geolocation accuracy of nearly one nautical mile (2 km) (SatMagazine, 2023). In this study, the RF data spatially covered an area that was situated between 30°N to 40°N and 120°E to 130°E(Fig. 2). RF data contained 12 pieces of information including ID, frequency, latitude, longitude, reliability, and accuracy which were obtained by using the frequency ranged from 3024.0 to 9424.8 MHz. The RF dataset consists of two time periods, which are the standard time ±1 second with different frequencies, and the reliability ranges from 1 to 100.

Table 1. The radiometric frequency (RF) data used in this study

Fig. 2. Position of ships within the study area obtained from the radio frequency (RF) datasets on 25 September 2023. (a) RF1: Ship information was collected at 03:24:26 UTC within frequency 3036.7–3076.1 MHz (red dots), and at 03:24:28 UTC within frequency 9375.7–9423.3 MHz (green dots). (b) RF2: Ship information was collected at 12:18:56 UTC within frequency 3024.0–3076.5 MHz (red dots), and at 12:18:58 UTC within frequency 9378.4–9424.8 MHz (green dots). (c) RF3: Ship information was collected at 13:54:42 UTC within frequency 9374.2–9420.9 MHz (red dots), and at 13:54:44 UTC within frequency 3033.5–3077.0 MHz (green dots).

In this research, S-AIS Data (https://spire.com/maritime/) collected on 25 September 2023 (3 hours before and after the RF dataset time) for the Yellow Sea were used. The S-AIS data spatially ranged between 30° to 42° latitude and 120° to 136° longitude that covers the southern waters of Busan and Ieodo Ocean Research Station. It contained a total of 32 pieces of information, including maritime mobile service identity (MMSI), ship type, latitude, longitude, speed, direction, cog (route), nationality, origin, and destination. In this study, MMSI, ship name, and ship type were grouped and treated as one ID to classify ships.

2.2. Methodology

Fig. 3 depicts the schematic representation of the ship identification process by matching the RF and S-AIS data. Before identifying ships using RF and S-AIS data, a preprocessing step was conducted within S-AIS data. Vessels were filtered out if there was no available data within a timeframe of 3 hours before and after the RF datasets’ time range. Additionally, S-AIS data unrelated to the study area was omitted. Afterwards, to synchronize with the target time of the RF dataset, the S-AIS data was subjected to interpolation at 30-minute intervals. Linear interpolation techniques were applied to estimate the vessel positions at the specific target times, and the Dead Reckoning (DR) locations were also determined.

Fig. 3. Overall flowchart of matching the radio frequency (RF) and Satellite-AIS (S-AIS) data for ship identification. DR: Dead Reckoning, S-AIS: Satellite-based Automatic Identification System.

Subsequently, one-to-one matching was performed between the RF dataset and the interpolated DR points. This matching process involved assessing the distance between each RF data point and its corresponding interpolated DR point. Following the initial matching step, the overall interpolated trajectory of the S-AIS data was plotted, providing a comprehensive view of the vessel movements. Vessel matching was conducted based on the proximity between the RF data points and the interpolated DR points along the trajectory. To ensure accuracy, matching was carried out in multiple sections, including distances of 2 km, 3 km, and 6 km from the DR points. Furthermore, to prevent repeated matches, the vessel IDs matched in each section were excluded from subsequent matching attempts. Lastly, the matching results at each frequency range of each dataset were merged.

Fig. 4 illustrates an example of the matching process between RF data and interpolated DR points to identify the vessel with the MMSI number 538006544, named TRF MONGSTAD, it is classified as a tanker designed for transporting chemicals, indicated by the vessel type TANKER_CHEMICALS. The vessel has a length of 184 meters and a width of 27 meters. It is registered under the flag of the Marshall Islands, denoted by the flag_country. This vessel emits signals in two frequency bands: 3041.7 MHz at 03:24:26 UTC and 9411.3 MHz at 03:24:28 UTC, facilitating its identification and tracking through RF data analysis.

Fig. 4. Example of matching procedure for MMSI 538006544 by using radio frequency (RF) and S-AIS data at 30-minute intervals on 25 September 2023, within the 6 km distance category. (a) Matching of the ship at 03:24:26 UTC. (b) Matching at 03:24:28 UTC. (c) Combined matching at 03:24:26 and 03:24:28 UTC. The red dot indicates the initial position of the ship from the interpolated S-AIS. MMSI: Maritime Mobile Service Identity, S-AIS: Satellite-based Automatic Identification System.

3. Results

3.1. Matching Result of First Radio Frequency Dataset (RF1)

In the RF1 dataset, the total number of ship information was found 1,023 among which 460 ship information was obtained at 03:24:26 UTC, and 563 ship information was obtained at 03:24:28 UTC. On the other hand, in the case of S-AIS data, a total of 86,419 ships were included in the matching process. At 03:24:27, during the data matching phase, 48,672 ships passed through the initial filtering based on time criteria, while 28,108 ships passed through regional range filtering.

The matching results from RF1 and S-AIS data showed a total of 798 identified ships (Fig. 5). At 03:24:26 UTC, 352 ships were identified out of 460 ships, and at 03:24:28 UTC, 446 ships out of 563 were detected. The matching results from the RF1 dataset and interpolated S-AIS data revealed a matching rate of 76.5% at 03:24:26 UTC and 79.2% at 03:24:28 UTC. The overall matching rate for the RF1 dataset was determined to be 78%. Additionally, the matching rate within the 2 km and 3 km distance was 55% and 19%, respectively.

Fig. 5. Ship identification result after matching the first radio frequency dataset (RF1) and Satellite-AIS (S-AIS) data at 30-minute intervals on 25 September 2023. The blue line represents the interpolated ship trajectories from S-AIS. S-AIS: Satellite-based Automatic Identification System.

Fig. 6 illustrates the matching outcomes within an enlarged area in the Yellow Sea, displaying the trajectories of four fishing vessels. They are presented in sequential order (from top) with their respectiveand MMSI, and ship names: MMSI 412343786, LUJIAOYU60538; MMSI 412323001, LURONGYU55088; MMSI 412348806, LUJIAOYU60177; and MMSI 12431206, ZHERUIYU 01355. The matching results indicated that three ships, with MMSI numbers 412323001, 412348806, and 12431206, were successfully identified by using the RF data and interpolated DR points. Additionally, the ship with MMSI 412323001 was matched by both RF data obtained at 03:24:26 UTC and 03:24:28 UTC, respectively.

Fig. 6. Matching result within a small area from the first radio frequency dataset (RF1) and interpolated S-AIS on 25 September 2023. The amber, green, red, and blue lines represent the interpolated trajectories of the ship at 30-minute intervals with the MMSI numbers 412343786, 412323001, 412348806, and 12431206, respectively. S-AIS: Satellite-based Automatic Identification System.

3.2. Matching Result of Second Radio Frequency Dataset (RF2)

In the RF2 dataset, a total of 953 ship’s data was found, and among them, 394 and 409 ship’s information were obtained at 12:18:56 UTC and 12:18:58 UTC, respectively. For S-AIS data at 12:18:57 UTC, a total of 25,945 ship’s information was utilized for matching which were selected from 50,194 ships after conducting the time and region filtering, respectively.

The matching process revealed the identification of a total of 803 ships (Fig. 7). At 12:18:56 UTC, 394 ships were identified from a dataset of 487 ships, and at 12:18:58 UTC, 409 ships were detected out of 466. The highest matching rate was found in RF2, with 80.9% at 12:18:56 UTC and 87.8% at 12:18:58 UTC, respectively. The RF2 dataset exhibited an overall matching rate of 84.3%. Moreover, matching rates of 51% and 25% were observed within distances of 2 km and 3 km, respectively.

Fig.7. Ship identification result after matching the second radio frequency dataset (RF2) and Satellite-AIS (S-AIS) data at 30-minute intervals on 25 September 2023. The blue line represents the interpolated ship trajectories from S-AIS. S-AIS: Satellite-based Automatic Identification System.

3.3. Matching Result of Radio Frequency Dataset 3 (RF3)

In the RF3 dataset, a total of 1,107 ship data records were found. Specifically, 396 ship’s records were acquired at 13:54:42 UTC, while 711 ship’s records were obtained at 13:54:44 UTC. On the other hand, from S-AIS data at 13:54:43 UTC, a total number of 25,726 ships out of 47,136 were selected for matching after accomplishing the time and region filtering, respectively.

The matching result depicted that a total of 825 ships were identified (Fig. 8). Besides, it was found that 317 ships were identified from a dataset of 396 ships at 13:54:42 UTC while 508 ships out of 711 ship data were detected at 13:54:43 UTC. The matching of RF3 and S-AIS data showed matching rates of 80.1% at 13:54:42 UTC and 71.4% at 13:54:44 UTC, respectively. The total matching rate for the RF3 dataset was found 74.5%. Additionally, within the 2 km distance, the matching result was 44%, whereas within the 3 km distance, it was 23%. After analyzing the overall result from each RF dataset across different frequencies, it was found that the average matching rate for RF and S-AIS data on September 25, 2023, was calculated to be 79.3%, with no duplicated RF matching results observed. However, there is a difference in the coverage between RF and AIS data.

Fig. 8. Ship identification result after matching the third radio frequency dataset (RF3) and Satellite-AIS (S-AIS) data at 30-minute intervals on 25 September 2023. The blue line represents the interpolated ship trajectories from S-AIS. S-AIS: Satellite-based Automatic Identification System.

4. Discussion

To enhance navigation safety, AIS has been mandated for new vessels since 2002 under the Safety of Life at Sea amendments by the IMO. Despite being subject to line-of-sight limitations, AIS signals can propagate over hundreds of kilometers, though they are vulnerable to radio interference and attenuation. Utilizing satellite-mounted receivers, S-AIS expands global tracking capabilities, addressing the limitations of terrestrial AIS in oceanic observations. However, vessels intent on operating undetected can simply switch off their AIS transponders, halting signal transmission. To address this issue, RF data can play a crucial role in verifying a ship’s AIS data.

In this study, we conducted a matching process using both RF and S-AIS data to identify vessels operating in the Yellow Sea and the Korea Strait. From the analysis, it was found that the average matching rate for the overall RF and S-AIS data on 25 September 2023 was 79.3%. The lower matching rate was attributed to the absence of S-AIS data within the RF coverage region. Therefore, S-AIS coverage was taken into consideration to assess the matching rate within the overlapped region between RF and S-AIS data, spanning from 32°N to 34.5°N and 123°E to 127.5°E. Within this region, the matching results from the RF1 dataset and S-AIS data displayed a matching rate of 80.2% at 03:24:26 UTC and 87.3% at 03:24:28 UTC. The highest matching rate was exhibited by the RF2 and S-AIS data, with 98.7% at 12:18:56 UTC and 92.7% at 12:18:58 UTC, respectively. RF3 and S-AIS data depicted matching rates of 88.5% at 13:54:42 UTC and 88.4% at 13:54:44 UTC. Finally, a considerably higher average matching rate of 89.3% was obtained for the entire RF and S-AIS data within the overlapping region.

Furthermore, the matching scenario of vessels from the RF1, RF2, and RF3 datasets within an enlarged area was illustrated to comprehend the matching rate. Fig. 6 depicts the matching results of four fishing vessels from RF1 within a specified area, with three successfully matched; however, the fishing vessel with MMSI 412343786 was not matched. This discrepancy could be due to the distance-based matching method, leading to the vessel remaining unmatched. Future efforts will focus on resolving this issue. On the other hand, within the RF2 and S-AIS data (Fig. 9a), the presence of two fishing vessels, denoted by MMSI 440154850 (red line) and MMSI 440148940 (green line), was identified, and successful matching was achieved.

Fig. 9(b) depicts an enlarged view of the RF3 dataset, illustrating the interpolated trajectories of four vessels. Among these, two were cargo vessels, represented by MMSI 312040000 (red line) and MMSI 441571000 (green line), while the remaining two were fishing vessels, identified by MMSI 440125620 (amberline) and MMSI 440100880 (blue line). From the RF3 and S-AIS data, two cargo vessels and one fishing vessel were successfully matched, while the fishing vessel with MMSI 440125620 remained unmatched due to the absence of RF data for this specific vessel.

Fig. 9. Matching scenario within a small area from radio frequency (RF) dataset and interpolated S-AIS data on 25 September 2023. (a) Matching results from the second radio frequency dataset (RF2) and interpolated S-AIS. (b) Matching result from third radio frequency dataset (RF3) and interpolated S-AIS. The red, green, amber, and blue lines represent the interpolated trajectories of the ship at 30-minute intervals. S-AIS: Satellite-based Automatic Identification System.

5. Conclusions

AIS data are globally used to prevent maritime accidents and ensure safe sailing. However, vessel information may become unavailable if AIS transponders are intentionally or unintentionally deactivated. In such cases, RF data becomes significant for acquiring ship information and verifying the ship’s AIS position. This study presents a matching process utilizing RF and S-AIS data for ship identification. The matching results revealed that out of 3,083 ships within the RF coverage, 2,426 ships were successfully matched, resulting in an average matching rate of 79.3% for the entire RF dataset. Moreover, S-AIS coverage was considered to determine the matching rate in the overlapping region between RF and S-AIS data, resulting in an average matching rate of 89.3%. Above all, any duplicate matched for the same ship was not found within the RF matching result.