Improvement of Fishing Vessel Detection Performance Using SAR and Trajectory Data

다국어 초록 (Multilingual Abstract)

Maritime Domain Awareness (MDA) is essential to ensure maritime security based on multi-platform data. Synthetic Aperture Radar (SAR) generated weather-independent image data covering extensive marine regions, and SAR was widely utilized for ocean surveillance. Automatic Identification System (AIS) provides ship type and the current location from real- time static and dynamic information, which becomes helpful in avoiding ship accidents by analyzing the behavior of ship trajectory. Both data were considered efficient tools for ship monitoring as a part of MDA, whereas monitoring of small fishing vessels has significant difficulties due to their technical limitation, such as image resolution and failure of equipment. Therefore, this thesis suggests an improvement method for small fishing vessel detection and identification performance from each independent data of SAR and trajectory data with AIS/V- Pass (Small fishing vessel tracking system). For improving the detection performance of small fishing vessels, a polarimetric combination method was proposed using Sentinel-1 Single Look Complex (SLC) images. Initially, a combination of VH and VV polarization channels was exploited to minimize azimuth smearing and ambiguities occurring in SAR for detecting merchant ships (fusVH). Second, reflection symmetry (sym) was utilized to create an image for detecting small fishing vessels that satisfies condition: when the difference between sym and fusVH was less than 7.2 dB and sym was greater than -18.03 dB, the pixel values in fusVH was replaced to sym (fusSym). Then, average filtering was applied to fusSym using each component of complex conjugate for sym. For ship detection, adaptive thresholds were applied to each image by categorized Radar Cross Section (RCS). Final small fishing vessel position was confirmed after filtering merchant ship obtained from fusVH. The results were evaluated using 14 scenes from 2021 to 2023 and the average matching results were calculated to 0.85 for matching rate, 0.24 for false alarm rate and 0.86 score of Area Under the receiver operating characteristic Curve (AUC), which was achieved the highest performance compare with other detectors. The detection of 10m length fishing vessel was demonstrated through field experiment. Using ship trajectory data, an enhanced ship-type classification model that employs a sequential processing methodology integrating Hidden Markov Model (HMM), Deep Neural Network (DNN), and Convolutional Neural Network (CNN) techniques was developed. Four different ship types—fishing boat, passenger, container, and other ship—were classified using multiple ship trajectory features extracted from the AIS and small fishing vessel tracking system. For model optimization, both ship datasets were transformed into various formats corresponding to multiple models, incorporating data enhancement and augmentation approaches. Speed over ground, course over ground, rate of turn, rate of turn in speed, berth distance, latitude/longitude, and heading were used as input parameters. The HMM–DNN–CNN combination was obtained as the optimal model and achieving classification performances of 97.54%. Application for fishing vessel tracking based on multi-satellite sensors and trajectory data was performed by the proposed small fishing vessel detection and identification method, and this result is expected to contribute to effective ship monitoring.

목차 (Table of Contents)

• 1. Introduction １
• 1.1 Background １
• 1.2 Related Works and Limitations ２
• 1.3 Research Purpose ６
• 2. Improvement of Small Fishing Vessel Detection from Sentinel-1 SAR ８
• 2.1 Materials and Methodology ８
• 2.2 Results and Discussion １９
• 3. Identification of Fishing Vessels Based on Ship Trajectory Feature ２８
• 3.1 Materials and Methodology ２８
• 3.3 Parameters-Based Trajectory Analysis ４０
• 3.3 Results and Discussion ４４
• 4. Discussion ４９
• 4.1 Demonstration of 10 m Small Fishing Vessel Detection ４９
• 4.2 Application to Ship Tracking Using Sentinel-1 and VIIRS ５０
• 5. Conclusion ５３
• Relevant Publication ５４
• References ５５