Integrated Detection of Ships and Their Wakes Using Sentinel-2A Imagery and Automated Machine Learning Approaches

This study investigates advanced methods for improving ship and wake detection using satellite imagery, focusing on the coastal area of Yeongdo, Busan, South Korea. Sentinel-2A multispectral data, specifically Bands 2, 3, 4, and 8, were utilized to enhance detection accuracy, particularly for small vessels. The Ship Detection Index (SDI) was introduced to refine the detection process, and Automated Machine Learning (AutoML) was implemented via the Tree-based Pipeline Optimization Tool (TPOT) to automate data preprocessing, model selection, and hyperparameter optimization. K-Nearest Neighbors (KNN) was identified as the optimal model for ship detection, whereas the Multilayer Perceptron (MLP) was chosen for wake detection, outperforming established methods such as convolutional neural network (CNN). A novel shadow differentiation algorithm was also developed to distinguish ships from their shadows, further improving detection precision. Testing across multiple scenes confirmed the robustness of the models, revealing high precision, recall, F1-score, and intersection over union (IoU). This study demonstrates the effectiveness of integrating AutoML with optical satellite data for maritime situational awareness; its potential applications include marine monitoring, traffic management, and illegal fishing detection.