A distributed acoustic sensing technology method for ship feature extraction based on multitask learning

To address the challenges in monitoring ”dark ships”—vessels that turn off their Automatic Identification System (AIS)—and the limitations of traditional remote sensing methods due to weather and lighting, this study investigates the application of Distributed Acoustic Sensing (DAS) technology for all-weather wide-area maritime surveillance. The method leverages existing submarine communication cables as a continuous sensor array to acquire subtle vibration signals caused by vessel movement in real time. The challenge of raw DAS signals having a low Signal-to-Noise Ratio (SNR) and being easily overwhelmed by strong marine background noise, a comprehensive data processing and feature enhancement pipeline was developed: this involves suppressing high-frequency noise using sliding window mean filtering and adaptive filtering, followed by applying techniques such as Z-score normalization and the Frangi enhancement filter to significantly enhance the acoustic and wake features of vessels, transforming the one-dimensional time-series data into high-contrast two-dimensional spatiotemporal waterfall maps. To enable automated and comprehensive extraction of vessel information from these enhanced DAS images, this research constructed a multi-task learning framework where a single model performs real-time vessel detection, type classification, and simultaneous heading estimation through regression. Experimental results demonstrate that this integrated methodology, combining DAS sensing, feature enhancement, and intelligent recognition, can efficiently and accurately extract multi-dimensional vessel information from complex backgrounds, confirming the effectiveness of DAS technology for maritime comprehensive situational awareness and providing key technical support for ”dark ship” detection.