Near Real-Time Radio Frequency (RF) Data Analysis Pipeline for Aiding Marine Domain Awareness and Surveillance

Maritime Domain Awareness (MDA) is primarily driven by an Automatic Identification System (AIS), an automated tracking system that identifies and displays other vessels in the vicinity. There have been several implementations of AIS, nevertheless this algorithm demonstrates a near real-time Machine Learning (ML) pipeline that detects and clusters the radio frequencies (RF) received from a Software Defined Radio (SDR) implementation of the maritime AIS, operated by USM. The hardware setup consists of an SDR, raspberry-Pi, and a super wide band antenna. The near real-time analytic pipeline reads the RF signals every five minutes and performs a feature-learning based random forest classification of the detected RF frequencies. The model with the selected feature-set and 500 estimators (yielding the lowest Out-Of-Bag error rate of 0.10) was used to test the streaming signals every five minutes. With a 96% classification accuracy, this SDR-ML pipeline demonstrated the ability to be used in port surveillance and security by detecting the center frequencies and bandwidths of the local active signals around the port in near real-time. Further, by integrating the AIS tracking and detection, the source of a suspicious RF signal can also be classified which would play a key role in ensuring MDA.