Passive Acoustic Monitoring of Arctic Surface Ships with DeepPAM

As new remote underwater acoustic sensor suites are installed and begin streaming unprecedented volumes of data, automated analysis becomes increasingly important. Automated passive acoustic monitoring (PAM) for contacts of interest such as surface ships or marine mammals is desirable. To this end, we are developing DeepPAM, a deep learning based passive acoustic monitoring system that can detect and classify multiple simultaneous surface ships in passive acoustic data collected in Barrow Strait in the Canadian Arctic. Herein, we present early results of DeepPAM trained and tested to detect and classify 20 unique vessels in data from the Northern Watch seabed arrays. The effects of varying spectral input data pre-processing on performance is explored, followed by preliminary applications of DeepPAM to data from the Barrow Strait Real Time Observatory.