Researcher(s)
- Eric Quezada, Computer Science, University of Delaware
Faculty Mentor(s)
- Mohsen Badiey, Electrical and Computer Engineering, University of Delaware
Abstract
The accurate detection and classification of marine mammal vocalizations is a critical component of passive acoustic monitoring (PAM) for marine conservation and ecological research. However, validating algorithm performance in the complex, noisy conditions of the open ocean is a significant challenge. By using synthetic acoustic signals (controlled replicas of vocalizations) we can eliminate environmental variables and establish a reliable ground truth for rigorous system evaluation.
This project documents the initial trials of the University of Delaware’s Passive Localized Underwater Transiting Observing (PLUTO) system. A series of synthetic bearded seal moans and harp seal trills were transmitted and recorded in a controlled environment. The resulting dataset provided a unique opportunity to directly assess the detection and classification algorithms developed by the University of Delaware’s – Ocean Acoustic Engineering Lab. This controlled testing allows us to establish a crucial baseline for our tools’ performance before their deployment in real-world scenarios.
Our analysis focuses on processing these recorded signals and comparing the outputs of our algorithms against the known ground truth. We will evaluate performance using key metrics such as precision and recall to identify and diagnose specific areas for improvement, such as reducing false positives or increasing signal recognition in low signal-to-noise ratio conditions. This work is a foundational step in validating and refining the PLUTO system, paving the way for the development of more robust and accurate PAM tools capable of operating effectively in complex marine environments.