Gruden Pina, Nosal Eva-Marie, Henderson E Elizabeth
Ocean and Resources Engineering, University of Hawai'i at Manoa, Honolulu, HI, 96822, USA.
Whale Acoustic Reconnaissance Project, NIWC Pacific, San Diego, CA, 92152-5001, USA.
Sci Rep. 2025 May 13;15(1):16505. doi: 10.1038/s41598-025-00535-z.
Passive acoustic monitoring (PAM) is a key technology for studying marine mammal populations. PAM typically generates large volumes of data that contain signals from multiple overlapping sources. To extract meaningful information from these data, automated tools are required that can cope with multiple sources, missed detections, and false alarms. This paper presents the Multiple-Animal Model-Based Acoustic Tracking (MAMBAT) framework, which integrates model-based localization with Bayesian multi-target tracking to automatically track multiple sound sources using acoustic data from wide baseline arrays. MAMBAT leverages a "Track-before-Localize" strategy followed by a "Localize-then-Track" strategy that does not require detection, classification, or association steps. The framework's effectiveness is demonstrated through application to real-world datasets that contain multiple sperm whales from two ocean basins. MAMBAT advances our ability to monitor marine mammal distribution, abundance, and behavior, with potential to provide valuable information for conservation and management efforts.
被动声学监测（PAM）是研究海洋哺乳动物种群的一项关键技术。PAM通常会生成大量数据，这些数据包含来自多个重叠声源的信号。为了从这些数据中提取有意义的信息，需要能够处理多个声源、漏检和误报的自动化工具。本文介绍了基于多动物模型的声学跟踪（MAMBAT）框架，该框架将基于模型的定位与贝叶斯多目标跟踪相结合，以利用来自宽基线阵列的声学数据自动跟踪多个声源。MAMBAT采用“先跟踪后定位”策略，随后是“先定位后跟踪”策略，该策略不需要检测、分类或关联步骤。通过将该框架应用于包含来自两个海洋盆地的多头抹香鲸的真实世界数据集，证明了其有效性。MAMBAT提高了我们监测海洋哺乳动物分布、数量和行为的能力，有可能为保护和管理工作提供有价值的信息。
