What question did this study set out to answer?

The study aims to develop an efficient method for detecting, classifying, and localizing bowhead whale calls during their migration.

May 14, 2026

Machine learning-based detection, classification, and localization of bowhead whale calls during their annual fall migration across the Beaufort Sea

Key Points

The study aims to develop an efficient method for detecting, classifying, and localizing bowhead whale calls during their migration.
Deployment of Directional Autonomous Seafloor Acoustic Recorders (DASARs) in the Beaufort Sea.
Implementation of a machine learning-based approach using a Faster R-CNN pre-trained on ResNet50.
Utilization of image intensity matching for localization of calls among multiple DASARs.
The machine learning model outperformed existing methods, showing improved accuracy in call detection and classification.
Utilized a dataset comprising 24 years of annotated bowhead whale call data for training.
Triangulation of detected calls enabled effective localization across the monitoring area.

Abstract

Annually since 2001, Greeneridge Sciences Inc. has deployed Directional Autonomous Seafloor Acoustic Recorders (DASARs) in the Beaufort Sea offshore of the North Slope of Alaska for the passive acoustic monitoring of bowhead whales (Balaena mysticetus). These studies aim to understand the potential impact of anthropogenic sounds on bowhead whales during their fall migration. Due to the large volume of acoustic data and the labor-intensive effort traditionally performed manually by human analysts, an efficient yet accurate method of identifying bowhead whale calls is desired. This work presents a machine learning based method for automated detection, classification, and localization of bowhead whale calls. Detection and classification is achieved through a faster region-based convolutional neural network (Faster RCNN) pre-trained on ResNet50. FasterRCNN was chosen over YOLO object detection models because high detection and classification accuracy was prioritized over real-time detection capability. Localization is achieved by image intensity matching of detected calls among DASARs and triangulating bearing estimates of associated calls. This model and software package outperforms existing automated approaches partly due to the large dataset available, which includes 24 years of annotated data of bowhead whale calls categorized by call type, as well as other marine mammal calls and anthropogenic sound sources.

Ask AI

Mark Helpful

Bookmark

Relay