Computer Vision for Bioacoustics: Detection of Bearded Seal Vocalizations in the Chukchi Shelf Using YOLOV5
Christian D. Escobar-Amado, Mohsen Badiey, Lin Wan
Abstract
Year-round recordings of bearded seal calls were collected in the northeastern edge of the Chukchi Continental Slope (Alaska, within the Arctic Circle) in 2016–2017, 2018–2019, and 2019–2020. While the underwater vocalizations of bearded seals are often analyzed manually or using automatic detections manually validated, in this article, a detection and classification system (DCS) based on the You Only Look Once Version 5 (YOLOV5) algorithm is proposed. With YOLOV5, the network learns how to detect and classify these marine mammals' calls using the principle of computer vision for object detection in images where bounding boxes enclose the objects of interest. During training, validation, and testing, YOLOV5 achieved an accuracy of 96.54%, 93.36%, and 93.87%, respectively. The DCS was applied to the three-year-long dataset, and an analysis of the vocal behavior of the bearded seals showed that there exists a geographical dependence where this species prefers shallower water depths in the Chukchi Continental Slope. Another advantage of using YOLOV5 over other typical DCS is that the predicted bounding boxes have embedded statistical information about the vocalization, such as the duration, bandwidth, and center frequency of the signals. This additional information equips biologists with statistical data that facilitate the analysis of animal vocal behavior.