Acoustic estimation of the manatee population and classification of call categories using artificial intelligence

• DOI: 10.3389/fcosc.2024.1405243
• Publication Date: 2024-06-11T16:45:38Z
• Authors: Sebastian Schneider; Lorenzo von Fersen; Paul Wilhelm Dierkes
• Keywords: Manatee

The population sizes of manatees in many regions remain largely unknown, primarily due to the challenging nature conducting visual counts turbid and inaccessible aquatic environments. Passive acoustic monitoring has shown promise for wild. In this study, we present an innovative approach that leverages a convolutional neural network (CNN) detection, isolation classification manatee vocalizations from long-term audio recordings. To improve effectiveness call detection classification, CNN works two phases. First, recording is divided into smaller windows 0.5 seconds binary decision made as whether or not it contains call. Subsequently, these are classified distinct vocal classes (4 categories), allowing separation analysis signature calls (squeaks). Signature further subjected clustering techniques distinguish recorded individuals estimate size. was trained validated using recordings three different zoological facilities with varying numbers manatees. Three methods (community classifiers HDBSCAN) were tested their suitability. results demonstrate ability accurately detect effectively classify categories. addition, our study demonstrates feasibility reliable size estimation HDBSCAN method. integration offers promising way assess populations visually autonomous devices. differentiate between categories will allow ongoing important information such stress, arousal, calf presence, which aid conservation management critical habitats.