Detecting free-ranging river dolphin using deep learning based algorithms

Abstract

Indus river dolphin is an endangered freshwater river dolphin and is included in IUCN red list of endangered species. Canal-strandings is one of the major issues that results in high mortality of the dolphin because the dolphins that are stranded into canals during flood season cannot return to the river during low-water season. As the dolphin uses high frequency echolocation clicks for navigation in muddy river waters, passive acoustic monitoring can be used for detection and localization of the dolphin and can assist in conservation efforts. In this context, we compare performance of variants of deep learning (LSTM) based event detection algorithms for dolphin click detection with conventional algorithms (such as the Teager Energy based click detection and Envelope Derivative Operator based click detection) and analyze the possible scenarios where machine learning based methods may assist in better performance by automatic detection of reflected clicks.