Monitoring Frog Communities: An Application of Machine Learning

Abstract

Automatic recognition of animal vocalisations would be a valuable tool for a variety of biological research and environmental monitoring applications. We report the development of a software system which can recognise the vocalisations of 22 species of frogs which occur in an area of Northern Australia. This software system will be used in unattended operation to monitor the effect on frog populations of the introduced Cane Toad. The system is based around classification of local peaks in the spectrogram of the audio signal using Quinlan's machine learning system, C4.5 (Quinlan 1993). Unreliable identifications of peaks are aggregated together using a hierarchical structure of segments based on the typical temporal vocalisation species ' patterns. This produces robust system performance. Problem Description Since the unfortunate introduction of the Cane Toad (Bufo marinus) to Australia, its abundance and continuing spread through northern Australia have been the cause of considerable concern. It is a voracious predator taking a wide range of prey. Cane Toads also possess poison glands which can kill unwary animals which attempt to prey on them. Although