Nirosha Pitakande Gedara

Citation

Thesis Title
Wavelet-Based Birdsong Recognition for Conservation

Read article at Massey Research Online:

An important part of conservation is knowing what animals are present in an environment, and how common they are. For birds, this relies largely upon their calls. Automatic methods of recording birds in their natural environment have improved massively in the last decade, but automatically recognising which birds are calling is difficult. Two features that make the problem particularly challenging are the fact that the field recordings are noisy and the birds are often a long way from the microphone making their calls faint. Mrs Pitakande Gedara developed a method to denoise field recordings using a mathematical technique known as wavelets, significantly improving the quality of the recordings. A scalable method was devised to automatically identify a target species from continuous field recordings. She validated the method on automated recordings of four New Zealand bird species and an external dataset, demonstrating high sensitivity and robustness even compared to human experts.

Supervisors
Professor Stephen Marsland
Dr Amal Punchihewa
Professor Isabel Castro