Abstract
This work is concerned with the problem of detecting right whales from autonomous surface vehicles (ASVs) and investigates the effectiveness of a range of deep learning methods. Given the success of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) across many different applications, these form the basis for comparison. From the input audio, power spectral features are extracted and consideration is given to how their time resolution and frequency resolution affects the detection accuracy and the number of points that need to be processed which is an important consideration within the limited processing power on an ASV. The effect of downsampling the input audio before feature extraction is also investigated. Tests establish that CNNs consistently give best performance on the detection task with accuracy of over 92% compared to around 90% with RNNs. Furthermore, tests measuring the processing time for detection found the CNN to be three times faster than
the RNN
the RNN
Original language | English |
---|---|
Publication status | Published - 2019 |
Profiles
-
Jason Lines
- School of Computing Sciences - Associate Professor in Computing Sciences
- Data Science and AI - Member
- Smart Emerging Technologies - Member
Person: Research Group Member, Academic, Teaching & Research
-
Ben Milner
- School of Computing Sciences - Senior Lecturer
- Interactive Graphics and Audio - Member
- Smart Emerging Technologies - Member
Person: Research Group Member, Academic, Teaching & Research