Math Colloquium: Research Presentations
Thursday, October 12, 2017 at 11:00am
VanderWerf Hall, 102 27 Graves Place, Holland, MI 49423-3617
“What Bird Was That? Feature Extraction of Recorded Bird Songs for Neural Networks” by Sarah Seckler, Summer 2017 Math Research student
In the past, researchers at Hope have worked towards identifying birds from recorded bird songs through using wavelets, image processing and neural networks.The general aim of our project is to extend this work to provide greater computational efficiency and accuracy in identification of bird songs. In this talk I will focus on taking a recorded bird song signal and extracting data from it to make it a suitable input for a neural network. This feature extraction process will involve using wavelets and related methods to create an image called a scalogram, encoding the key aspects of the sound including frequency and time. Our work focuses primarily on finding more efficient ways to extract these images, allowing us to analyze much larger data sets.
“Name That Bird: Using Neural Networks to Identify Birds” by Russell Houpt, Summer 2017 Math Research student
Can a computer learn to identify a bird by analyzing samples of its song? This research explores how neural networks can be used to identify different birds from recordings of their songs. We explore convolutions, wavelets, and neural networks, how they work together, and what techniques were employed to teach the programs how to quickly and accurately identify birds. In earlier work, a research group at Hope College made progress on this question by using neural networks to classify bird songs on a somewhat limited scale. Our results extend this work by using similar techniques on larger data sets, improving the accuracy and speed of the analysis, and modifying the existing algorithms to take advantage of multiple core computers.
No recent activity