Automatic Multiple-Stage Classification of Sleep Stages of Songbirds

• Activity Number: 466
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Hong Xu*+ and Zhiyi Chi and Daniel Margoliash and Sylvan Shank
• Companies: Nanyang Technological University and University of Connecticut and The University of Chicago and The University of Chicago
• Keywords: non-homogeneous hidden Markov model (NH-HMM) ; EEG ; classification ; songbirds ; learning ; maximum-a-posterior (MAP)

Abstract

Electroencephalography (EEG) recording has been proved a powerful tool to uncover latent states of brain. Songbirds have been studied to provide important models on vocal and language learning. In this work, we aimed to classify multiple stages for the sleep of songbirds. The data consisted of whole-night recordings of EEG during sleep of songbirds at sampling frequency of 1kHz. With such large datasets, the conventional manual classification used by biologists takes long time and is hard to achieve an objective standard. We developed various techniques to automatically analyze the EEG. In particular, we proposed a non-homogeneous hidden Markov model (NH-HMM) for the estimated spectral density in multiple frequency bands and used the maximum-a-posterior (MAP) criterion for classification. Its effectiveness was further demonstrated by both simulations and application to the EEG re