Committee: 

Dr. Anderson, , Advisor         

Dr. Davenport, Chair

Dr. Clements

Dr. Daley

Abstract: The objective of the proposed research is to enable detection and  classification of anomalies in poultry production facilities through  sparse, dictionary-based representations of the auditory environment.   This research focuses primarily on the detection of respiratory disease  in chickens, but should be extensible to other types of problems or  conditions that may arise.  The scale of the poultry industry creates significant financial potential for the development of signal processing  techniques capable of monitoring varying production settings with  minimal human intervention. However, there are numerous challenges to  the application of audio signal processing in this domain that must be  addressed.  For example, the vast majority of the data is unlabeled, and  it contains significant amounts of noise.  Conditions throughout the day  also cause short-term changes in the auditory environment, and the aging  of the chickens cause drift over the long-term.  The proposed work seeks  to overcome these challenges through a frozen dictionary approach to disease detection and several novelty detection approaches based on  dictionary learning for sparse representations.