Sparsity-Based Representation for Classification Algorithms and Comparison Results for Transient Acoustic Signals
Technical Report,01 Jun 2015,30 Jun 2016
US Army Research Laboratory Adelphi United States
Pagination or Media Count:
In this report, we propose a general sparsity-based framework for the classification of transient acoustic signals this framework enforces various sparsity structures like joint-sparse or group-and-joint-sparse within measurements of multiple acoustic sensors. We further robustify our models to deal with the presence of dense and large but correlated noise and signal interference i.e., low-rank interference. Another contribution is the implementation of deep learning architectures to perform classification on the transient acoustic data set. Extensive experimental results are included in the report to compare the classification performance of sparsity-based and deep-network-based techniques with conventional classifiers such as Markov switching vector auto-regression, Gaussian mixture model, support vector machine SVM, hidden Markov model HMM, sparse logistic regression, and the combination of SVM and HMM methods SVM-HMM for 2 experimental sets of 4-class and 6-class classification problems.