Multiaspect target detection via the infinite hidden Markov model

Area of Science:

• Signal Processing
• Machine Learning
• Acoustics

Background:

• Traditional target detection methods often struggle with complex scattering phenomena.
• Hidden Markov Models (HMMs) offer a framework for sequential data analysis but require pre-defined states.
• The need for adaptive and robust methods in analyzing wave scattering data is critical.

Purpose of the Study:

• To introduce a new multiaspect target detection method utilizing an infinite hidden Markov model (iHMM).
• To model wave scattering from targets using an iHMM with an infinite number of states.
• To infer posterior distributions on the number of states and collectively learn target-dependent states.

Main Methods:

• Developed a multiaspect target detection framework based on the infinite hidden Markov model (iHMM).
• Employed Dirichlet processes (DPs) to define HMM transition matrix rows, linked via a hierarchical Dirichlet process.
• Utilized a Gibbs sampler for learning and inference within the iHMM framework.

Main Results:

• Successfully inferred a full posterior distribution on the number of states associated with targets.
• Demonstrated collective learning of target-dependent states within the iHMM.
• Applied the framework to analyze measured acoustic scattering data, validating its effectiveness.

Conclusions:

• The proposed iHMM provides a powerful and flexible approach for multiaspect target detection.
• This method effectively handles the complexity of wave scattering by adaptively determining the number of states.
• The framework shows promise for real-world applications, particularly in acoustic target analysis.