Plume Localization Using Fuzzy Hidden Markov Model: An Efficient Decoding Method

This paper considers the problem of finding the location and propagation path of chemical source using multiple intensity sensors. Source origin localization is treated as the sequence estimation problem over a fuzzy hidden Markov model. The main advantage of using fuzzy hidden Markov model is its inherent nonstationary behavior with time varying state transition probabilities which characterizes the uncertainties in advection and diffusion of the plume source. Unfortunately, estimation algorithms of Viterbi type can be very inefficient due to the large size of state space. Motivated by a recent theoretical result on low complexity maximum likelihood sequence detection, we propose a greedy heuristic algorithm to obtain a candidate source path and search only for state sequences within a constrained Hamming distance from the candidate path. Our approach is applicable to fuzzy hidden Markov decoding problem with a general class of fuzzy measures and fuzzy integrals. From simulation study, we found that our algorithm has plume localization error close to that using fuzzy Viterbi algorithm. It is very effective for state estimation over a long observation sequence when the decoding error probability is small.