Bionic Nose Active Sensing Based On Partially Observable Markov Decision Process

As a simulation of biological olfactory system, bionic nose is a device that can recognize simple or complex gas and a concentrated expression of artificial olfactory researches, mainly consists of gas sensor array, data preprocessing system and pattern recognition system. Compared with the biological olfactory system, the implementation of existing bionic nose system is primarily a passive process, but as an active process, biological olfactory adjusts the velocity of the gas flowing into the nasal passages and makes olfactory receptor cells in the nasal cavity closest to an array of patterns already stored in the brain. Currently, active sensing in machine vision and other fields have achieved rapid development and application. But the application in the field of bionic nose still in its infancy and have a little relevant research, have great development potential, value and significance.Currently, researches on the bionic nose focus on optimizing the system architecture, extracting feature vectors, improving pattern recognition algorithms and modulating temperature of gas sensor actively and so on. About those researches, there is the following deficiency to improve the time efficiency and identification efficiency of the bionic nose:Researchers indulge in extensive researches from the perspective of active temperature modulation and optimizing system architecture of the bionic nose, ignoring the contact of biological olfactory and behavior and the chromatographic effect of gas molecule in mucosa, the information obtained can’t response to identify the characteristics of the object from the essence, the recognition tasks of the bionic nose lost the value in use. Based on the research of the research status of the active sensing, the principle of the bionic nose, the relationship between biological sniffing and olfactory perception, the chromatographic effect of nasal mucosa, the basic theoretical knowledge of the partially observable Markov decision model (POMDP) and the hidden Markov model (HMM), this paper puts forward an active sensing thought based on the velocity modulation and the chromatographic effect of nasal mucosa and an active sensing model based on the POMDP. In this paper, the main work is as follows:This paper introduces the concept of the active sensing from the biological active sensing, studies the research status of the active sensing in bionic nose. Having some researches on the relationship between biological sniffing and olfactory perception, the chromatographic effect of nasal mucosa, and putting forward an active sensing thought based on the velocity modulation and the chromatographic effect of nasal mucosa.On the basis of existing work, the paper designs and implements a hardware system of bionic nose based on the velocity modulation and software-based information collection system of the hardware system. And through velocity modulation experiment and analysis of bionic nose, have verified that the research of the active sensing about bionic nose has sufficient theoretical basis and the high reliability of the bionic nose system.Using the active sensing thought based on the velocity modulation and the chromatographic effect of nasal mucosa, this paper proposes a model named active flow velocity sensing model (ASM), namely a bionic olfactory model with active sensing function. The model combines the advantages of POMDP and HMM, integrates an active sensing strategy based on Bayesian risk, can take actions actively or make decisions according to its environment and prior knowledge of the environment. In addition, the difference between the model and other active sensing model based on HMM is making the continuous HMM discrete for modeling the sensory array.Finally, this paper applies the ASM algorithm to olfactory recognition task, and takes some experiments to four kinds of VOC gases respectively, and makes a comparison between the ASM algorithm and the pattern recognition algorithm used in bionic nose commonly. The experimental result shows that the ASM algorithm has a good stability and high recognition performance, and can complete the identification process quickly compared to other pattern recognition algorithms, and verifies the feasibility of the active sensing thought based on the velocity modulation and the chromatographic effect of nasal mucosa and ASM based on POMDP in the information processing of bionic nose.