Article
A Novel Subspace Alignment-Based Interference
Suppression Method for the Transfer Caused by
Different Sample Carriers in Electronic Nose
Zhifang Liang
1,
* , Fengchun Tian
2
, Ci Zhang
2
and Liu Yang
1
1
School of Communication and Information Engineering, Chongqing University of Posts and
Telecommunications, Chongwen Road 2nd, Nan’an District, Chongqing 400065, China;
yangliu@cqupt.edu.cn
2
School of Microelectronics and Communication Engineering, Chongqing University, 174 ShaZheng Street,
ShaPingBa District, Chongqing 400044, China; fengchuntian@cqu.edu.cn (F.T.); ZhangCi@cqu.edu.cn (C.Z.)
* Correspondence: liangzf@cqupt.edu.cn
Received: 9 October 2019; Accepted: 4 November 2019; Published: 7 November 2019
Abstract:
A medical electronic nose (e-nose) with 31 gas sensors is used for wound infection detection
by analyzing the bacterial metabolites. In practical applications, the prediction accuracy drops
dramatically when the prediction model established by laboratory data is directly used in human
clinical samples. This is a key issue for medical e-nose which should be more worthy of attention.
The host (carrier) of bacteria can be the culture solution, the animal wound, or the human wound.
As well, the bacterial culture solution or animals (such as: mice, rabbits, etc.) obtained easily are
usually used as experimental subjects to collect sufficient sensor array data to establish the robust
predictive model, but it brings another serious interference problem at the same time. Different
carriers have different background interferences, therefore the distribution of data collected under
different carriers is different, which will make a certain impact on the recognition accuracy in the
detection of human wound infection. This type of interference problem is called “transfer caused by
different sample carriers”. In this paper, a novel subspace alignment-based interference suppression
(SAIS) method with domain correction capability is proposed to solve this interference problem.
The subspace is the part of space whose dimension is smaller than the whole space, and it has some
specific properties. In this method, first the subspaces of different data domains are gotten, and then
one subspace is aligned to another subspace, thereby the problem of different distributions between
two domains is solved. From experimental results, it can be found that the recognition accuracy
of the infected rat samples increases from 29.18% (there is no interference suppression) to 82.55%
(interference suppress by SAIS).
Keywords: electronic nose; subspace alignment; interference suppression; transfer
1. Introduction
As an artificial nose, electronic nose (e-nose) can analyze the gas characteristics quickly by
simulating the biological olfactory system. It consists of a gas sensor array and an appropriate pattern
recognition system. The basic structure of an e-nose system is similar to that of the biological olfactory
system, in which the gas sensor array, the signal processing unit and the pattern recognition unit
respectively simulate the olfactory receptor, the olfactory bulb and the cerebral cortex in the biological
olfactory system of the mammal [
1
]. Due to its portability and ease of use, e-nose technology is
superior to other detection methods, such as chemical detection methods, gas chromatography and
mass spectrometer, etc. [
2
]. With the development of sensing technology and intelligent algorithms,
e-nose technology has been widely used in many areas, such as disease diagnosis by detecting specific
Sensors 2019, 19, 4846; doi:10.3390/s19224846 www.mdpi.com/journal/sensors
