Gas detection method based on self-adaption of semi-supervised domain

Abstract

The invention discloses a gas detection method based on the self-adaption of a semi-supervised domain. The gas detection method comprises the following steps: preprocessing gas data signals acquired by a gas sensor; constructing a feature subspace by virtue of the preprocessed signals; establishing a combined kernel function according to the constructed feature subspace; selecting label-free samples in a target domain; and training a classifier according to the selected label-free samples in the target domain, and carrying out gas identification according to the trained classifier. According to the gas detection method, a selection strategy for the label-free samples in the target domain is provided by constructing the self-adaptive kernel function of the domain according to the time sequence characteristic of the data of the gas sensor, and thus the data and drifting of the gas sensor can be effectively processed; the new kernel function not only considers that the subspace adjacent to a source domain and the target domain deserves larger weight, but also considers the middle data between the source domain and the target domain, and the drifting of the gas sensor is described by virtue of a Grassmann geometrical flow pattern, so that the influence of the drifting on the gas detection and identification is effectively avoided.

Description

Technical field [0001] The invention belongs to the technical field of gas detection, and specifically relates to a gas detection method. Background technique [0002] The use of an artificial olfactory system (electronic nose) for gas detection and recognition is one of the research hotspots in the field of artificial olfaction at home and abroad. It has a wide range of applications in air pollution detection, chemical plant detection and monitoring, battlefield gas detection and medical food detection, etc. . [0003] The electronic nose is mainly composed of two parts: a gas sensor array and a pattern recognition system. A plurality of gas sensors form an array, which produces chemical and physical reactions on the sensor film through the adsorption of molecules filled with gas, and converts the reactions into electrical signals to obtain response signals. The response signal is used as the input of the pattern recognition system after data processing and analysis. The pattern...

AI Technical Summary

Problems solved by technology

[0004] However, with the long-term use of the electronic nose, the response signal collected by the gas sensor will drift seriously, that is, the response signal data obtained by the same concentration and type of gas in the same environment at different time periods is different, which increases the detection and identification of gas. It is also one of the difficult problems that the current artificial sense of smell needs to solve.

Drift causes the distribution of test data and training data to be inconsistent, which leads to low recognition rate of the learned recognition algorithm

Most of the existing drift correction methods use some kind of linear transformation and calibration reference gas to eliminate the drift of the sensor, but the actual drift is not linear, and the calibration reference gas increases the maintenance cost of the system

Images