Research on Electronic Nose Drift Suppression Algorithm based on Classifier Integration and Active Learning

Abstract

In the field of electronic nose(E-nose) research, the underlying gas sensor is affected by environmental changes, aging of its own devices, sensor poisoning and other factors, which will cause the detection value to drift. Besides, the need for a large number of labeled samples in the pattern recognition algorithm will lead to excessively high model training costs. In order to solve the problems mentioned above, a method that combines classifier integration and active learning to reduce the model training cost by reducing the number of labeled samples is proposed in this paper. Using this method, the trend of sensor drift is captured by classifier integration, the number of single-labeled samples is dynamically adjusted, and finally the drift of the gas sensor array is suppressed. From the experiment results, it can be found that the sensor drift can be satisfactorily solved by the proposed method.