A Sparse Reconstruction Domain Transfer Method for Interference Suppression in Artificial Olfactory System

In recent years, the electronic nose (E-nose) technology has made a great progress in practical application. However, the interferences, such as the interferences from non-target gas, manufacturing deviation and sensor drift, have a negative impact on the detection accuracy during the real-time detection process, which will affect the industrialization and commercialization of E-nose. Thus, it is necessary to research the method of interference suppression for realizing the long-term and stable detection. This paper proposes a sparse reconstruction domain transfer (SRDT) method to suppress the interference in E-nose. The proposed method is to find the optimal subspace for each dataset, and sparsely reconstruct the transformed data after transforming to the optimal subspace, which can realize the knowledge transfer in the data domains with and without interference information. Three different datasets of E-nose are used to verify the effectiveness of the proposed method, including the benchmark sensor drift data, the dataset collected by three identical sets of E-nose systems and the dataset of background interference caused by carrier gas. The average recognition rate is 92.97% in the first dataset. The recognition rate is 93.75% and 90.27% separately for the two tasks of the second dataset, and 85.16% and 84.62% separately for the two tasks of the third dataset. The experiment results show that different types of interference in E-nose can be satisfactorily suppressed.