A collaborative classification algorithm with multi-view terahertz spectra

Abstract

Terahertz time-domain spectroscopy (THz-TDS) has been widely used for food and drug identification. Several types of THz spectral data are usually obtained by using THz-TDS. It is crucial to make full use of these multi-view THz spectra. In this paper, a novel feature selection method based on optimal interval search followed by complementary feature seeking is proposed, which ensures to find the optimal characteristic peak without omitting other complementary features and reduces the dimension of data to improve efficiency. On the basis of the novel feature selection, we further propose a collaborative decision-making method for multi-view THz spectra. It utilizes the validation values of different THz spectrum’s that obtained in the process of the feature selection to construct weights, and applies a decision-level weighted fusion with these weights to make full use of the information provided by various THz spectra. By using the proposed feature selection and multi-view THz spectra collaboration, a few dozens of effective features can be extracted from hundreds of THz spectral features, and the discrimination of the origin, year and variety of three traditional Chinese medicine herbs can be obtained with higher classification accuracies. The collaborative use of multi-view THz spectra is demonstrated to have more reliable classification performance than that of single THz spectrum.