Mass spectra separation for explosives detection by using PLCA with an attenuation model

We propose a new method to separate mass spectra into individual chemical compounds for explosives detection. The conventional method based on probabilistic latent component analysis (PLCA) is effective in many cases because the method can solve the problems of non-negativity and non-orthogonality by using sparsity of the domain of explosives detection. However, multiple compounds tend to be merged into a single basis component, and a single compound tends to be split into multiple basis components in error because the method does not model temporal structure of mass spectra. In this paper, first, we introduce a separation method based on shift-invariant PLCA (SIPLCA) in order to model temporal structure. Next, in order to prevent overfitting, we introduce an attenuation envelope such that it imposes a temporal constraint by focusing on the fact that the intensity of chemical compounds is attenuated with time after passing through the detector. Experimental results indicate that the proposed method outperforms the PLCA-based conventional method and other simple SIPLCA-based methods.