A Decision Fusion for Source Detection Based on Multimodal Supervised Spectral Unmixing

This paper addresses the problem of source detection in unknown chemical mixtures in the context of multimodal measurements of spectral data. The proposed approaches are based on supervised linear spectral unmixing under non-negativity and sparsity constraints with adapted variants of the orthogonal matching pursuit algorithm. Two detection strategies are introduced: fusion of independent unimodal detection results and fusion by a joint multimodal decomposition and detection. Results are evaluated using a real database of ion mobility mass spectrometry (IMMS) data. A significant increase of the detection accuracy is obtained using the joint decomposition based decision as compared to the single modality detection results.