The absorption spectra of BWA/CWA often heavily overlap with each other and with absorption spectra of harmless species. The traditional approach of spectral discrimination usually involves estimation of concentration of each constituent, wherein the first- and second derivatives are being used as the spectrum features and the linear relationship between these features and the concentrations is sought by, e.g., the partial least squares or principal component regression. These algorithms may not be suitable for real-time early warning detection of BWA/CWA in the gaseous/liquid environments, especially taking into account the inevitable presence of environmental constituents with unknown spectra. In this paper, we present a new approach suitable for ragged, real-time spectral discrimination. In this approach, we are using an independent component analysis (ICA) technique to unmix the mixture spectra into independent spectral components. In order to classify the components, we have developed a special feature extraction algorithm based on a complex wavelet transform. We have tested the procedure experimentally using a ragged fiber-optics spectrometer working in the NIR region (800 - 1000 nm), and mixtures of organic liquids. The obtained results clearly demonstrate the applicability of the proposed system to the early warning "trigger"-type detection suitable for real-time environmental monitoring.