A Novel Method for Spectral Signal Pattern Recognition

In this paper, we present a novel method for spectral signal pattern recognition. Considering the characteristics of spectral signal, we designed a statistical mixture model which combined several radial basis function neural networks, and built a cost function for the recognition system. We derived the EM-like algorithm to estimate the model parameters through optimizing the system cost function. Consequently, a pattern recognizer was constructed for spectral feature recognition. Based on the real-world Raman spectral signals, we investigated the proposed algorithm for estimating mixture model parameters, then verified the constructed spectral pattern recognizer. In addition, we also investigated various methods to construct input feature vectors for the recognizer, such as characteristic wavelength with corresponding spectral intensity, or principal component analysis, and the effect of these input vectors applied to spectral pattern recognizer. Experimental results show that the model parameters can be estimated by the proposed algorithm effectively, and the constructed spectral pattern recognition model has a higher recognition accuracy. The method proposed in this paper for spectral pattern recognition may find relatively wide applications since it is based on a more general consideration.