Features-oriented filtering of biological signals

This chapter is focused on neural signal processing, specifically on mergin g classic neural filtering processes with the neural features extraction me thods. Neural signal filtering, pattern recognition and control are still v iewed as subsequent, completely separate stages in these types of tasks. Mo dalities of merging these subfields and methods are investigated. The objec tive of this chapter is to show how hybrid techniques can handle merging th e signal filtering and conditioning level with the recognition level, as we ll as possibly the control level. The merging technique concerns mainly the design of an intelligent link bet ween the neural filter and the features extractor. The emphasis is on the a ssociated learning systems, which are able to form a useful semantic link, and to the degree of generality of the presented methods. The basic concepts related to features space filtering are introduced in th e first two sections of this chapter. In the following sections, two differ ent neural network features extractors are described: the Principal Compone nt Analysis (PCA) network and the Radial Basis Functions (RBF) network. The se features extractors are used for filtering applications in hybrid neural systems in the fifth section. Several results of ECG signal filtering in f eature spaces are presented in the sixth section. Concluding remarks and an appendix end the chapter.