Wavelet neural networks - innovative processing strategy for classification and quantification models

Biologically inspired technologies for artificial sensing have been investigated in the last years using different kinds of sensors and applying novel computational paradigms, in order to emulate the behavior of their biological counterparts. In this sense, the use of advanced processing techniques is essential in interpreting the obtained information. Among the available processing tools, the combination of Wavelet Transform (WT) and Artificial Neural Networks (ANNs) has been adopted as innovate strategy for processing complex signals, especially when they are composed of a large number of measures and include non-linear content. Wavelet Neural Networks (WNNs) are a new class of networks that combine wavelets and neural networks in order to improve tasks such as classification, prediction and modeling, taking benefit of both paradigms. This chapter describes the fundamentals of WNNs and their advantages in the treatment of signals, especially when feature extraction stage and/or signal preprocessing are mandatory to build an appropriate calibration or classification model. This chapter will highlight the successful applications of these networks in areas such as electronic tongues and myoelectric analysis. To demonstrate the computing power of the WNNs, two case studies from our laboratories are presented. The first relates to the quantification of analytes using a voltammetric electronic tongue and the second involves the classification of motor unit spikes at different levels of muscle contraction from myoelectric records obtained with a needle electrode. © 2012 by Nova Science Publishers, Inc. All rights reserved.