Abstract: Awavelet neural network based on wavelet analysis can be used to represent chemical signals. The adaptive wavelet neural network using the continuous wavelet transform has problems of a high redundancy and slow training, and the compactly supported orthogonal wavelet neural network using the discrete wavelet transform is difficult to be applied, because the compactly supported orthogonal wavelet function with analytic form is hard to build. Based on the idea of the compactly supported orthogonal wavelet network, a novel algorithm using the compactly supported B-spline function instead of the compactly supported or thogonal wavelet function is proposed. It has been applied to the compression and de- noising of chemical signals. Compared with the adaptive wavelet neural network, the speed of our algorithm was enhanced greatly.

Key words: Wavelet neural network, B-spline function, Compression, De-noising