Optical neural networks based on wavelet transform

The Fourier transform has been widely applied in the optical signal processing, yet it is just fit for analyzing the stationary signals. By extending the Fourier transform into wavelet transform, a new type of filter is proposed and its analogy to neural networks is developed. Optical neural networks (ONNs) are the new type networks, which possess good capacity of super parallel processing, signal transmission and high-density connecting lines. Although neural networks' implementations have been limited by the availability of high-resolution optical devices, by virtue of simple optical architectures for the wavelet transforms, the new neural network is easy to implement in large-scale by applying photoelectric technology. In this paper, the basic principles of ONNs and optical wavelet transform (OWT) are presented respectively, and the principle and structure of their combination-optical neural networks based on the wavelet transform are also proposed. For the optical neural networks and optical wavelet transforms, their optical implementations have many unique superiority, yet theirs combination takes on characteristics better than such structures just using neural networks or wavelet transform. Furthermore, their application perspectives are predicted in the paper.