Adaptive Optical Neural Network Classifier Systems

摘要

We present two adaptive opto-electronic neural network hardware architecturescapable of exploiting parallel optics to realize real-time processing and classification of high-dimensional data. The architectures are based on radial basis function neural networks that employ on-line training techniques to offer robustness to noise and optical system imperfections. A binary-input data classifier is presented first and the issues of system imperfections, device characterization, and system noise are addressed. The experimental results from the optical system are compared with data from a computer model of the system in order to identify critical noise sources and to indicate possible areas for system performance improvements. A grayscale-input classifier is then proposed for handling 8 bit input data to broaden the range of applications of the classifier. An optical wavelet transform system intended for use as a multi-resolution image preprocessor for the classifiers is then presented and discussed.