Ann-based classifier of features produced by computer generated hologorams

摘要

The paper presents the ANN-based pattern recognition system with computer generated hologram (CGH) used as a feature extractor. Features obtaind by standard and optimized CGH are classified using multi layer perceptron network. Experiments with gradient and stochastic learning rules, as well as different hidden layer s izes for this system are presented. The objective in these experiments, were to classify the distortion of quasi-monomode optical fiber from speckle images taken when this distortion occurred.