Rough Sets with Real Valued Attributes in Evolutionary Optimization of Holographic Ring Wedge Detector Dedicated for Neural Network

The paper presents application of neural network in a hybrid, high speed, pattern recognition system. The feature extraction part is built as a grating based holographic ring wedge detector and the classifier is a probabilistic neural network. Since the feature extractor can be produced with relatively low costs from computer generated high resolution masks, such masks should be designed specifically to given recognition task. This requires automatic knowledge acquisition and processing with the goal of optimization of the feature space dedicated for subsequent use of neural network classifier. Appropriate methodology, proposed by the author in earlier works, has been enhanced by novel author’s modification of the notion of indiscernibility relation in theory of rough sets. New, generalized version of this relation, makes possible natural application of discrete type of rough knowledge representation into problems operating in continuous space and therefore using, like neural networks do, real valued data. Key-Words: hybrid systems, neural networks, rough sets, holographic ring-wedge detectors, image processing, pattern recognition