Optical perceptron learning for binary classification with spatial light rebroadcasters

摘要: Binary classification of an object in a two-dimensional image is considered. A spatial light rebroadcaster shown to be advantageous for learning this case because it can store the weights and permit upward downward adjustments. Two algorithms, based on perceptron, are modification perceptron algorithm developed so that only positive needed. This convenient intensity only. The modified converge finite number steps linear separable classes. Optical experiments show four characters two groups, which alternative groupings used robustness. In second group complements used, convergence equally fast. Adding results from original complementary patterns provides discrimination superior obtained using either its own.