Optical Neural Network Classifier Architectures

摘要: Abstract : We present an adaptive opto-electronic neural network hardware architecture capable of exploiting parallel optics to realize real-time processing and classification high-dimensional data for Air Force Hostile Target Identification (HTI). This utilizes a grayscale-input radial basis function based on previously demonstrated binary-input version. The greyscale-input capability broadens the range applications classifier by allowing it handle 8 bit input data. characterized key component this system, variable phase retarder, found that uniformity changed less than 7% with applied voltage. An optical wavelet transform preprocessor is also discussed. produces reduced feature set multiwavelet images improve training times discrimination network. design uses joint correlator (JTC) provide cross correlations multiple images. experimental results JTC which used four generated spatial light modulator. then propose using functions as perform extraction. from retarder characterization work were be in software simulation system determine its feasibility. However, remains unfinished project was canceled due budget cuts.