Energy function for learning rotation invariant in a cascaded neural network model
Shengjiang Chang, Kwok‐Wo Wong, Wenwei Zhang, Yanxin Zhang · Optik · 1999
A cascaded neural network model is employed to perform multi-target rotation invariant classification. Based on the model, a novel algorithm for training binary interconnection weight matrix (IWM) using Hopfield network is proposed. Computer simulations show that this approach is an effective method. Compared with other existing methods. the proposed algorithm offers a faster rate of convergence. A real-time optical processing system for the implementation of the neural network is presented. By hiasing the interconnection weights to yield a non-negative weight matrix and adding a threshold subchannel, the optical system can realize, in real-time, the bipolar weighted summation in a single channel. Preliminary experimental results are shown.