DECISION-BASED NEURAL NETWORKS WITH SIGNAL IMAGE CLASSIFICATION APPLICATIONS

Supervised learning networks based on a decision-based formulation are explored. More specifically, a decision-based neural network (DBNN) i s proposed, which combines the perceptron-like learning rule and hiera rchical nonlinear network structure. The decision-based mutual trainin g can be applied to both static and temporal pattern recognition probl ems. For static pattern recognition, two hierarchical structures are p roposed: hidden-node and subcluster structures. The relationships betw een DBNN's and other models (linear perceptron, piecewise-linear perce ptron, LVQ, and PNN) are discussed. As to temporal DBNN's, model-based discriminant functions may be chosen to compensate possible temporal variations, such as waveform warping and alignments. Typical examples include DTW distance, prediction error, or likelihood functions. For c lassification applications, DBNN's are very effective in computation t ime and performance. This is confirmed by simulations conducted for se veral applications, including texture classification, OCR, and ECG ana lysis.