| Statistical pattern recognition has been successfully applied in several application areas for many years. The primary goal of statistical pattern recognition is classification, where a pattern feature vector is assigned to one of the finite number of pattern classes. This study assumes that each pattern class is characterized by a normal probability distribution on the measured features. This research addresses the use of a perceptron neural network model employing principal component analysis for implementing a statistical pattern recognizer.; The proposed neural network model contains a single hidden layer with a small number of nodes. The number of hidden nodes is determined by the desired percentage of the variation in the selected principal components, which are obtained from the principal component analysis of the feature vector training sample. The weights between the input layer and hidden layer are derived from the principal component analysis of the feature vector training sample. The weights between the hidden layer and the output layer are computed by using the delta-rule iteratively in the training procedure.; The significance of this research lies in the fact that the proposed pattern recognizer significantly reduces the number of training iterations and the computational time of training compared to the conventional back-propagation training method. Meanwhile, the expected accuracy is still preserved.; An application of this proposed model is tested to classify a 256 x 256 pixel aerial image, which contains several normally distributed pattern classes. Each pattern point is treated as a vector of nine features, eight neighbors and itself. In the training procedure, a training sample with several pattern points for each class is selected. After the training procedure is done, every pattern point of the image, including the training pattern points, is classified through the proposed pattern recognizer. |
