Millimeter wave radar stationary-target classification using a high-order neural network

摘要

A type of artificial neural network is investigated which has been called a high-order network for application to the millimeter wave (MMW) radar stationary target classification problem. The high-order network, like the multilayer perceptron, provides a minimum mean square error (MMSE) estimate of the optimal discriminant; however, the high order network has the advantage of ease of training. This network can be trained via iterative gradient descent and also by a closed form one-pass solution. Using real beam Ka-band radar field data, the authors compare the classification performance of the high-order network with that of a Gaussian classifier for several conditions. The high-order network performance shows improvement over the Gaussian classifier, and very attractive results were obtained with the one-pass solution.