Detection and classification of buried dielectric anomalies using a separated aperture sensor and a neural network discriminator

The problem of detection and classification of buried dielectric anomalies using artificial neural networks was considered. Several methods for training and data representation were developed to study the trainability and generalization capabilities of the networks. The effect of the architectural variation of the network performance was also studied. The principal component method was used to reduce the volume of the data and also the dimension of the weight space. Simulation results on two types of targets were obtained which indicated superior detection and classification performance when compared with the conventional methods.<>