Neural networks for sequential discrimination of radar targets

Perceptron neural networks are applied to the problem of discriminating between two classes of radar returns. The perceptron neural networks are used as nonlinearities in two threshold sequential discriminators which act upon samples of the radar return. The neural network´s training phase eliminates the impractical task of estimating high-order probability density functions when designing a discriminator; consequently discriminators with memory are easily obtained. The discriminators using neural networks for their nonlinearities significantly outperform the optimal memoryless discriminators of Geraniotis (1989). The discriminators constructed with neural networks made no classification errors in 10000 trials from each hypothesis. These discriminators also used a significantly smaller expected number of samples to make their decisions than did known discriminators