In this research a novel approach for target recognition was developed that allows rapid detection of targets with known spectral signatures in hyperspectral images. The algorithm is based on a radial basis neural network (RBNN) configured for specific target signatures or a series of target spectral signatures. The main advantages of using radial basis neural networks for target detection of hyperspectral data are faster training and better selectivity compared to common backpropagation neural networks, significant reduction in overall image processing time, and greater accuracy. In addition, an adaptive band selection technique was used that was expected to improve the efficiency of the algorithm by reducing the dimensionality of the data. In the adaptive band selection technique, optimal spectral bands were selected using Mahalanobis distance and noise variance. It was determined that band selection based only on Mahalanobis distance did not improve the performance of the RBNN algorithm, and actually degraded the performance. |