Neural network based automatic target processing and recognition

This work is developed in the context of the SAPRI project-an integrated system for the automatic acquisition, manipulation and interpretation of radar images based on advanced neural networks, pattern recognition and other computational intelligence techniques. This is a joint work resulting from the cooperation between five academic institutions and the Naval Research Institute (IPqM) in Brazil the objectives of which are the definition and implementation of a five-module environment-image acquisition, pre-processing, recognition, tactical analysis and knowledge base of nautical documents-for automatic target processing and recognition. A set of results are described with respect to the solutions investigated in the pre-processing and recognition stages of the SAPRI project. Image processing techniques are investigated based on the Karhunen-Loeve and Walsh-Hadamard transforms for the extraction of statistical and geometrical features, respectively. Three different solutions are investigated for the automatic target recognition task, namely the multilayer perceptron (MLP) and radial basis function (RBF-DDA) based on neural network models and the k-nearest neighbour technique from statistical pattern recognition. Extensive experiments on real world data are used to test these two image processing techniques in conjunction with the three image classifiers