Neural approaches to ship target recognition
Michael Raymond Inggs, Andrew Robinson · 2002
This paper summarizes current research into the applications of neural networks for radar ship target recognition. Three very different neural architectures are investigated and compared, namely; the feedforward network with backpropagation, Kohonen's (1990) supervised learning vector quantization network, and Simpson's (see IEEE Trans on Neural Networks, vol.3, no.5, p.776-787, 1992) fuzzy min-max neural network. In all cases, preprocessing in the form of the Fourier-modified discrete Mellin transform is used as a means of extracting feature vectors which are insensitive to the aspect angle of the radar. Classification tests are based on both simulated and real data. Classification accuracies of up to 93% are reported.