Classification of sea-ice images using a dual-polarized radar

Abstract

The classification of the returns of a ship-borne like- and cross-polarized radar system into one of four categories, first-year ice, multilayer ice, icebergs, and shadows cast by icebergs is described. The data sets are digitized images obtained from a dual-polarized noncoherent Ku-band (16.5-GHz) radar used on the northern tip of Baffin Island, Canada. By using both the like- and cross-polarized radar inputs, classifier accuracy is improved compared to previous classifiers using only a single input. In particular, the use of both polarizations significantly improves the discrimination between icebergs and multilayer ice. In order to combine the like- and cross-polarized inputs, four classifiers are used: a one-dimensional classifier using the composite image formed by fusing the two polarization inputs with principal components analysis; a two-dimensional Gaussian classifier; and two neural network classifiers (the multilayer perceptron and the Kohonen feature map classifier). The results are compared to the classification based on a single like- or cross-polarized input.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>