A Multivariate Approach to Iceberg and Ship Classification in HH/HV ASAR Data

Abstract

Spaceborne synthetic aperture radar (SAR), with its wide area coverage and all-weather operation, is an ideal sensor to provide iceberg surveillance in support of safe shipping and offshore operations. Reliable ship/iceberg discrimination in SAR imagery is at least as important as detection since misclassification can result in expending significant resources for investigation or avoidance. To address this need, the authors have undertaken research to facilitate effective target discrimination using SAR multi-polarization data. The results presented here are for iceberg and ship classification for ENVISAT advanced synthetic aperture radar (ASAR) HH/HV data. Target classification was achieved by maximizing the a posteriori probabilities obtained from Bayes's rule. The maximum likelihood Gaussian classifier was used to model the probability of an unknown target belonging to either the iceberg or ship class. The feature selection algorithms, sequential forward selection (SFS), genetic algorithm (GA), and exhaustive search (ES) were evaluated for optimization of a feature space dependant multivariate classifier. The results from this study show for dual polarized HH/HV ASAR, icebergs and ships can be classified with a 93.5% accuracy using a two-class maximum likelihood model. As well, for the small sample set of 201 iceberg and ship targets presented here, suboptimal feature selection algorithms such as the SFS, GA, and exhaustive ranked search (ERS) are considered. These feature selection methods were considerably less computationally expensive to run than the global exhaustive search and were found to have converging results for both accuracy and features selected.