Case-Based Sonogram Classification
David W. Aha, Patrick Harrison · 1994
This report replicates and extends results reported by Naval Air Warfare Center (NAWC) personnel on the automatic classification of sonar images. They used novel case-based reasoning systems in their empirical studies, but did not obtain comparative analyses using standard classification algorithms. Therefore, the quality of NAWC results were unknown. We replicated the NAWC studies and also tested several other classifiers (i.e., both case-based and otherwise) from the machine learning literature. These comparisons and their ramifications are detailed in this report. Next, we investigated Fala and Walker's two suggestions for future work (i.e., on combining their similarity functions and on an alternative case representation). Finally, we describe several ways to incorporate additional domain-specific knowledge when applying case-based classifiers to similar tasks.