Improved change detection through post change classification: A case study using synthetic hyperspectral imagery

Abstract

Change detection is a well studied problem and well accepted taxonomies, although not formalized, exist in the literature to some degree. The basic taxonomy includes pre-processing, change detection and post processing. The final stage typically addresses the selection of appropriate thresholds, this work extends it to encompass classification in order to reduce false alarms. This effort leverages synthetic data generation capabilities to investigate the feasibility of the proposed postchange classification methodology to distinguish significant and insignificant change results produced from change detection analysis. Results demonstrate that post-change classification improves false alarm performance for a principal component analysis-based change detector by nearly 2-orders of magnitude for cases when high detection rates are required.