Anomaly detection for hyperspectral imagery based on vertex component analysis

While detecting low probability anomalies with Low Probability Detection (LPD), the high miss rate was discovered since the anomaly detection was effected badly by noise resulting from selecting orthogonal eigenvectors as background spectrum in the low-frequent space. This paper presented an approach that the background was replaced with endmembers obtained with Vertex Component Analysis (VCA) and spectral vectors observed were projected onto subspace that is orthogonal to the background subspace, which suppressed the background effectively and made anomalies obvious. The experiment on the AVIRIS data shows that the miss rate is reduced obviously and the detection capability is enhanced, moreover, the miss rate is decreased 30 percent of LPD by simulation on part of the AVIRIS data.