LWIR and MWIR images dimension reduction and anomaly detection with locally linear embedding

Abstract

In recent years, hyperspectral imaging has been widely used in remote sensing. As opposed to many other imaging instruments, hyperspectral receivers can measure the radiation coming from the earth in very narrow and frequent intervals. In this study, long-wave infrared (LWIR) and mid-wave infrared (MWIR) hyperspectral images were used to determine the effect of dimensionality reduction on anomaly detection. On eleven MWIR and seven LWIR images of various noise levels, two dimensionality reduction methods, namely the local linear embedding (LLE) and principal component analysis (PCA) were compared. After dimension reduction, dual window Reed-Xialoi (DWRX) algorithm was used for anomaly detection. On several images, it was observed that locally linear embedding gives better results when compared to principle component analysis, especially on hyperspectral images with higher noise levels.