An Improved Method for Oil Spill Detection

Abstract

Constrained energy minimization (CEM) extracts target signatures from hyperspectral images and multispectral images by suppressing the energy of background using a FIR filter. While the energy of background is described by the correlation matrix of the whole image, the target signatures could also be suppressed if they occupy a large proportion of the image. In this respect, we come up with an approach, which abandons possible target pixels based on spectral angle mapper (SAM) and then uses the left pixels to construct the correlation matrix. Based on this approach, we propose a new method named SAM-hCEM for target detection, which improves hierarchical CEM, a variation of CEM, and exhibits impressive performance for oil spill detection on MODIS datasets.