Abstract

The segmentation of hyperspectral images (HSIs) is being used in many fields from target detection to classification. In this paper, we propose a new affinity matrix for the normalized cuts algorithms that takes into account both the hyperspectral and LiDAR data for segmentation. The affinity matrix uses both the spatial-spectral as well as the elevation information; and our results show that the segmentation is much more accurate and can distinguish objects better than a plain normalized-cuts algorithm. We show the improvement gained by adding the LiDAR data onto the hyperspectral data, and discuss the parameters selection strategies.

© 2016 Optical Society of America