Using distributed compressed sensing to derive continuous hyperspectral imaging from a wireless sensor network

Abstract

Hyperspectral imaging is a powerful and versatile tool to gather geo-information. Currently, main contenders for large-scale imaging campaigns are airplanes and satellites equipped with hyperspectral cameras. However, long-term, continuous monitoring of specific areas is hard to accomplish with these techniques: satellites only pass over areas rarely, airplanes require pilots, and all depend on weather conditions and suffer from regulatory overhead. Ground-based monitoring with a Wireless Sensor Network (WSN) is a promising addition which delivers data more continuously because it may be deployed permanently. But hyperspectral sensors are usually considered too expensive and energy-consuming to be used for a WSN. We suggest an alternative approach, using cheaper multispectral sensors and combining them to get hyperspectral-like spectral resolution while preserving the spatial resolution of the WSN. We evaluate our approach, comparing different algorithms for processing the data on datasets gathered in situ and via remote sensing.