Multisensor fusion for the accurate classification of vegetation in complex ecosystems

Abstract

The use of geospatial tools to monitor natural ecosystems is a fundamental task to preserve the environment. In this context, remote sensing data can provide a valuable source of information to complement field observations, offering frequent and accurate imagery to support the mapping and monitoring of natural areas. The growing availability of hyperspectral (HS) data can provide a valuable solution but the spectral richness provided by hyperspectral sensors is usually at the expense of spatial resolution. To alleviate this inconvenience, instead of satellite platforms, airborne sensors can be considered. In this work, the accurate mapping of a complex shrubland ecosystem has been accomplished using multisensor imagery. Specifically, airborne CASI data (68 bands and 75 cm of pixel size) has been fused with an orthophoto (25 cm) to increase the spatial detail. A comprehensive analysis of 11 sharpening algorithms has been performed and, to improve the Support Vector Machine (SVM) classification accuracy, different input features have been considered. Excellent results have been achieved and the importance to improve the spatial resolution has been demonstrated.