Hyperspectral Data Classification to Support the Radiometric Correction of Thermal Imagery

The derivation of surface temperature from thermal images requires a proper modelling of the spectral characteristics of the observed surfaces, in particular emissivity. Several possible approaches have been developed in literature. A first category of methods relies on the availability of multiple bands in the thermal region, while a second family of methods, which can be applied also with a single channel sensor, requires the derivation of emissivity values from ancillary data. The methodology, discussed in the present paper, involves the use of hyperspectral images acquired by an AISA Eagle 1 K sensor installed on board an aircraft platform. Data are composed of 61 bands in the visible and near-infrared region. A supervised classification approach was adopted to derive a map of the main materials appearing in the scene, with special attention to roofing materials. The presented analyses were performed in a portion of the urban area of Treviso (Italy), where two aerial surveys, one with a thermal sensor and the second with the AISA sensor, were carried out in 2011. All the presented activities were conducted in the framework of the European project “EnergyCity - Reducing energy consumption and CO2 emissions in cities across Central Europe”.