Study of pansharpening methods applied to hyperspectral images of sediment cores (Poster)

Spectroscopic and imaging sensors are very useful methods in analytical chemistry because they are fast, cost effective, very informative analysis. Recent works search to fused them to create a new sensor with different spectral range to increase spectral and thus chemical information to create robust and precise prediction models. Remote sensing already used fusion methods to increase spatial resolution for spectral sensors. In this paper, we propose to use pixel level data fusion methods on laboratory sensors to check their availability to increase spatial information with low effect on both dimensions (spectral and spatial). The methodology presents two steps, first the registration to fit spatially the sensors and then the fusion step to estimate each sensor at the optimal resolution. The proposed method was used on three sediment cores, that are living sample which can move, crack. They are imaged sequentially with two sensors that do not overlap spectrally: visible near infrared VNIR (400-1000 nm, pixel size: 60 mu m), short wave infrared SWIR (1000-2500 nm, pixel size: 190 mu m). The registration step allows to have a correlation above 0.9 with still spatial defect bring by the samples that cannot be removed. The twenty-one state of the art pixel level data fusion methods seems to be less relevant than a bicubic interpolation for the case of the laboratory hyperspectral images of sediment cores.