Hyperspectral imaging for lake sediment cores analysis

PosterThe aim of this presentation is to overview some applications of hyperspectral imaging for core sediment analysis in paleoenvironmental studies. Sampling methods (millimetre or centimetre) and routine analyses are destructive and non-spatially resolved methods that consume time and material. Hyperspectral Imaging is a way to have the advantages of spectroscopy (non-destructive, fast analysis) and of imaging (high resolution, information is spatially referenced). coupling hyperspectral imaging with data mining methods makes possible to study several proxies at micrometric scale in each area of the core.Two hyperspectral cameras are used, a Visible-Near InfraRed VNIR (spectral range: 400-1000nm, spatial resolution: 60μm) and a Short Wave InfraRed SWIR (spectral range: 1000-2500nm, spatial resolution: 189μm). The two datasets produced can be fused in a unique one used to model environmental proxies. This methodology was achieved on a core from the lake Le Bourget (Western Alps, 53cm long and 9cm width). Quantitative prediction models can be made with partial least squares regression PLSR. This method links spectra with a reference analysis by the creation of a regression model. Assuming a scale homogeneity, it can be spread to all the spectra of the hyperspectral image to predict high spatially resolved proxies. Total Organic Carbon and Grain Size class models have been developed with a validation determination coefficient of 0.86 for TOC and 0.85 for clay. Concentration maps are used to study variation inside each stratigraphic unit event at the scale of laminae.These datasets can be used for classification. Based on pattern recognition and artificial neural network, it is possible to classify the type of lithology defined by the user, for example: summer or winter lamina, floods with labelled areas of less than 1% of the image. For varved sediments, this method can be used to count the varve and apply statistics on them