A new smoothness based strategy for semi-supervised atmospheric correction: Application to the léman-Baïkal campaign

The estimation of reflectance from hyperspectral data, a process called atmospheric correction, is a critical first step in hyperspectral image processing. Most of the current methods need additional measurements of ground reflectances or atmosphere. Without these measurements, unsupervised methods as Quick Atmospheric Correction (QUAC) can be applied but the observed scene has to be composed by several distinct materials. When studies on specific materials are conducted, no such large diversity occurs. To solve the atmospheric correction in that case, we propose a new atmospheric correction method, called Smoothing Technique for Empirical Atmospheric Correction (STEAC), using the smoothness property of the reflectance. This method is benchmarked against QUAC method on images acquired during the Leman-Baikal campaign. Results shows that this new method outstands QUAC method, both in terms of accuracy and stability with respect to the scene heterogeneity.