Application of Nonlinear Feature Normalization on Combined Hyperspectral and Lidar Data

Mitigating nonlinear effects, e.g., due to shadows, variations in illumination conditions, and angular dependencies of spectral signatures is an important topic in hyperspectral remote sensing. In this paper, we apply the Nonlinear Feature Normalization on a combined data set consisting of 128 spectral bands and a weighted digital elevation model. The NFN transforms the data set to a new linear basis and by that mitigates nonlinearities. Evaluation is done by applying the Spectral Angle Mapper to the original and the NFN-transformed data. Different parameter combinations are tested to find the best classification results. Additionally, a Random Forest approach is calculated to compare the results.