Feature-Level Based Hyperspectral and Lidar Data Fusion for Urban Analysis

Multisensor data fusion has recently become a relevant topic in remote sensing. Versatile hyperspectral sensors and lidar scanners not only separately deal with land cover classifications but promise to improve object and material classification synergistically. Our study presents a combined workflow for land cover and roof material classification based on the fusion of hyperspectral and lidar data. An innovative 3D ResU-Net incorporates continuous high-dimensional data based on 3D convolutional operations. Our results show that feature-level-based multisensor data fusion has high potential, especially in heterogeneous urban, dynamic, and highly complex environments.