Object-based Analysis of Lidar Geometric Features for Vegetation Detection in Shaded Areas

The extraction of land cover information from remote sensing data is a complex process. Spectral information has been widely utilized in classifying remote sensing images. However, shadows limit the use of multispectral images because they result in loss of spectral radiometric information. In addition, true reflectance may be underestimated in shaded areas. In land cover classification, shaded areas are often left unclassified or simply assigned as a shadow class. Vegetation indices from remote sensing measurement are radiation-based measurements computed through spectral combination. They indicate vegetation properties and play an important role in remote sensing of forests. Airborne light detection and ranging (LiDAR) technology is an active remote sensing technique that produces a true orthophoto at a single wavelength. This study investigated three types of geometric lidar features where NDVI values fail to represent meaningful forest information. The three features include echo width, normalized eigenvalue, and standard deviation of the unit weight observation of the plane adjustment, and they can be derived from waveform data and discrete point clouds. Various feature combinations were carried out to evaluate the compensation of the three lidar features to vegetation detection in shaded areas. Echo width was found to outperform the other two features. Furthermore, surface characteristics estimated by echo width were similar to that by normalized eigenvalues. Compared to the combination of only NDVI and mean height difference, those including one of the three features had a positive effect on the detection of vegetation class. * Corresponding author