Land Cover Classification Using Full-waveform LiDAR Data and Remotely Sensed Imagery

摘要

This study integrates geometric information derived from waveforms and spectral information derived from remotely sensed imagery, in order to improve the classification accuracy of surface features. Fundamental waveform observations, such as echo amplitude and pulse width are echo features of interest. In addition, a terrain feature, slope, was of interest in land cover mapping. In order to remove noisy feature information, mean filter was applied to all the features. A Decision Tree classifier was chosen to implement a classification procedure, with the NDVI feature calculated from a Worldview-2 image. A comparison was made between those LiDAR data with and without mean filters over echo analysis. It was found that the overall accuracy of the classification was improved by 7% with the mean filter; the Kappa value was improved by 5.92%. With a NDVI feature added, it is possible to further discriminate artificial and natural ground.