POTENTIAL USE OF MULTISPECTRAL AIRBORNE LIDAR DATA IN LAND COVER CLASSIFICATION

摘要

Multispectral LiDAR systems either in the form of separate multi-sensor or true multispectral sensors are currently emerging in the market. The multispectral LiDAR sensors operate at different wavelengths that allow recording a diversity of spectral reflectance from land features. This capability improves and extends the applications of LiDAR data. In this context, an improvement of land cover classification is presented from multispectral airborne LiDAR data acquired by the first commercial multispectral airborne LiDAR sensor, the Optech Titan. The sensor operates at three wavelengths; mid-infrared (MIR): 1550 nm, near-infrared (NIR): 1064 nm, and green: 532 nm. In this paper, three intensity images were first created from the collected point clouds at each wavelength, as well as the digital surface model (DSM). A maximum likelihood classifier was then applied to each intensity image separately, combined three-intensity images, and combined three-intensity images with DSM. An urban area, covering Oshawa, Ontario, Canada, was tested to classify the terrain into six classes namely; buildings, trees, roads, grass, soil, and wetland. The classification results were validated using 200 randomly selected reference points. The classification results showed that using single intensity image from wavelengths 1550 nm, 1064 nm, or 532 nm leads to overall classification accuracies of 34.0%, 48.5%, and 41.5%, respectively. The overall classification accuracy was improved to 65.5% using the combined three-intensity images. Moreover, the overall classification accuracy was increased to 72.5% by incorporating the height LiDAR data (i.e., DSM image). In further investigation, a radiometric correction model has been applied to the intensity data to improve the classification accuracy. The overall classification accuracy was improved to 69.0% from the combined three-intensity images and 78.0% by adding the DSM. The results achieved are promising and indicating that the use of multispectral LiDAR data can improve the land cover classification accuracies.