Classification of Spruce and Pine Trees Using Active Hyperspectral LiDAR

Résumé

Most forest inventories based on the use of remote-sensing data produce the required species-specific information by fusing data from different sources (e.g., Light Detection And Ranging (LiDAR) and spectral data). We tested an active hyperspectral LiDAR instrument in a laboratory measurement of spruce and pine trees to find out whether these species could be separated by means of combined range and reflectance measurements. An analysis focused on those pulses that had penetrated through the foliage improved the classification accuracies of the species with otherwise highly similar reflectance properties. Based on a careful selection of the classification features, 18 spruce and pine trees could be classified with accuracies of 78%-97% using independent training and validation data acquired by separate scans. The results denote the potential of using active hyperspectral measurements for species classification.