Usability of a Sunlit - Shaded Area Separation in Individual Tree Species Classification

Individual tree species classification can be done using Light Detection and Ranging (LIDAR) and aerial multichannel images simultaneously. LIDAR is typically used to derive shape features and aerial images are employed to extract colour features from a dataset. Different lighting conditions are a common problem when images are utilised: the position of the sun has a significant impact on the backscatter radiation of the trees. Methods and results that work on a set of data can be ineffective on another dataset if the method is not robust in changing lighting conditions. The current solutions to shading induced problems are spectral rationing and the use of a subset of well illuminated tree pixels. We present a method, Illumination Dependent Colour Channels (IDCC), for separating the sunlit and shadowed side of a tree sample in this paper. We test if the features derived from this separation can be used effectively in individual tree species classification. Pixel values of the digital aerial image are mapped to Canopy Height Model (CHM) derived 3D representation of the trees. The sun position together with the mapping aircraft position are then utilised to determine the sunlit and shadowed parts on the tree model surface. The classification results are computed on a separate test set and compared with a reference method. Improved overall classification accuracy of 11.0 percentage points were achieved with the used dataset, when results were compared with the ones acquired from two separate reference methods. * Correspondence author.