Leaf Wood Separation of TLS Point Cloud of Mangroves

Terrestrial Laser Scanner (TLS) captures 3D details of forests actively and accurately. Mangroves grow in hostile environments, making it difficult to obtain structural parameters of individual trees, using conventional survey methods which are crucial for the estimation of Above Ground Biomass. Researchers have attempted to find foliage filtering methods using radiometric or geometric features or a combination of both. When applied to mangroves, most of these methods produce erroneous results. Our study establishes a method to separate foliage from the trunk by finding the crucial features and the optimal spatial scale for deriving those features. It uses a combination of both radiometric and geometric features like eigenvalues, Omnivariance, anisotropy, etc. at 3 different spatial scales and classified using. Random Forest classifier. The method was repeated for sets of increasing and decreasing spatial scales and with sets of features based on importance. Our method gave an accuracy of 0.98 and a Cohen-Kappa score greater than 0.95.