A ONE-AGAINST-ALL EXTRACTION OF COCOS NUCIFERA AT INDIVIDUAL TREE CROWN LEVEL VIA SUPPORT VECTOR MACHINE CLASSIFICATION USING LIDAR DATA

摘要

Philippines is the second largest producer of Cocos nucifera, also known as coconut, in the world, with an average production of 15 billion nuts per year corresponding to a hundred billion pesos. Being one of the major crops in the country, coconut accounts for 26% of the total agricultural land, corresponding to at least 3.5 million hectares. As significant declines in the production have been charted since 2016 due to climate-related incidents and infestations, it is high time that we introduce efficient and accurate data as inputs to resources management in the country. However, managing these much of coconut resources scattered on large geographic area is inefficient if we use data gathered through manual counting, and inaccurate of we resort to rough estimations. As the Philippine government embarks on the acquisition of LiDAR data achieving an equivalent 1 meter grid resolution, this study seeks to achieve classification of coconut trees at the individual tree crown level by performing Object-Based Image Analysis (OBIA) on a simple LiDAR-derived first-return highest-elevation model without the aid of spectral data. Support Vector Machine classification in a one-against-all approach has been implemented for the simplicity of the classification process. The methodology produces highly accurate tree count estimates on selected study sites in San Antonio, Quezon, reaching at least 90% on 16 study areas, without incorporating other remotely-sensed data and without using complex procedures. The outputs of this research suggests that agricultural resources mapping at individual tree level achieves high accuracies even when using LiDAR data alone. This study may also pioneer on "one agricultural class per classification" approach in the improvement of existing agricultural resources maps.