Application of Support Vector Machine (SVM) and Quick Unbiased Efficient Statistical Tree (QUEST) Algorithms on Mangrove and Agricultural Resource Mapping using LiDAR Data Sets

摘要

Accurate mapping of mangrove and agricultural areas is necessary for effective planning and management of ecosystems and resources. While expert interpretation has been the typical method of classifying data sets, more efficient, objective, and faster methods of classification are required. This study applied the two classification techniques namely Support Vector Machine (SVM) and Quick Unbiased Efficient Statistical Tree (QUEST) algorithms for mapping mangrove and agricultural resources using LiDAR data. Ten LiDAR data sets were used for mangrove delineation. Each data set had a total of 90 ground-truth samples (30 per class) and 150 training points (50 per class) grouped into three classes: Mangroves, Other Vegetation and Non-Vegetation. Using Lastools software CHM, DSM, DTM, Intensity, Hillshade, Numret and Slope derivatives of the three LiDAR blocks were generated. eCognition software was used to perform classification of mangroves. A paired t-test was done to compare the accuracy of these two algorithms to determine which performed better in classifying mangroves. For agricultural resource mapping, LiDAR data sets for Tagum City and Panabo City were analysed. These areas contain large banana, coconut, and mango plantations. Statistical analyses showed that SVM performed better than QUEST in mangrove delineation. In agricultural resources mapping on the other hand, results showed that SVM and QUEST combined improved the general overall accuracy for Tagum and Panabo Cities to 97% and 96%, respectively. The agricultural land cover extracted could be used for a more accurate and effective resource management and monitoring of the cities' agricultural land. Both SVM and QUEST have a potential to improve the overall accuracy of LiDAR blocks in both mangrove and agricultural areas.