University of Twente Proceedings
Mangrove classification using support vector machines and random forest algorithm: a comparative study
Campomanes, F. and Pada, A.V. and Silapan, J. (2016) Mangrove classification using support vector machines and random forest algorithm: a comparative study. In: GEOBIA 2016 : Solutions and Synergies., 14 September 2016 - 16 September 2016, University of Twente Faculty of Geo-Information and Earth Observation (ITC) .
|
PDF
1MB |
| Event: | GEOBIA 2016 : Solutions and Synergies., 14 September 2016 - 16 September 2016, University of Twente Faculty of Geo-Information and Earth Observation (ITC) |
| Abstract: | Mangrove forest ecosystems fulfil a number of important functions like supporting the conservation of biological diversity by providing habitats, nurseries, and nutrients for animal species. In the Philippines, mangrove forests are declining due to the growth of aquaculture production. Mangrove forests are slowly being replaced by fishponds. An accurate inventory of what are left of these natural resources is important to know how we can conserve and manage them. This study aims to compare the performance of support vector machines (SVM) with random forest (RF) algorithm in automatically classifying mangrove forests using LiDAR data and orthophotographs in an object based approach. The site is a 36 sq. km. coastal area in Manapla, Negros Occidental, Philippines. Various derivatives were created from the LiDAR data like the pit-free canopy height model (CHM) and intensity. The CHM was used in contrast split segmentation to distinguish between ground and non-ground objects. Only the non-ground objects were segmented further knowing that majority of mangroves are tall. Inventory of short mangroves is not yet included in this study. The non-ground objects were further segmented using multiresolution segmentation with the CHM and RGB bands of the orthophoto using a scale of 15. The non-ground class was further separated into four classes namely: mangrove, built-up, other trees, and sugarcane. 120 training points and 30 validation points per class were collected by visual inspection using the orthophoto as reference. Several features of the training objects were computed from both the LiDAR and orthophoto derivatives and used for classification. SVM with radial basis function was used to classify the rest of the image and resulted in an overall accuracy of 95.83%. For mangroves, its precision and recall reached 83.33% and 100%, respectively. In the SVM classification, mangroves were confused with other trees and sugarcane. Another machine learning algorithm, random forest (RF) was used to classify the same area to compare their performance and accuracy. Using the same features, the RF classification achieved an overall accuracy of 99.1667%. For mangroves, the RF classification obtained 100% and 96.70% for its precision and recall, respectively. The RF classifier confused other trees with mangroves which caused the error. The accuracies from both machine learning algorithms show that the RF classifier performed better than the SVM classifier and further implies the potential of using RF in classifying mangroves in other areas. |
| Item Type: | Conference or Workshop Item (Paper) |
| Link to this item: | https://doi.org/10.3990/2.385 |
| Conference URL: | https://www.geobia2016.com/ |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page