Object-Based Image Analysis for Coral Reef Benthic Habitat Mapping with Several Classification Algorithms

Object-based classification technique (OBIA) in Landsat image analysis is assumed to be able to increase classification accuracy result compared with a pixel-based classification technique. Most of the previous research was conducted using conventional (pixel-based) classification technique such as maximum likelihood algorithm. The objective of this research was to determine accuracy values of coral reef habitat classification based on OBIA algorithms such as Support Vector Machine (SVM), Random Tree, Decision Tree (DT), Bayesian, and k-Nearest Neighbour (KNN). The Landsat 8 OLI satellite imagery acquired on 17 October 2013 on coral reef area in Morotai island, North Maluku Province, Indonesia was analysed in this study. Field data was obtained in October 2012 using transect photo technique. The field data was then classified employing OBIA classification within 7 (seven) classes i.e., sand-rubble, rubble, coral, sand, sand-algae, seagrass, and rubble-sand. Results showed that the highest values of overall accuracy and Kappa on coral reef habitat mapping was produced by SVM algorithm which reached 73% and 0.64 followed by RT, KNN, Bayesian, and DT algorithms of 68% and 0.59, 67% and 0.56, 66% and 0.53, and 56% and 0.46, respectively. Those results showed that the classification methods based OBIA approach within > 6 classes produced better accuracy than the pixel based classification technique. The 1st International Symposium on LAPAN-IPB Satellite for Food Security and Environmental Monitoring. Procedia Environmental Sciences, Vol. 24. Hal. 222-227