BENTHIC HABITAT CLASSIFICATION AND MAPPING USING SUPPORT VECTOR MACHINE ALGORITHM IN HINATUAN, SURIGAO DEL SUR, PHILIPPINES Michelle V. Japitana and James Earl D. Cubillas Phil-LiDAR 2.B.14 Project, College of Engineering and Information Technology, Caraga State University, Butuan City, Philippines, Email: michelle.japitana@gmail.com Email: jamesearl_cubillas@yahoo.com KEY WORDS: benthic habitat mapping, image processing, Support Vector Machine ABSTRACT: This study demonstrates the application of classification techniques using Support Vector Machine (SVM) for benthic habitat mapping. The orthophotos of the coastal area of Hinatuan, Surigao Del Sur, which undergone quality checking, was used for this study. Optimization procedure was performed in matrix laboratory (MATLAB) software with parallel computing to help hasten the process due to its enormous size. The study area is composed of four datasets, namely: (a) Blk66L005, (b) Blk66L021, (c) Blk66L024, and (d) Blk66L0114. The image used for collecting samples for SVM procedure was Blk66L0114 in which a total of 134,516 sample objects of mangrove, possible coral existence with rocks, sand, sea, fish pens and sea grasses were collected and processed. The collected samples were then used as training sets for the supervised learning algorithm and for the creation of class definitions. The learned hyperplanes separating one class from another in the multi-dimensional feature space can be thought of as a super feature which will then be used in developing the C (classifier) rule set in eCognition software. The feature used for SVM algorithm are the following: CIE L*a*b*, RGB Intensity, and One Dimensional Scalar Constancy. The classification results of the sampling site yielded an accuracy of 98.85% which confirms the reliability of remote sensing techniques and analyses employed to orthophotos like the color transformation and illumination correlation and the use of SVM classification algorithm in mapping benthic habitats. 1. INTRODUCTION Benthic habitat mapping is an important tool for several studies. It is interesting from a geographical perspective to study the trends of local landscape changes, anthropogenic disturbance on benthic organisms, and climate changes. Such Benthic maps are also important for the economy, providing marine assessments for coastal management and ecological analysis. Various area utilizations can be efficiently planned by having prior knowledge about certain habitats and their changing tendencies especially that coastal areas represent a very dynamic case regarding their locations. The need for high resolution maps in the management of tropical environments is increasing and emphasized by the rapid anthropogenic development often occurring in coastal zones (S. Chauvaud, C. Bouchon & R. Maniere, 1998). Remote sensing plays a predominant role in coastal management and monitoring especially in mapping marine resources such as sea grasses, coral reefs, and mangroves. Advances in sensor design and data analysis techniques are making remote sensing systems practical and attractive for use in research and management of coastal ecosystems, such as wetlands, estuaries, and coral reefs (Klemas, 2011). Also, the current procedures and techniques in classification algorithms has now presented an opportunity to improve traditional methodologies in classifying benthic habitats. According to the Directory of Remote Sensing Applications for Coral Reef Management (2010), there are certain requirements for data archiving and imagery used for creating benthic habitat maps. Shown in Table 1 are the requirements for each class of benthic habitat. It shows the requirements for specific benthic habitats which concerns certain limitations using those data sources. Using prescribed datasets, it is often necessary to download data files and analyze locally, a difficult and time-consuming process if data with different formats and resolutions were used. With this, using orthophotos with resolution of 0.5m might aid in generating benthic habitats. Image analysis of remotely sensed data is the science behind extracting information from the pixel within a scene or an image. Multispectral sensors for remote sensing are designed to capture the reflected energy from various objects on the ground in the visible and the infrared wavelengths of the electromagnetic (EM) spectrum of the sun. Some of the sensor ranges extend all the way into the thermal spectral range, whereas most of the commercial sensors today primarily capture data in the visible and near-infrared regions of the EM spectrum. (Navulur, 2009).Traditional image analysis techniques are pixel-based techniques and explore the spectral difference of various features to extract the thematic information of the interest to the end user. There are instances where objects that are the size of a pixel need to be identified; typical applications involve feature extractions comprised of multiple pixels such as roads, buildings, crops, and others.