Comparing Object-based and Pixel-based Classifications for Benthic Habitats Mapping in Pari Islands

• Published in: Journal of physics. Conference series 2018-11, Vol.1114 (1), p.12049
• Main Authors: Anggoro, A, Sumartono, E, Siregar, V P, Agus, S B, Purnama, D, Supriyono, Puspitosari, D A, Listyorini, T, Sulistyo, B, Parwito
• Format: Article
• Language: English
• Subjects: Accuracy; Classification; Ground truth; Habitats; Image resolution; Image segmentation; Islands; Mapping; Pixels; Satellite imagery; Support vector machines
• ISSN: 1742-6588; 1742-6596
• DOI: 10.1088/1742-6596/1114/1/012049

Object-based classification methods suitable for high resolution satellite imagery Worldview-2 provides a valid alternative to pixel-based methods. This study compares the results of an object-based to pixel-based classification for benthic habitats map in Pari islands of DKI Jakarta-Indonesia. The object-based approach involved segmentation of image data into object at multiple scale levels. The objects of the research were assign classes using training object to the support vector machines classification (object-based and pixel-based classification). Ground truth habitats for accuracy assessment using confusion matrices of both classification were undertaken based on 230 reference sites. An optimized segmentation was performed to get optimum classification result. The results showed, the overall accuracy on object-based classification was 75%, while the pixel-based classification produce the low overall accuracy was 61%. This research suggest that the object-based technique could be a promise approach for mapping benthic habitats, where the information obtained from this research was more accurate.