Share Email Print

Journal of Applied Remote Sensing

Kernel parameter variation-based selective ensemble support vector data description for oil spill detection on the ocean via hyperspectral imaging
Author(s): Faruk S. Uslu
Format Member Price Non-Member Price
PDF $20.00 $25.00

Paper Abstract

Oil spills on the ocean surface cause serious environmental, political, and economic problems. Therefore, these catastrophic threats to marine ecosystems require detection and monitoring. Hyperspectral sensors are powerful optical sensors used for oil spill detection with the help of detailed spectral information of materials. However, huge amounts of data in hyperspectral imaging (HSI) require fast and accurate computation methods for detection problems. Support vector data description (SVDD) is one of the most suitable methods for detection, especially for large data sets. Nevertheless, the selection of kernel parameters is one of the main problems in SVDD. This paper presents a method, inspired by ensemble learning, for improving performance of SVDD without tuning its kernel parameters. Additionally, a classifier selection technique is proposed to get more gain. The proposed approach also aims to solve the small sample size problem, which is very important for processing high-dimensional data in HSI. The algorithm is applied to two HSI data sets for detection problems. In the first HSI data set, various targets are detected; in the second HSI data set, oil spill detection in situ is realized. The experimental results demonstrate the feasibility and performance improvement of the proposed algorithm for oil spill detection problems.

Paper Details

Date Published: 13 April 2017
PDF: 11 pages
J. Appl. Rem. Sens. 11(3) 032404 doi: 10.1117/1.JRS.11.032404
Published in: Journal of Applied Remote Sensing Volume 11, Issue 3
Show Author Affiliations
Faruk S. Uslu, Yildiz Technical Univ. (Turkey)

© SPIE. Terms of Use
Back to Top