Title: Mauritius oil spill detection using transfer learning approach for oil spill mapping and wind impact analysis using Sentinel-1 data
Authors: Koushik Das; Prashanth Janardhan; Harish Narayana
Addresses: National Institute of Technology Silchar, Cachar, Assam, 788010, India ' National Institute of Technology Silchar, Cachar, Assam, 788010, India ' Visual and Transparent Infra Private Limited, Mysore, Karnataka, India
Abstract: The oil spill detection and mapping using Sentinel-1 (S-1) data for the Mauritius oil spill event have been done in this study. The convolutional neural network (CNN)-based on pre-trained models such as AlexNet, VGG-16, and VGG-19, have been used to classify the S-1 images by the transfer learning approach. The S-1 images are classified into two classes: with and without the oil spill. Then, the oil spill detection was done in the sentinel application platform (SNAP), and the oil spill mapping was done in ArcGIS. The VGG-16 network performs the best among the other pre-trained networks with an accuracy of 96.88%, precision of 95.92%, and recall of 97.92%. The impact of wind on the spreading of oil is also analysed using remote sensing and GIS techniques. It has been observed that the spreading of oil doesn't only depend on sea wind but also other environmental factors.
Keywords: transfer learning; convolutional neural network; CNN; Sentinel-1; sentinel application platform; SNAP; oil spill; image classification; remote sensing; GIS.
DOI: 10.1504/IJHST.2024.142019
International Journal of Hydrology Science and Technology, 2024 Vol.18 No.4, pp.421 - 444
Received: 07 Feb 2023
Accepted: 03 Sep 2023
Published online: 07 Oct 2024 *