Deep neural network for oil spill detection using Sentinel-1 data: application to Egyptian coastal regions

Ahmed, Somia B.; ElGharbawi, Tamer; Salah, Mahmoud; El-Mewafi, Mahmoud

doi:10.1080/19475705.2022.2155998

Cited by 8 publications

(1 citation statement)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, an adversarial learning approach is implemented to forecast the wind field correction and oil-spill drift detection [28]. Later several machine learning algorithms were utilized for detecting the oil spill using Sentinel-1 data [29][30][31][32].…”

Section: Introductionmentioning

confidence: 99%

Clustering of Marine Oil-Spill Extent Using Sentinel-1 Dual-Polarimetric Scattering Spectrum

Paul,

Dey,

Marino

et al. 2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Oil spills pose a significant threat to the maritime ecosystem. Identifying an oil spill is vital to assess its spread and drift to nearby coastal areas. Synthetic Aperture Radar (SAR) sensors are viable for mapping and monitoring marine oil spills. This study proposes a new technique that utilizes the dual-polarimetric Sentinel-1 SAR data. The method is based on projecting the 2 × 2 covariance matrix onto distinct random realizations of the normalized scattering configuration. We then obtain the dual-polarimetric spectrum of the scattering-type parameter, θDP. The θDP spectrum is then used in the unsupervised K-means clustering technique to segment oil spills from the rest. The cluster findings are then compared to the accuracies obtained using the standard scattering-type parameters from the eigen-decomposition approach (VV, VH) intensities and Otsu thresholding of [H + α + A] parameter. We demonstrate the proposed approach by clustering marine oil-spill extent over parts of India, Kuwait, the United Arab Emirates, and the Mediterranean Sea obtained by Sentinel-1 SAR images. We observed that the clustering accuracy of the proposed technique outperforms the ones obtained from the channel (i.e., VV and VH) intensities, Otsu thresholding of [H + α + A] parameter, and the eigen-decomposition based method. The proposed approach improves the overall accuracy by ≈ 8 % and ≈ 20 % respectively over different study areas.

show abstract