Application of Lipschitz Regularity and Multiscale Techniques for the Automatic Detection of Oil Spills in Synthetic Aperture Radar Imagery

Friday, 18 December 2015: 17:45
309 (Moscone South)
Olaniyi A Ajadi, University of Alaska Fairbanks, Fairbanks, AK, United States, Franz J Meyer, University of Alaska Fairbanks, Geophysical Institute, Fairbanks, AK, United States and Marivi Tello, German Aerospace Center DLR, Munich, Germany
This research presents a promising new method for the detection and tracking of oil spills from Synthetic Aperture Radar (SAR) data. The method presented here combines a number of advanced image processing techniques in order to overcome some common performance limitations of SAR-based oil spill detection. Principal among these limitations are: (1) the radar cross section of the ocean surface strongly depends on wind and wave activities and is therefore highly variable; (2) the radar cross section of oil covered waters is often indistinguishable from other dark ocean features such as low wind areas or oil lookalikes, leading to ambiguities in oil spill detection.

In this paper, we introduce two novel image analysis techniques to largely mitigate the aforementioned performance limitations, namely Lipschitz regularity (LR) and Wavelet transforms. We used LR, an image texture parameter akin to the slope of the local power spectrum, in our approach to mitigate these limitations. We show that the LR parameter is much less sensitive to variations of wind and waves than the original image amplitude, lending itself well for normalizing image content. Beyond its benefit for image normalization, we also show that the LR transform enhances the contrast between oil-covered and oil-free ocean surfaces and therefore improves overall spill detection performance. To calculate LR, the SAR images are decomposed using two-dimensional continuous wavelet transform (2D-CWT), which are furthermore transformed into Holder space to measure LR. Finally, we demonstrate that the implementation of wavelet transforms provide additional benefits related to the adaptive reduction of speckle noise.

We show how LR and CWT are integrated into our image analysis workflow for application to oil spill detection. To describe the performance of this approach under controlled conditions, we applied our method to simulated SAR data of wind driven oceans containing oil spills of various properties. We also show applications to several real life oil spill scenarios using a series of L-band ALOS PALSAR images and X-band TerraSAR-X images acquired during the Deep Water Horizon spill in the Gulf of Mexico in 2010. From our analysis, we concluded that the LR and CWT have distinct advantages in oil spill detection and lead to high performance spill mapping results.