Cascade SVM Based Oil Detection in SAR Images

Synthetic aperture radar (SAR) is well adapted to detect ocean pollution independently from daily or weather conditions. Oil slicks have a specific impact on ocean wave spectra because the presence of oil slicks can induce a damping of the backscattering to the sensor and a damping of the energy of wave spectra. Thus oil slicks can be discernible from the radar image. Several algorithms are applied for local segmentation of oil slicks but most solutions are tailored for specific applications. This paper describes a multi-scale kernel-based fusion approach by using textural and statistical features. This cascade svm approach reduces the problems of speckle and sea clutter and preserves subtle variations of oil slicks. The experimental results carried out on SAR images prove the effectiveness of proposed approach.