Oil Slick Extraction And Coverage Area Calculation

In recent years, oil spill detection by satellite remote sensing has gained popularity and is of great interest for researchers. Synthetic Aperture Radar (SAR) offers a good means to detect oil slicks, since oil covered water appear as dark spots on radar images. The advantage of SAR over other remote sensing systems is that it can function under different weather conditions. The two problems in oil spill extraction are SAR images are inherently affected by multiplicative speckle noise, which is due to the coherent nature of scattering phenomena, and oil look-alikes on radar images which cause ambiguity. So, speckle reduction is an essential procedure for oil spill detection, object tracing and object recognition. Nonlinear thresholding technique on Wavelet coefficients gives significant better de-noising results than linear de-noising techniques. The challenge is to choose the reasonable threshold value for each subband of wavelet coefficients. In this dissertation paper, we present a novel approach of speckle noise removal using wavelet packet. The results of the proposed approach are compared with current state-of-the-art speckle filtering methods as well as discrete wavelet transform (DWT) based soft and hard thresholding methods. Comparisons are based on both visually and the peak signal to noise ratio. For the second problem, we concentrate on the use of automatic identification technique based on fuzzy classifier to discriminate between oil slicks and look-alikes. In this approach fuzzy classifier identifies the out put of the segmented images and segmentation is accomplished through smoothing and thresholding. Since smoothing removes noises and so is the details, the resultant blur image can represent the original image up to some degrees. So, it influences the later processes in identification process, e.g. the area of the object is one of the factors in the decision rules. In this work, the smoothing process is replaced with denoising via wavelet packet approach and very small objects are removed by area filtering method. The segmentation results via the proposed method are compared with those from smoothing. The proposed method gave more approximate images to the original ones and fuzzy classifier could give more precise decisions. Finally, the coverage area of the identified oil spill is calculated via region labeling method.