Oil spill identification based on DAM UNet model using MATLAB

This research focuses on oil spill identification leveraging a DAM UNet model implemented in MATLAB. The process initiates with comprehensive pre-processing steps, including the extraction of amplitude VV polarization, radiometric correction, and the application of speckle filtering using a Lee sigma filter with a 7x7 window. Further refinement involves terrain correction to enhance the accuracy of the dataset. The culmination of these steps results in a meticulously processed oil-detected image. Subsequently, the UNet semantic segmentation algorithm is employed for effective segmentation of the image, optimizing the identification of oil spill regions. The DAM UNet model incorporates a denoising algorithm, specifically the Wiener2 filter, to further enhance the clarity of the segmented image. Finally, the Peak Signal-to-Noise Ratio (PSNR) value is calculated, providing a quantitative measure of the fidelity between the original and denoised images. This holistic approach, encompassing pre-processing, semantic segmentation, denoising, and quantitative evaluation, contributes to the robustness and accuracy of oil spill identification in remote sensing imagery.

Keywords: Oil spills Images, deep learning, image processing, UNet and Sematic Segmentation, Convolutional neural network, Denoise Technique and PSNR.