The Study Of Marine Oil Spills Classification In SAR Images Based On Texture Analysis

Synthetic aperture radar is the most applicable spaceborne sensor for operational oil spill detection, mainly because of its all-weather, all-day detection capability, wide coverage and so on.Nowadays SAR images are extensively used for the determination of oil spills in the marine environment. It is often difficult to distinguish unambiguously mineral oil floating from real oil spills on the sea surface in radar images. Detecting oil spills from so many images is a very heavy job. This paper is based on the work of building a pattern recognition system for the oil spills images which are caught by SAR.This paper detects oil spills from one raw SAR image and researchs some critical problems including preprocessing image, choosing SAR image's characteristics and detecting algorithms. This paper using texture alalysis method based on GLCM to obtain the adaptable characteristics and window size.The adaptable characteristics are choosen from SAR images for classification. Author's tasks in this thesis here are building minimum distance classification model, maximum likelihood model and improved Back-Propagation Neural Network (BPNN) model in order to research the better method to detect oil spills. This paper is mainly to choose the ANN method and improve the BPNN method by using Matlab tools to model the classification result.This paper uses classification modle evalucation method. By evaluating, the improved BPNN is the best method whether from the whole precision or from Kappa coefficient. And also show that ANN has a brilliant feature in this field.