Oil Spill Classification Based On SVM In Hyperspectral Remote Sensing Images

Marine oil spills have a serious pollution to the marine environment and may cause incalculable ecological disasters.For this reason,many countries pay great attention to the preventation and dealing with the oil spill accidents.Airborne and spaceborne remote sensing is one of the most useful tools to the tasks of oil spill detection,in which hyperspectral remote sensing is now becoming increasingly important nowadays.Faced with the urgent needs and the practical difficulties in hyperspectral oil spill monitoring,this thesis aims at improving the oil detection ability with hyperspectral remote sensing,following the trends of domestic and international research communities.This thesis focus on the oil spill detection methods with Support Vector Machine(SVM)model,which has important theoretical and practical values.The main works in this thesis are listed as follows:First,current studies on oil spill monitoring with hyperspectral remote sensing are reviewed in detail,including the preprocessing of hyperspectral imagery,the feature extraction of the spilled oil,and the SVM classifier.The aforementioned content gives a well foundation for further study of oil spill detection technology.Second,after the analysis of the oil detection mechanism in hyperspectral images,typical hyperspectral images located at Deep Water Horizon in Gulf of Mexico(denoted as DWH)are used to analyze the spectral similarities between spilled oils and natural seawaters in multi-resolution and multi-temporal images.A novel method is proposed to decrease the spectral difference between multi-temporal images,and based on which the spectral model could be designed to detect the spilled oil in hyperspectral images.Finally,based on the spectral characteristics of oil spills in hyperspectral images,the detection method of the oil on multi-temporal images with SVM classifier is proposed,which comprises the following components: kernel type selection,kernel parameter settings,and spectral bands selection.The proposed method is implemented with MATLAB programming language using the LibSVM package.Experiments were conducted on the AVIRIS hyperspectral Image to evaluate the oil detection performance of the proposed SVM method.