Research On Marine Oil Spill Detection Method Based On Hyperspectral Remote Sensing

Marine oil spills are one of the most serious problems of marine pollution,causing serious harm to the marine environment and the production and living of nearby residents.Once an oil spill occurs,the most urgent task is to quickly monitor the oil spill,grasp the oil spill dynamics in time,and prepare for emergency rescue and decision-making.Among the many marine remote sensing technology monitoring methods,hyperspectral remote sensing has become an effective marine oil spill monitoring tool due to its the integration of "space and spectrum",high spectral resolution,and wide spectral information range.In this paper,starting from two aspects: the traditional method based on spectral index and the deep learning method based on Convolutional Neural Network(CNN),the research of hyperspectral remote sensing marine oil spill detection method is carried out.The main work is as follows:1.Extraction and analysis of spectral index of hyperspectral oil spill image.At present,the spectral index algorithm is an important hyperspectral oil spill detection method.This article summarizes 13 commonly used spectral indices.Due to the limitation of the data band range,11 spectral indices are extracted from the feature image,probability density function(PDF)curve and J-M distance(Jeffreys-Matusita distance)analyzes the ability of different indices to distinguish thick oil film,thin oil film and seawater from three aspects.The research results show that the traditional spectral index algorithm has great limitations in oil spill detection,and it is difficult to accurately distinguish between thick oil film,thin oil film and seawater.2.Carried out the research of hyperspectral oil spill detection algorithm based on decisionlevel fusion of convolutional neural network.In order to carry out decision-level fusion,two hyperspectral oil spill detection algorithms based on convolutional neural networks are studied.They are oil spill detection algorithms based on Linear Discriminant Analysis(LDA)CNN and multiscale input(MSI)CNN.After confirming the effectiveness of the above two algorithms for oil spill detection,the results of the two algorithms are fused at a decision level.The experimental results show that this method can obtain higher oil spill detection results than a single classifier,and has good practical value.3.A hyperspectral oil spill detection algorithm based on spectral-spatial features integrated network is proposed.Specifically,1-D and 2-D CNN models have been employed for the extraction of the spectral and spatial features,respectively.During the stage of spatial feature extraction,three consecutive convolution layers are concatenated to achieve the fusion of spatial features.Next,the extracted spectral and spatial features are concatenated and fed to the fully connected layer so as to obtain the joint spectral-spatial features.In addition,L2 regularization is applied to the convolution layer to prevent overfitting,and dropout operation is employed to the full connection layer to improve the network performance.The experiment was firstly conducted on the hyperspectral public data set Pavia University to verify the effectiveness of the algorithm.Eventually,the experimental results upon oil spill datasets demonstrate the strong capacity of oil spill detection by this method,which can effectively distinguish thick oil film,thin oil film and seawater.