Extraction Oil Spill Information From Optical Remote Sensing Image Using Multiple Characteristics

The continuous development of the maritime transportation industry and theexploitation of offshore petroleum resources, leading to marine oil spills frequently.Oil spill effects on marine ecosystems and human severously, monitoring oil spillsituation and extracting the oil film information accurately have importantimplications to guide the clean-up work. Rapid large-scale monitoring, dataaccessibility, information-rich of optical remote sensing make itself to be have anirreplaceable position in the field of monitoring oil spill using remote sensing. And theoptical remote sensing is predicted to be the main mode of the future offshore oil spillmonitoring. The use of optical remote sensing monitoring now mainly using thespectral characteristics and ignoring the spatial characteristics of the image. Using thespatial characteristics is the hot point of current remote sensing applications, also theresearch breakthrough point of monitoring marine oil spill by satellite remote sensing.In this paper, we analyze multispectral data collected by the Chinese HJ-1satellite of the sea area in the Penglai19-3oil spill affected by the June2011oil spillaccident. The work and achievements are summarized as follows:(1) Exploring effect features in addition to optical features in the use of opticalremote sensing monitoring oil spill, and selecting multiple features for extracting oilspill information. After radiometric calibration, atmospheric correction, filtering andother preprocessing, extracting spectral characteristics, GLCM texture feature, LBPtexture feature, Sobel direction feature of image.(2) Using support vector machine remote sensing classification model to extractthe oil spills information. Enter a different combination of characteristics to supportvector machine model to extract the oil spill and calculate the recognition accuracy ofthe model using the test sample. The results showed that, the number of integratedfeature and the classification accuracy is not a positive correlation, it means the morenumber of integrated feature not necessarily lead to the higher recognition accuracy.Film recognition accuracy tends to stable when there have three features wereintegrated. Considering the overall classification accuracy, the film misclassification error, the film recognition accuracy and algorithm execution speed, we think thecombination of spectral feature and GLCM texture feature is the best method toidentify oil spill information using HJ-1satellite multispectral data.(3) Using Dalian Xingang oil spill accident occured in July16,2010verifies theapplicability and superiority of the proposed method. We get seven HJ-1satellitemultispectral images, using the support vector machine model which combinatespectral features and GLCM texture features to analyzing three oil spill images. Theresult shows that the method has applicability and superiority. According to extractedoil spill distribution information, we draw a Dalian Xingang Sea oil spill drift anddiffusion diagram, which presents a good effect about oil spill drift and diffusion.