Parameter analysis of texture feature in oil spill detection based on SAR

Oil spill in ocean has been one of the primary factors which lead to deteriorations of marine ecological balance. The analysis of marine oil spill detection has become an important subject of the marine environmental protection. However, traditional methods to extract oil spill merely base on the spectral information of optical image or backscattering coefficient of SAR image respectively, which may lead to misclassification of different objects with same spectrum or similar roughness. Hence, texture information is combineal with the traditional image information to improve the extraction accuracy of oil spill. In the process of texture analysis, there are many parameters which will directly affect the extraction accuracy. So it is important to select appropriate parameters. In this paper, we choose three SAR images in the same orbit which covered the Bohai Sea area in 2006 as data resource, and use method based on gray level co-occurrence matrix (GLCM) to analyze texture feature. Because GLCM-based texture analysis method can perceive the surface of image well and describe the texture feature in detail by gray correlation of pixels, it is more suitable for marine oil spill detection in SAR images. Then we discuss, experiment, select and verify the parameters of texture analysis. Finally, this paper selected four parameters include local stationary, non-similarity, contrast and change as the texture feature statistics, determined the value of these parameters, used neural network classification considered both texture feature and backscattering coefficient of SAR. With the classification accuracy up to 80.65%, the method combined the traditional information with the texture information to extract oil spill turn out to be feasible and effective and also laid a good foundation for the future study on marine oil spill detection.