Oil Spill Detection And Recognition Based On SAR Image

In recent years, marine environment is threatened by oil spills from transport vessels and offshore oil drilling platform. Some governments have built all-time marine environment surveillance systems to monitor possible oil spill accident. With the development of satellite remote sensing technologies, SAR(Synthetic Aperture Radar) with a variety of image processing techniques can detect and recognize large-scale oil film at sea surface accurately and timely. This article deeply studies the key problem of oil spill detection based on SAR image, including image filter, image segmentation, interested-of-area(IOA) extracting, oil film target recognition, reducing false alarm rate, etc. The principal results in this article are as follows:This article proposes an improved CV(Chan-Vese) model algorithm based on watershed algorithm for SAR image segmentation to extract IOA with oil spill from large field of view SAR image including land, sea and oil spilled sea area. First the Gamma MAP adaptive filter is apllied to suppress the specle noise of SAR image. Then the watershed algorithm is used to mark the homogeneous areas and makes the image only contains sea area with and without oil spill by filling processing. Last, the CV model is employed to segment the image to get candidate oil spill area. To avoid the CV algorithm failure by grey level of SAR image is not uniform, the grey scale smoothing is carrying out to improve the CV algorithm and the noise impact is reduced.An oil spill target detection algorithm in SAR image with the BOV(Bag of Word) model as key technique is proposed for the SAR image with large field of view and high complexity. First, the low level features of IOA in SAR image are extracted to construct the visiual dictionary by K-means clustering algorithm. Then a training sample library is constructed by the sea areas with or without oil spill targets, the low-level features in sample areas are extracted and matched with visual words in the visual dictionary. The frequency histogram of visual words of two classes is stated. Last, a classifier is designed to recogonize the oil spill areas from IOAs. The research and analysis of the application of BOV in SAR image interpretation can help to deeply understanding hige-level semantic of SAR image.Another oil spill detection algorithm based on image context information with Markov random field(MRF) is presented to use context informtions to improve the accuracy of image segmentation and target recogonition, and suppress false alarm rate. First, Markov statistical model is applied to describe the context probability relationship among different neighborhood systems including local pixels or pixel blocks. Then, the oil spill data statistics model of SAR image is constructed based on the equivalent principle of MRF and Gibbs distribution. Last, the best segmentation with maximum a posteriori(MAP) by iterative condition model(ICM) algorithm according to the local and global probability. The impact and improvement of initial label field for the MRF context model is discussed deeply. The initial label field is established by saliency map. The multi-scale SAR image with large view of field is analyzed by visual pyramid. The calculating of Gibbs distribution’s segmentation function is important. The energy function describes the relation between pixles in segmentation. The parameters of potential function of energy function are discussed in detail and verified by experiments. Simulation experiments show that the proposed algorithm provides more accuracy in segmentation and detection, improves the suppression of noise and false alarms, conserves the texture details of SAR image with high resolution and small scale.