海上溢油合成孔径雷达探测研究

发布时间：2018-10-09 09:57

【摘要】：在合成孔径雷达(SAR)图像中区分溢油和类油现象是溢油SAR探测的关键任务。实现该任务一般可分为3步:首先是提取油膜和类油膜的特征;然后筛选出有助于油膜和类油膜分类的关键特征;最后构造有效的分类器进行模式识别以便做出准确的判别。本文基于2011年蓬莱19-3油田溢油事故期间的15景SAR图像提取了138个油膜和类油膜样本的几何特征、背景特征、散射特征和纹理特征,将Fisher判别率和序列前向选择方法相结合,筛选出背景后向散射系数标准差、逆差距、能量和后向散射系数的均值四个关键特征组成的特征子集。在此基础上,为提高分类器的精度,将决策树模型CART算法与Bagging技术相结合,通过随机抽样给出多个维数相同大小的训练数据集从而建立多个决策树模型,以投票的方式对油膜和类油膜样本进行分类;最后,文中采用了五折和十折交叉验证方法对油膜和类油膜的分类结果进行评估,研究显示基于Bagging的决策树方法的油膜和类油膜分类的平均精度在85%以上,且将文中所用基于Bagging的CART决策树分类算法与经典CART决策树分类算法及神经网络分类算法相比较,发现本文所用方法的分类精度较高,从而表明了该方法在溢油SAR探测方面的可行性。
[Abstract]:The key task of oil spill SAR detection is to distinguish oil spill from oil spill in synthetic Aperture Radar (SAR) images. The task is generally divided into three steps: first, extracting the oil film and oil-like film characteristics; then screening out the key features that are helpful to the classification of oil film and oil-like film; finally, constructing an effective classifier for pattern recognition in order to make accurate discrimination. Based on 15 SAR images of oil spill accident in Penglai 19-3 oilfield in 2011, the geometric features, background features, scattering features and texture features of 138 oil film and oil-like film samples were extracted. The Fisher discriminant rate was combined with the method of sequence forward selection. The characteristic subsets of four key features, standard deviation of backscattering coefficient, inverse gap, energy and mean value of backscattering coefficient, are selected. On this basis, in order to improve the accuracy of the classifier, the decision tree model (CART) algorithm is combined with Bagging technology, and several decision tree models are established by random sampling of training data sets of the same dimension. Oil film and oil-like film samples were classified by voting. Finally, 50% discount and 10% cross-validation were used to evaluate the classification results of oil film and oil-like film. The research shows that the average accuracy of oil film and oil film classification based on Bagging decision tree method is more than 85%, and the CART decision tree classification algorithm based on Bagging is compared with classical CART decision tree classification algorithm and neural network classification algorithm. It is found that the classification accuracy of the method used in this paper is high, which shows the feasibility of this method in oil spill SAR detection.
【作者单位】：中国海洋大学信息科学与工程学院;
【基金】：海洋公益性科研专项(201505002) 国家自然科学基金项目(61501520)资助~~
【分类号】：TN957.52

【相似文献】