引入纹理特征的多光谱遥感影像海面油膜信息提取

Oil spill information extraction based on textural features and multispectral image

针对单纯依靠光谱特征油膜提取精度低、雷达影像油膜提取易受海况条件及假目标影响的问题,提出了一种结合光谱特征与纹理特征的多光谱遥感影像油膜信息提取方法。以2011年6月蓬莱19-3油田溢油事故为研究对象,选用HJ-1星CCD遥感数据,利用灰度共生矩阵获取影像纹理特征,采用SVM模型对结合纹理特征与光谱特征的影像进行分类,提取出研究区油膜信息,并将分类提取结果与仅依靠光谱特征的SVM模型分类结果进行了比较。结果表明:引入纹理特征的SVM模型分类总精度达到90.29%,比仅依靠光谱特征的分类精度提高了12.41%;纹理特征的参与降低了原影像噪声对分类结果的影响,油膜边缘提取更加清晰,油膜中心呈连续面状分布,引入纹理特征的SVM模型可有效地用于多光谱遥感影像海面油膜信息提取。

The existing methods of oil spill information extraction have many problems.For example,extraction only based on spectral characteristics is difficult to obtain high accuracy,and sea conditions and false targets have seriously influence on the study that depends on radar data.A model combined with textural features and spectral characteristics based on the support vector machine （SV M）classification was designed to extract oil spill information,using H J-1 optical satellite image of Penglai 19-3 oil spill accident in 2011 as study data.At first,textural features were calculated through gray-level cooccurrence matrix,then the model was used to classify and analyze oil spill information extraction accuracy by comparing it with single spectral characteristics classification.The total classification accuracy of the former method has risen to 90.29 ％,which was 12.41％ higher than the later.Therefore,using this method can reduce noise information and improve the precision of classification.In addition,the marginal area of oil spill looks more clearly and the central area distributes more coherent.The study result indicated that combined with textural features and spectral characteristics,this method was effective for the oil spills information extraction.