论文部分内容阅读
提出了一种组合中间层特征(Middle Level Feature,MLF)和支持向量机(Support Vector Machine,SVM)的全极化合成孔径雷达(Synthetic Aperture Radar,SAR)监督地物分类方法。选择监督方法的目的是直接区分实际地物类别,中间层特征在非监督聚类结果中获取,用于跨越底层特征与地物类别间的语义鸿沟。统计以某像素为中心的特征支持区域内各“中间成分”的占比作为该像素的MLF。这里“中间成分”对应于基于底层极化特征得到的非监督聚类类别。在覆盖武汉地区的Radarsat-2全极化数据上,与基于经典全极化特征的SVM监督分类方法进行了对比,研究了不同中间成分获取方法以及特征支持窗口对于分类性能的影响,结果显示:该方法有很好的性能并有进一步提升的空间。
A Synthetic Aperture Radar (SAR) supervised object classification method combining Middle Level Feature (MLF) and Support Vector Machine (SVM) is proposed. The purpose of choosing the monitoring method is to directly distinguish the actual feature categories, and the middle-level features are obtained from the unsupervised clustering results to cross the semantic gap between the underlying features and the feature categories. The proportion of each “intermediate component ” in the feature support area of a pixel-centered feature is counted as the MLF of the pixel. The “middle component” here corresponds to the unsupervised clustering category based on the underlying polarization features. The Radarsat-2 total polarization data in Wuhan area are compared with the SVM supervised classification method based on classical total polarization. The influence of different middle-component acquisition methods and feature support window on the classification performance is studied. The results show that: This method has good performance and room for further improvement.