中文摘要：

针对高分辨率遥感影像中道路交叉口难以用固定几何和光谱特征描述的问题,该文提出了一种基于像元结构指数的交叉口提取方法。采用自适应光谱异质性阈值来提取像元形状特征,较好地适应了不同交叉口之间的场景特征差异;构建了像元形状与交叉口结构的量化映射关系,赋予特征语义信息,从而保证了提取结果的合理性。基于城区复杂场景高分遥感影像的交叉口提取实验表明：所提方法能够准确定位交叉口中心点,并检测道路交叉结构;另外,该方法对于交叉口支路发生变化的情况也具有适用性,定性与定量分析表明了该文方法的有效性。

英文摘要：

Aiming at the problem that it is hard to describe road intersection with specify geometry pattern and spectral characteristics, this paper proposed a method based on intersection pixel structure in- dex. The adaptive spectral heterogeneity threshold was introduced to extract pixel shape feature which was better adapted to the characteristic differences between the intersections of different scenes. Semantics fea ture was constructed by mapping between pixel shape and intersection structure ensuring validity of extrac tion. Experiment was conducted on high resolution satellite image which contained various intersec tions. Results showed that locations of intersection center points and intersections＇ structure were well ex- tracted. This method is also applicable for extraction task of changed intersections. The qualitative and quantitative analysis showed effectiveness of this method.