一种顾及多特征的建筑物变化检测方法

A Building Change Detection Method with Multi-feature

建筑物变化检测干扰因素太多,导致算法鲁棒性差,自动化程度较低。针对高分辨率卫星数据特点,提出了一种顾及多特征的建筑物变化检测方法,综合利用建筑物指数特征、光谱变化特征、阴影特征等解决高分辨率遥感影像的建筑物变化自动检测问题。该方法将变分建筑物指数的差异作为建筑物变化决策因子,多元变换(multivariate alteration detection,MAD)变量作为光谱变化因子,在D-S (Dempster-Shafer)证据理论框架下建立概率模型,采用像素级、特征级和对象级相结合的策略进行建筑物变化检测;同时将阴影作为辅助因子进行场景理解的变化信息后处理,进而去除一些非建筑物变化的高亮度区域。并利用两套ZY-3数据进行了相关实验,得到了较好的结果。

There are too many interference factors in building change detection, which leads to poor robustness and low degree of automation of algorithm. A building change detection method with multi-feature for high resolution satellite data is proposed in this paper. The method considers the building index feature, spectral change feature, shadow feature, et al. Difference of building index is used as the building change indicator and the MAD(multivariate alteration detection) variables are used as the spectral change index. The Dempster-Shafer fusion theory is adopted to combine two change indicators to get change probability map, and then combination of pixel-level, feature-level and object-level are used for the automatic building change detection. The shadow feature is also used to assist understanding the scene which can remove some high bright none-building areas. The experiments are conducted on ZY-3 satellite images and acquire good result. The experimental results demonstrate the effectiveness of the algorithm and the overall procedure.