一种正样本单分类框架下的高分辨率遥感影像建筑物变化检测算法

A new building change detection algorithm of high-resolution remote sensing image based on one-class classifier framework

高分辨率遥感影像建筑物变化检测对于城市规划、城市执法等方面具有非常重要的意义。文中提出一种基于正样本单分类器学习框架下的建筑物变化检测算法。首先提取影像形态学建筑物指数特征(MBI),通过卡方变换将其与光谱特征进行融合;然后利用一种单分类器完成建筑物变化初始判定;最后利用改进的长宽比形状特征完成最终建筑物变化判定。通过与现有建筑物变化检测算法、传统的多分类器算法对比,本文算法精度有一定提升并具有更强的鲁棒性。

Building change detection algorithm of high-resolution remote sensing image is very useful on urban planning, urban law enforcement and so on. This paper proposes a building change detection algorithm based on a positive sample one-class classifier learning framework. Firstly, this paper extracts the feature index (MBI) of image morphological structure, which is combined with the spectral feature by chi-square transformation. Then, a single classifier is used to complete the initial decision of building change. Finally, the improved aspect ratio shape characteristics is used to complete determining the final building changes. Compared with the existing building change detection algorithm and the traditional multi-classifier algorithm, the accuracy of the algorithm is improved,which has stronger robustness.