基于对象BOW特征的高分辨率遥感影像变化检测方法

针对传统基于像素的变化检测方法的缺点,以及底层特征表现能力不足等问题,提出一种基于对象BOW特征的变化检测方法。首先,将经过预处理操作的两期影像进行波段组合得到组合后影像,再考虑地物光谱特征和几何空间信息对组合后影像进行多尺度分割,获得相对应的对象基元;同时,分别提取两幅影像的底层特征（包括影像各波段的均值和方差以及灰度图像的6种纹理特征）。其次,将对象视作文档,像素的特征向量视作单词,利用BOW模型构建影像对象的中层表达,即对象的BOW特征。最后,通过相似性度量算法比较相应对象的BOW特征,从而识别出影像上的变化区域。本文利用2组WorldView-2影像进行了检验,结果表明本文方法的变化检测结果较为完整,精度优于对比方法。本文方法基本能够满足变化检测的需求,为高分辨率遥感影像上的数据挖掘分析提供了有效的手段。

Change Detection Method for High Resolution Remote Sensing Images Based on BOW Features Representation

A novel change detection method is proposed based on BOW (Bag of Words) features of the image objects, aiming at the drawbacks of traditional change detection methods and the poor interpretation capacity of basic features. First of all, a simultaneous segmentation method which took both spectral features and geometric information into consideration was applied to two images that were previously preprocessed and layer-stacked to acquire corresponding image objects; then mathematical mean and variance of every image band, and six textures from its panchromatic image were extracted as basic features. Moreover, the BOW representation of each image object was constructed by regarding objects as documents and basic features of pixels as words. Finally, similarity measurement was used to compare BOW features to define changed and no-changed areas. In this paper, experimental results on two sets of WorldView-2 images showed that the proposed method demonstrated superiority in completeness of detection result and outperformed the methods for comparison in accuracy assessment. In all, the proposed method can basically meet the needs of change detection work and provide an effective way for data mining and analysis of high resolution remote sensing images.