摘要
利用遥感数据进行城市变化检测时,由于城市是包含建筑物、道路、绿地以及水体等多种地物类型的综合体,同时与农村居民地等地物类型具有相似的光谱特征,因此,传统的单纯基于光谱信息的变化检测方法很难取得理想的效果。将空间信息加入到变化检测中,可提高变化检测的精度,但常用的加入纹理的方法容易产生边缘效应。本文提出一种基于图像分割的变化检测方法,在该方法中使用一种基于组分的多尺度形态学梯度,具有对噪声不敏感、边缘不会变厚等优点;同时与现有方法进行比较。实验结果表明,该方法在加入图像空间信息的同时避免了加入纹理等空间信息所产生的边缘效应,能够有效地提高城市变化检测的精度。
Urban change detection using spectral information alone proves to be unsatisfactory owing to the high spectral variability of the urban area and the similarity between the urban area and the rural area. The addition of the spatial information would improve the accuracy of change detection for urban area. In this paper, a method based on image segmentation was proposed and evaluated for urban change detection with post-classification com- parison technique. In this method, a component-based multi-scale multi-spectral gradient algorithm was applied in image segmentation, which was insensitive to noise and could extract the various fineness of the edges; and the combination of segmentation result and pixel-based classification was used to get a better classification result for post-classification comparison. Furthermore, it was also compared with other two methods： one with spectral data alone and the other with the addition of texture features. The results show that the method based on image segmentation provides a significant improvement in overall accuracy and Kappa coefficient of urban change detection because it can not only incorporate the spatial information but also avoid the edge effect with the addition of texture.
出处
《地球信息科学》
CSCD
2008年第1期121-127,共7页
Geo-information Science
关键词
城市变化检测
分类后比较法
图像分割
urban change detection
post-classification comparison
image segmentation
