摘要:
后验概率变化矢量分析(change vector analysis in posterior probability space,CVAPS)方法没有顾及到遥感影像波段之间和多时相之间的光谱相关性,可能会造成信息丢失而降低影像变化检测的精度.因此,结合多元变化检测(multivariate change detection,MAD)技术与 CVAPS方法,提出一种改进的土地利用/覆盖变化(land use/cover change,LUCC)分类自动更新方法.首先,引入MAD技术来降低多光谱影像波段间相关性的影响,从而改善对像元变化检测的精度,增强LUCC分类自动更新过程中训练样本的可靠性,提高LUCC分类自动更新的精度;然后,为减少分类图中"椒盐"噪声的影响,进一步利用迭代马尔科夫随机场(iterative Markov random field,IR-MRF)模型进行分类后空间邻域处理,以提高自动更新的精度.以福建省长汀县2013年获取的Landsat8影像数据以及相应的LUCC分类图为基准,利用2003年获取的Landsat5影像,对长汀县2003年的LUCC进行更新.实验结果表明,该方法的自动更新总体精度能够达到80%,比单独采用CVAPS方法的自动更新精度提高了约3%.
Abstract:
The method of change vector analysis in posterior probability space(CVAPS) does not take into consideration the correlation between the bands of remote sensing image, which may result in unreliable change detection. In view of such a situation,the authors introduced multivariate change detection(MAD)method and,in combination with CVAPS,proposed an improved method for automatic updating of land use / cover change(LUCC) classification. The method firstly introduces MAD to reduce bands-correlation for improving the reliability of train-samples and accordingly improving LUCC updating maps,and then included an iterative Markov random field(IR-MRF)model to fully employ the contextual information in post-processing to reduce the noise of"salt -and -pepper". Choosing Changting County of Fujian Province as the study area, the authors used Landsat5 TM and Landsat8 OLI data acquired in 2003 and 2013 respectively, and took OLI as the base image to update the classification map in 2003. The experimental results show that the proposed method significantly outperforms the CVAPS in that its overall accuracy could reach 80% with the improvement rate being about 3%.
下载: