文献类型：期刊文献

中文题名：对应分析在多时相遥感影像分类前变化检测中的应用--以江苏省盐城市为例

英文题名：Correspondence Analysis for Pre-Classification Change Detection Using Multi-Temporal Satellite Images:A Case Study in Yancheng City,Jiangsu Province

作者：刘春悦[1] 张树清[1] 江红星[2] 刘志明[3] 李晓峰[1] 那小东[1]

第一作者：刘春悦

机构：[1]中国科学院东北地理与农业生态研究所;[2]中国林业科学研究院森林生态环境与保护研究所;[3]东北师范大学城市与环境科学学院

年份：2008

卷号：30

期号：9

起止页码：1409-1414

中文期刊名：资源科学

外文期刊名：Resources Science

收录：CSTPCD;;北大核心:【北大核心2004】；CSCD:【CSCD2011_2012】；CSSCI:【CSSCI2008_2009】；

基金：联合国开发计划署(UNDP)/全球环境基金(GEF)项目(编号:CPR/98/G32);地球系统科学数据共享网项目(编号:2006DKA32300-16);国家“十一五”科技支撑重点项目(编号:2006BAD23B03)

语种：中文

中文关键词：对应分析;主成分分析;多时相遥感影像;分类前变化检测

外文关键词：Correspondence analysis; Principal component analysis; Multi-temporal satellite images ; Preclassification change detection; Yancheng

分类号：TP751;TP277

摘要：利用对应分析(Correspondence Analysis,CA)对江苏省盐城市及周边地区两个不同时相影像进行土地利用/土地覆被(landuse/landcover,LULC)变化检测。在对两幅辐射归一化影像采用CA算法进行变换后,将二者的第一主成分(Principal Component1,PC1)相减得到差值影像,利用总体精度(Overall Accuracy)曲线,经反复试验,最终设定±1.6δ作为上下限最佳阈值,对差值影像进行密度分割(DensitySlice),并提取出研究区域的植被变化信息。在利用混淆矩阵与主成分分析(Principal Component Analysis,PCA)所得的结果对比中发现,CA变换变化检测的结果要明显优于传统的PCA变化检测的结果。CA结果显示,CA变换变化检测结果的总体精度为89.60%,卡帕系数(Kappa Coefficient)为0.8194,比PCA变化检测结果卡帕系数提高近0.1,由此可见,CA变换可以作为用于检测土地覆被变化的一种有效的多元统计分析技术。
Correspondence analysis （CA） is a type of multivariate statistical analysis based on a data matrix with non-negative elements and related to principal component analysis （PCA）, which has recently been applied to pre-classification detection in satellite images. The CA arithmetic can be used to concentrate information from the original multi-temporal satellite images. This paper evaluates the accuracy of Land use/Land cover （LULC） change detection for Yancheng City in 1992 and 2002 using CA methods, and compares the results with PCA methods. The information of the principal component 1 of CA （CA-PCl） and PCA-PCl showed differences in 1992 and 2002. In CA-PC1, the loading of band 4 represented the highest value in both years, and is also the most vegetation-sensitive of the 7 spectral bands. The loadings of bands 3 and 7 took the second and third positions in the CA-PCl. However, the loading of band 5 showed the highest value in PCA-PCl and depended on its covariance with the other bands. In fact, bands 4, 3, and 7 are usually used in the calculation of vegetation or urbanization indexes. Thus, the CA-PCl contained the greenness information from the original images and was a better reflection of LULC change information. The difference images were acquired respectively by the first principal component （PCl） of the 1992 image subtracting that of 2002, which were transformed by the CA and PCA method after the images were radiometrically corrected. The threshold value （ ± 1.6δ） was defined based on the overall accuracy curve and repeated tests, and used to segment the difference images by density slices to obtain the land cover change region in the study area. The overall accuracy and Kappa Coefficient using the CA method are 89. 60% and 0. 8194, and that of PCA method are 85.60% and 0.7399 respectively. The CA-based change map produced a 10% higher Kappa Coefficient than the PCA base maps. Results showed that there was a 10.35km^2 increase in urban-related cover types between 1992 and 2002 in the area surrounding the old Yancheng city. Overall, the CA method had a higher accuracy and contained more information on LULC change than the traditional PCA method, which can be used for pre-classification detection in the multi-temporal satellite images as a powerful multivariate analysis technique. However, the CA method cannot be used to determine specific LULC types, which would be a valuable application.