基于对象变化矢量分析的土地利用变化检测方法研究

变化检测是资源和环境遥感应用的一个重要内容。在变化矢量分析法的基础上,本文提出采用变化矢量-主成分分析法提取研究区变化信息,首先,对不同时相的遥感影像进行差值运算得到差值影像,再对其进行主成分变换并选取主分量,最后,使用多尺度分割获取影像对象。在影像分割的基础上,采用变化矢量-主成分分析方法构建自动检测规则提取变化信息,并作精度评价。研究表明:基于对象的变化矢量-主成分分析方法不仅可克服传统的基于像元式方法难以利用空间信息的缺陷,有效避免了"椒盐"噪声,而且可将多波段差值信息经主成分变换后集中在几个累计贡献率较高的主成分分量上;同时,结合了变化矢量法与主成分分析法的优点,与单独使用变化矢量分析法相比提取精度明显提高。

Research on Land Use Change Detection Based on an Object-oriented Change Vector Analysis Method

Land cover/use change in Taizhou City is studied in this paper. Due to human or natural factors, land cover/use constantly changes. With the development of remote sensing technology, change detection is one of the important applications on resource and environment remote sensing. In this paper, we propose a Change Vec-tor-PCA analysis method based on CVA to extract change information in the study area. First, we compute the relative difference between t1 and t2 remote sensing image of Taizhou to obtain the difference image, then em-ploy PCA to this difference image and select principal components for the extraction of change information. Next, the image objects are obtained by multi-scale segmentation combining with the spectral and spatial charac-teristics and suitable segmentation scale. And, the change information is extracted automatically with the rules derived from a classification tree on the basis of Change Vector-PCA analysis of the objects. Last, we have an ac-curacy evaluation according samples which reaches the mapping requirement. The reason of missed or false de-tection is that some land cover types is significantly affected by the reason, such as farmland, grass. The result shows that Change Vector-PCA based on objects is more superior than the traditional methods based on pixel in terms of utilizing the spatial information and avoiding“salt”noises;and this method can compare several princi-pal components transformed by multi-band difference change information. Change Vector-PCA combines the ad-vantage of CVA and PCA. It is significant in improving the accuracy and automation level for change detection.