一种改进型面向对象的遥感变化检测方法

An Advanced Change Detection Method Based on Object-oriented Classification of Remote Sensing Imagery

针对人工变化发现效率低下和高分影像数据量大的现状,按照"先发现变化区域,再进行分类识别"的检测思路,提出了一种集成改进型变化矢量分析法和面向对象分类法的变化检测方法。该方法利用基于最大类间方差阈值搜索的改进型CVA法以自动快速提取出变化区域,再基于面向对象分类法对变化区域进行地物类型识别。为验证该方法的有效性,本文利用2009年和2013年两期IKONOS高分影像对重庆市沙坪坝区大学城建成区进行了变化检测实验,并采用变化检测一致性比率进行精度分析。实验结果表明,该方法对道路和建筑用地的变化检测精度可以达到85%以上,对耕地和水体的地类检测精度达到70%以上。可见,该方法在大区域的宏观土地利用遥感动态监测工作中具有一定的实用价值。

In order to deal with the inefficient detection of artifacts among the large amount of image data,we proposed an integrated improved change vector analysis and object-oriented classification method,which follows the idea that detecting changes in the region firstly and then performing classification.The method firstly uses a large law to determine the optimal threshold value to improve CVA method to automatically and rapidly extract the change regions.Then the change region is object-oriented classified based on feature type identification.To verify the validity of this method,we use the IKONOS images of university district of Chongqing area,where the images are acquired in 2009 and 2013,respectively.Experimental results show that the method can reach more than 85% accuracy on the changes of roads and construction sites detection;and under the influence of the Chongqing area polygon plots broken,the detection accuracy of arable land types and water bodies can still reach more than 70%.The results indicate that this method is reliable and efficient in land-use change detection,and has practical value in land use dynamic monitoring by remote sensing in large district.