基于面向对象和CART决策树方法的遥感影像湿地变化检测研究——以龙祥岛地区为例  

Wetland Change Detection Based on Object-oriented and CART Decision Tree Method——A Case Study of Wetlands in Longxiang Island

作　　者：马宇龙 林志垒[1] 

机构地区：[1]福建师范大学地理科学学院,福建福州350117

出　　处：《福建师范大学学报（自然科学版）》2017年第6期69-80,共12页Journal of Fujian Normal University：Natural Science Edition

基　　金：国家自然科学基金资助项目(41501451);福建省自然科学基金资助项目(2011J01265)

摘　　要：以乌龙江流域龙祥岛地区为研究区,以两期不同水位的Landsat-8 OLI影像与高分一号(GF-1)PMS1影像为数据源,经过面向对象的分割技术,以CART(Classification And Regression Tree,CART)决策树方法挖掘不同地类的特征指数阈值,提取出该地区滩涂湿地并根据水位值进行变化检测分析.研究表明:GF-1 PMS1传感器8 m分辨率影像可以与Landsat-8 OLI传感器15 m分辨率影像结合进行变化检测,总体分类精度分别达到0.89和0.92;CART决策树算法与面向对象的分类方法相结合可以获得很好的分类效果,对于湿地滩涂信息提取具有高效快捷的优势,这种方法随着多元遥感的发展将展现巨大的应用潜力;水位变化对湿地滩涂的影响可以通过遥感影像进行检测分析,两期影像T1与T2两时刻水位高程分别为3.34 m与3.56 m,水位变化0.22 m,T1时刻河流水面面积为1 811.09 hm^2,T2时刻河流水面面积为2 092.34hm^2,两期变化增加的水面面积为281.25 hm^2,未变化面积为1 778.47 hm^2,占T1时刻总面积的98.20%,T1时刻内陆滩涂转化为T2时刻河流水面的面积为244.91 hm^2,占T1时刻内陆滩涂总面积的34.60%.Through the segmentation of object oriented technique,we used Landsat-8 OLI image and GF-1 PMS1 image as data source and characteristic index threshold of different land which mined by CART（ Classification And Regression Tree,CART） decision types tree method to extract the wetland intertidal zone. According to the variation in water level,we made change detection analysis,and using Longxiang Island area of the Wulong River Basin as the study area. Research shows that： GF-1 PMS1 sensor 8 m resolution image can be combined with the Landsat-8 OLI sensor 15 m resolution image for change detection,the overall classification accuracy of GF-1 PMS1 image is 0. 89,the overall classification accuracy of Landsat-8 OLI image is 0. 92; CART decision tree algorithm and object-oriented classification method can be combined to obtain a good classification effect,for the extraction of wetlands information,this method has advantages of fast and efficient and show a huge application potential with the development of multiple remote sensing; The influence of water level changes on wetland tidal flat can be detected and analyzed by remote sensing image,thewater level of T1 and T2 is 3. 34 m and 3. 56 m respectively,water level changed 0. 22 m,the surface area of the river in T1 is 1 811. 09 hm^2,and the surface area of the river in T2 is 2 092. 34 hm^2,increased surface area is 281. 25 hm^2,unchanged area is 1 778. 47 hm^2 accounted for 98. 20% of total area in T1; from T1 to T2,the area of Inland tidal flats transformed into the river surface was 244. 91 hm^2,which accounted for 34. 60% of the total area of the inland tidal flat.

关 键 词：面向对象 CART决策树 GF-1影像 Landsat-8影像 湿地变化检测 水位对比 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]