北京林业大学学报2016，Vol.38Issue(11)：23-33,11.DOI:10.13332/j.1000-1522.20160054
基于Landsat 8 OLI辅助的亚米级遥感影像树种识别
Submeter remote sensing image recognition of trees based on Landsat 8 OLI support
摘要
Abstract
In order to study the validity of the object-based identification of tree species with high spatial resolution remote sensing image ( QuickBird) and multi spectral remote sensing image ( Landsat 8 OLI) coordinated, based on QuickBird high spatial resolution ( panchromatic 0. 61 m) remote sensing image and Landsat 8 OLI(30 m) remote sensing image, we used 2 segmentation schemes ( segmentation based on QuickBird remote sensing image with Landsat 8 OLI remote sensing image as an auxiliary or not) to do multi-scale segmentation, and compared the 2 segmentation schemes in the classification processing. This research applied nearest neighbor classification and support vector machine object-based classification methods, the same classification system, the unified segmentation scale and the same set of validation samples to classify tree species with 68 classification features in terms of spectral, texture and spatial extracted by QuickBird remote sensing image and Landsat 8 OLI remote sensing image, and then take use of Kappa coefficient, total accuracy, producer accuracy and user accuracy to evaluate the accuracy. The results showed that the segmentation result based on QuickBird high spatial resolution remote sensing image only was better than that based on QuickBird high spatial resolution remote sensing image and Landsat 8 OLI remote sensing image coordinated. The best segmentation threshold was 25 and the best merging threshold was 90. On the basis of the best segmentation threshold, applying Landsat 8 OLI multi spectral remote sensing image and QuickBird high spatial resolution remote sensing image together to take the object-based classification, the total accuracy of nearest neighbor classification method and support vector machine classification method was 85. 35% ( Kappa = 0. 701 3 ) and 88. 12% ( Kappa=0. 853 6 ) . And the total accuracy of the above two methods was 79. 67% ( Kappa =0. 693 9 ) and 83. 33% ( Kappa =0. 792 5 ) when using the QuickBird high spatial resolution remote sensing image only. Under the support of Landsat 8 OLI remote sensing image, object boundary of the classification result is clearer, the total accuracy and the accuracy of major tree species are significantly improved. The research results can effectively shorten the time and reduce the cost of investigation and survey, reduce labor intensity, improve the quality of products when it was applied in field forest survey and zoning.关键词
面向对象/QuickBird/森林类型/高空间分辨率Key words
object-based method/QuickBird/forest classification/high spatial resolution分类
农业科技引用本文复制引用
魏晶昱,毛学刚,方本煜,包晓建,许振宇..基于Landsat 8 OLI辅助的亚米级遥感影像树种识别[J].北京林业大学学报,2016,38(11):23-33,11.基金项目
国家自然科学基金项目(31300533)、“十二五”国家科技支撑计划项目(2012AA102001)、东北林业大学大学生创新项目(201410225152)。 （31300533）
