摘要
以湖南省攸县为研究区,利用2009、2010年SPOT5影像,在抽样可靠性指标为95%的情况下,设计系统抽样方案、分层抽样方案和简单随机抽样,通过SVM进行图像分类,并结合2009年湖南省连续清查数据对方案进行精度验证,得到适合研究区的抽样方案。结果表明:3种抽样方案中,适宜攸县森林资源调查的最优方案为,以抽样间隔为4 km×6 km(第Ⅰ层)、4 km×4 km(第Ⅱ层)、4 km×4 km(第Ⅲ层)进行的分层抽样,总体分类精度达到90.48%。其中,在系统抽样中,抽样间隔为4 km×4 km和2 km×2 km的方案总体精度均为88.10%,但前者训练样本数较少,表明在实际调查中,训练样本的数量与抽样的总体精度不是一直呈正相关。在分层抽样中,适合各层的最优抽样方案不一定相同,并且与系统抽样的最优方案也不一定相同。当抽样间隔相同时,分层抽样的总体精度要高于系统抽样的总体精度,但训练样本数少于系统抽样的训练样本数。所以在实际调查中,采用分层抽样较系统抽样,得到的精度较高,并且耗费的人力物力较少,较为高效。
Taking You County in Hunan Province as the research area,SPOT5 images in 2009 and 2010 were used to designed schemes on systematic sampling,stratified sampling and simple random sampling under the sampling reliability index of 95%. Classification of images by SVM,combined with continuous inspection data from Hunan Province in 2009 to verify the accuracy of the plan,to obtain a sampling plan suitable for the study area. Results show that among 3 sampling schemes,the optimal scheme for investigating forest resources in You County was 4 km × 6 km（ Layer Ⅰ）,4 km × 4 km（ Layer Ⅱ） and 4 km × 4 km（ Layer Ⅲ）. In addition,the overall classification accuracy reached 90. 48%. In systematic sampling,the overall accuracy of the schemes with sampling intervals of 4 km × 4 km and 2 km × 2 km was 88. 10%,but the number of training samples in the former was small. It showed that in actual survey,the number of training samples and the overall accuracy of sampling were not always positively correlated. In stratified sampling,the optimal sampling scheme for each layer was not necessarily the same,and it also had a difference with the optimal sample for systematic sampling. When the sampling intervals were the same,the overall accuracy of stratified sampling was higher than the overall accuracy of systematic sampling,but the number of training samples was less than the number of training samples sampled by the system.Therefore,in actual surveys,stratified sampling was more accurate than systematic sampling. Thus,it meant stratified sampling can consume less manpower and material resources and was more efficient.
作者
蒋仟
林辉
严恩萍
罗攀
Jiang Qian;Lin Hui;Yan Enping;Luo Pan(Research Center of Forestry Remote Sensing ＆ Information Engineering, Central South Lniversity of Forestry and Technology,Changsha Hunan 410004, China;Key Laboratory of Forestry Remote Sensing Based Big Data ＆ Ecological Security,Hunan Provincial Science ＆ Technology Department, Changsha Hunan 410013, China;College of Forestry, Central South Lniversity of Forestry and Technology, Changsha Hunan 410004, China)
出处
《西南林业大学学报（自然科学）》
CAS
北大核心
2018年第3期145-150,共6页
Journal of Southwest Forestry University：Natural Sciences
基金
国家十三五重点研发计划子课题(2017YFD0600902-4)资助
湖南省科技厅项目(2016TP1014)资助
中南林业科技大学青年自然科学基金(QJ2017002B)资助
关键词
森林
资源调查
系统抽样
分层抽样
支持向量机
SPOT5影像
攸县
forest
resource survey
systematic sampling
stratified sampling
Support Vector Machine
SPOT5 images
You County
