遥感影像预分类精度对地物面积空间抽样估算的敏感性分析
|
摘要
传统的地物面积测量受精度和效率制约,为此引入了结合遥感影像的空间分层抽样方法。首先以遥感影像的预分类结果作为模拟地物的真实分布,在地物外沿等概率随机添加不同比例的错分像元,从而获得准真实地物区的摸拟预分类结果,并依此设定各层等比例取样的样本入层标志,指导地物样本的选取,然后以抽中样本地物的准真实值之和按比例推算出总量。通过比较分析各水平含量的地物类别、不同预分类精度、层内随机和系统抽样下的多次总量估计精度及其稳定性变化情况,结果表明:该方法不需要背景数据库等先验知识,在预分类达到一定精度之上时,依分类区域设立层标志的分层抽样方法所获得的总量估计精度及标准差均好于无分类支持的随机和系统抽样;当预分类精度达到50%以上时,具有较高的成本效率比,其中在60%时,各类地物在0.5%抽样率、95%的置信度下可以保证估计量精度在92%以上。
The traditional ground area measurement is limited by accuracy and efficiency.Remote sensing technology helps improve both but alone still far from satisfaction.Therefore,we proposed a spatial stratified sampling method based on remote sensing.The basic idea is using up to date remote sensing images to guide the target stratified sampling by a general classification,other than by those obsolete background knowledge databases.In order to validate the improvement on accuracy of calculating the real quantity with actually it unknown(which is just our goal to estimate),we took the early classification map from the remote sensing image(in this experiment,we used part of one scene of TM image covering Beijing area,a size of 4800×4800 pixels,and classified into 5 different types,3 of which were chosen)as the laboratorial quasi-real target(proportions vary from 7% to 32%).The detail methods and operations are described in the following steps: Firstly,we played back simulated pre-classification result,which completely contains the target,by iteratively adding error class pixels around the outskirts of the target to a demanded proportion;secondly.Secondly,we arranged square boxes(with a size of 20×20 pixels)on the pre-classification image,excluded zero-target ones and divided the rest into 5 strata according to the proportion of in-box pre-classification target area(pixels),randomly or systematically chose the samples.And then we estimated the gross by summing up the actual pixels in each sample pro rata.Finally,we analyzed and compared the quality and variation of estimation accuracy repetitiously with different land cover types,different pre-classification precision levels,and two methods of random and system in stratum.The results mainly presented the relationship between estimation accuracy and pre-classification accuracy in each target type,which showed that the estimation accuracy degraded when the stratified sampling method was aided with rough pre-classifications(accuracy less than 40%),but remarkably reached higher accuracy and stabilization than those of unsupported random or systematic sampling methods with pre-classification above a certain accuracy level.For the former situation,the degradation is mainly caused by the extreme inconsistency of area distributing direction poorly classified,and does not take place in common classifications.In general,this method has its best cost-efficiency at 50% pre-classification accuracy,a case of which is that the accuracy of the estimators for each target class with all proportions at 0.5% sampling ratio level and 95% confidence level can be acquired above 92% when the accuracy of pre-classification reaches 60%.During the study,we innovated in the following aspects: first,we created a simulated classification map with an assumed given target,and this map worked properly to show up the real situation;second,with the help of pre-classification from remote sensing images,the stratified sampling method can be much more effective and precise,and less in need of prior knowledge.
The traditional ground area measurement is limited by accuracy and efficiency.Remote sensing technology helps improve both but alone still far from satisfaction.Therefore,we proposed a spatial stratified sampling method based on remote sensing.The basic idea is using up to date remote sensing images to guide the target stratified sampling by a general classification,other than by those obsolete background knowledge databases.In order to validate the improvement on accuracy of calculating the real quantity with actually it unknown(which is just our goal to estimate),we took the early classification map from the remote sensing image(in this experiment,we used part of one scene of TM image covering Beijing area,a size of 4800×4800 pixels,and classified into 5 different types,3 of which were chosen)as the laboratorial quasi-real target(proportions vary from 7% to 32%).The detail methods and operations are described in the following steps: Firstly,we played back simulated pre-classification result,which completely contains the target,by iteratively adding error class pixels around the outskirts of the target to a demanded proportion;secondly.Secondly,we arranged square boxes(with a size of 20×20 pixels)on the pre-classification image,excluded zero-target ones and divided the rest into 5 strata according to the proportion of in-box pre-classification target area(pixels),randomly or systematically chose the samples.And then we estimated the gross by summing up the actual pixels in each sample pro rata.Finally,we analyzed and compared the quality and variation of estimation accuracy repetitiously with different land cover types,different pre-classification precision levels,and two methods of random and system in stratum.The results mainly presented the relationship between estimation accuracy and pre-classification accuracy in each target type,which showed that the estimation accuracy degraded when the stratified sampling method was aided with rough pre-classifications(accuracy less than 40%),but remarkably reached higher accuracy and stabilization than those of unsupported random or systematic sampling methods with pre-classification above a certain accuracy level.For the former situation,the degradation is mainly caused by the extreme inconsistency of area distributing direction poorly classified,and does not take place in common classifications.In general,this method has its best cost-efficiency at 50% pre-classification accuracy,a case of which is that the accuracy of the estimators for each target class with all proportions at 0.5% sampling ratio level and 95% confidence level can be acquired above 92% when the accuracy of pre-classification reaches 60%.During the study,we innovated in the following aspects: first,we created a simulated classification map with an assumed given target,and this map worked properly to show up the real situation;second,with the help of pre-classification from remote sensing images,the stratified sampling method can be much more effective and precise,and less in need of prior knowledge.
引文
[1]Liu J Y.Macro-Investigation and Dynamic Research of ChineseResource Environmental Remote Sensing[M].Beijing:ChinaScience and Technology Press,1996.[刘纪远.中国资源环境遥感宏观调查与动态研究[M].北京:中国科学技术出版社,1996.]
[2]Wu B F,Li QZ.Crop Acreage Estimation Using Two IndividualSampling Frameworks with Stratification[J].Journal ofRemoteSensing,2004,8(6):551—569[吴炳方,李强子.基于两个独立抽样框架的农作物种植面积遥感估算方法[J].遥感学报,2004,8(6):551—569.]
[3]Minasny B,McBratney A B,Walvoort D J J.The VarianceQuadtree Algorithm:Use for Spatial Sampling Design[J].Computers and Geosciences,2007,33:383—392.
[4]WangJ F,Liu,J Y,Zhuan,D F,et al.Spatial Sampling De-sign for Monitoring the Area of Cultivated Land[J].Internation-al Journal ofRemote Sensing,2002,23(2):263—284.
[5]Wang J F,Haining R,Wise S.ADesign for Spatial Sampling ofChina Drought,Flood and Earthquake Disaster Monitoring[J].Progress in Natural Science,1999,9(4):336—345.[王劲峰,Haining R,Wise S.中国干旱,洪水,地震灾害监测空间采样设计[J].自然科学进展,1999,9(4):336—345.]
[6]Li L F,Wang J F,Liu J Y.Spatial Sampling Optimized Deci-sion-making of National Land Remote Sensing Investigation[J].Science in China Sec.D Earth Sciences,2004,34(10):975—982.[李连发,王劲峰,刘纪远.国土遥感调查的空间抽样优化决策[J].中国科学D辑地球科学,2004,34(10):975—982.]
[7]Liu HQ.Sampling Method With Remote Sensing for Monitoringof Cultivated Land Changes on Large Scale[J].Transactions ofthe Chinese Society ofAgricultural Engineering,2001,17(2):168—171.[刘海启.大尺度耕地变化监测的遥感抽样方法研究[J].农业工程学报,2001,17(2):168—171.]
[8]Jiao X F,Yang B J,Pei Z Y.Paddy Rice Area EstimationUsing a Stratified Sampling Method with Remote Sensing in Chi-na[J].Transactions ofthe Chinese SocietyofAgricultural Engi-neering,2006,22(5):105—110.[焦险峰,杨邦杰,裴志远.基于分层抽样的中国水稻种植面积遥感调查方法研究[J].农业工程学报,2006,22(5):105—110.]
[9]McBratney,A B,Webster,R.The Design of Optimal SamplingSchemes for Local Estimation and Mapping of Regionalized Varia-bles[J].Computers and Geosciences,1981,7:331—334.
[10]Brus,D J,Spatjens,L E E M,de Gruijter,J J.A SamplingScheme for Estimating the Mean Extractable Phosphorus Concen-tration of Fields for Environmental Regulation[J].Geoderma,1999,89:129—148.
[11]Brus,D J,de Gruijter,J J,van Groenigen,J W.DesigningSpatialCoverage Samples Using the K-means Clustering Algorithm[A].Lagacherie,P,McBratney,AB,Voltz,M.Digital Soil Mapping:An Introductory Perspective[C].Amsterdam:Elsevier,2006.
[12]Gallego F J,Stratified Sampling of Satellite Images with a Sys-tematic Grid of Points[J].International Society ofPhotogram-metry and Remote Sensing Journal ofPhotogrammetry and Re-mote Sensing,2005,59:369—376.
[13]Chen Y K,Hu R.The Theoretic Structure and ComputerizedRealization of Stratified Audit Sampling[J].The Theory andPractice ofFinance and Economics,2003,24(125):78—80.[陈月昆,胡荣.分层审计抽样的理论架构与计算机实现[J].财经理论与实践,2003,24(125):78—80.]
[14]Wang S G,Chen W H,Gao L D.Probability Theory and Math-ematical Statistics[M].Beijing:Science Press,2000.[王松桂,程维虎,高旅端.概率论与数理统计[M].北京:科学出版社,2000.]
[2]Wu B F,Li QZ.Crop Acreage Estimation Using Two IndividualSampling Frameworks with Stratification[J].Journal ofRemoteSensing,2004,8(6):551—569[吴炳方,李强子.基于两个独立抽样框架的农作物种植面积遥感估算方法[J].遥感学报,2004,8(6):551—569.]
[3]Minasny B,McBratney A B,Walvoort D J J.The VarianceQuadtree Algorithm:Use for Spatial Sampling Design[J].Computers and Geosciences,2007,33:383—392.
[4]WangJ F,Liu,J Y,Zhuan,D F,et al.Spatial Sampling De-sign for Monitoring the Area of Cultivated Land[J].Internation-al Journal ofRemote Sensing,2002,23(2):263—284.
[5]Wang J F,Haining R,Wise S.ADesign for Spatial Sampling ofChina Drought,Flood and Earthquake Disaster Monitoring[J].Progress in Natural Science,1999,9(4):336—345.[王劲峰,Haining R,Wise S.中国干旱,洪水,地震灾害监测空间采样设计[J].自然科学进展,1999,9(4):336—345.]
[6]Li L F,Wang J F,Liu J Y.Spatial Sampling Optimized Deci-sion-making of National Land Remote Sensing Investigation[J].Science in China Sec.D Earth Sciences,2004,34(10):975—982.[李连发,王劲峰,刘纪远.国土遥感调查的空间抽样优化决策[J].中国科学D辑地球科学,2004,34(10):975—982.]
[7]Liu HQ.Sampling Method With Remote Sensing for Monitoringof Cultivated Land Changes on Large Scale[J].Transactions ofthe Chinese Society ofAgricultural Engineering,2001,17(2):168—171.[刘海启.大尺度耕地变化监测的遥感抽样方法研究[J].农业工程学报,2001,17(2):168—171.]
[8]Jiao X F,Yang B J,Pei Z Y.Paddy Rice Area EstimationUsing a Stratified Sampling Method with Remote Sensing in Chi-na[J].Transactions ofthe Chinese SocietyofAgricultural Engi-neering,2006,22(5):105—110.[焦险峰,杨邦杰,裴志远.基于分层抽样的中国水稻种植面积遥感调查方法研究[J].农业工程学报,2006,22(5):105—110.]
[9]McBratney,A B,Webster,R.The Design of Optimal SamplingSchemes for Local Estimation and Mapping of Regionalized Varia-bles[J].Computers and Geosciences,1981,7:331—334.
[10]Brus,D J,Spatjens,L E E M,de Gruijter,J J.A SamplingScheme for Estimating the Mean Extractable Phosphorus Concen-tration of Fields for Environmental Regulation[J].Geoderma,1999,89:129—148.
[11]Brus,D J,de Gruijter,J J,van Groenigen,J W.DesigningSpatialCoverage Samples Using the K-means Clustering Algorithm[A].Lagacherie,P,McBratney,AB,Voltz,M.Digital Soil Mapping:An Introductory Perspective[C].Amsterdam:Elsevier,2006.
[12]Gallego F J,Stratified Sampling of Satellite Images with a Sys-tematic Grid of Points[J].International Society ofPhotogram-metry and Remote Sensing Journal ofPhotogrammetry and Re-mote Sensing,2005,59:369—376.
[13]Chen Y K,Hu R.The Theoretic Structure and ComputerizedRealization of Stratified Audit Sampling[J].The Theory andPractice ofFinance and Economics,2003,24(125):78—80.[陈月昆,胡荣.分层审计抽样的理论架构与计算机实现[J].财经理论与实践,2003,24(125):78—80.]
[14]Wang S G,Chen W H,Gao L D.Probability Theory and Math-ematical Statistics[M].Beijing:Science Press,2000.[王松桂,程维虎,高旅端.概率论与数理统计[M].北京:科学出版社,2000.]
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心