基于非光谱信息的龙口市土地利用遥感分类方法研究
|
摘要
随着遥感技术的飞速发展,遥感影像对土地利用分类起到了举足轻重的作用,分类的精度直接影响遥感影像的实用价值和实用水平,但是由于目视判读耗时长、需要人力资源大等缺点已经使人们越来越多关注遥感影像计算机分类的研究,如何提高土地利用计算机分类的精度也是越来越被关心的话题。但是遥感卫星获得的影像表现的只是自然综合体里面在二维平面上所表现的信息的集合,只利用遥感影像的这些光谱信息进行分类很难达到理想的效果。本文在中德科技合作与交流项目(2007DFB70200)和山东省自然科学基金(Y2008E10)的资助下,主要根据龙口市的实际情况,利用一些综合的遥感信息或者某些非光谱信息来辅助计算机进行分类,可以提高某种或者某几种地类的分类精度。主要研究内容和成果如下:
(1)在参阅前人的研究成果的基础的上,阐述了遥感影像分类的理论方法,分析了几种常见的计算机分类方法,介绍了遥感影像预处理的理论及方法,对遥感影像进行前期的预处理,并且分析了研究区遥感影像的特征及最佳波段组合。
(2)在资料收集的基础上,通过对研究区域遥感影像的研究以及实地考察,建立研究区域的训练样本,对其进行土地利用分类。通过分类发现遥感影像容易产生“同物异谱”或者“异物同谱”的现象,因此本文将“同物异谱”的地类分成多类,然后进行合并,对于“异物同谱”的地类并不好区分,因此合并成一类,利用非光谱数据再进行分类。
(3)探讨了龙口市遥感影像的在不同非光谱信息的支持的分类精度和不同的分类方法的精度。通过对研究区域的研究,发现龙口市遥感影像的“异物同谱”现象大多是由于地形因素的原因引起的,因此本文分别用DEM、坡度和坡度变化与影像进行融合,然后进行监督分类,发现融合DEM的影像分类精度较高,达到90.2%,kappa系数为0.8692;又对融合DEM的影像通过决策树法和神经网络法进行分类,发现神经网络分类的精度较高,总体精度为93.4%,kappa系数为0.9099。分类结果说明了根据研究区域的实际情况引入相应的非光谱信息改变其原来的光谱信息可以相应地提高分类精度,比传统的基于光谱信息进行分类的精度有很大的提高;神经网络分类方法要优于另外两种分类方法。
With the rapid development of remote sensing technology, remote sensing image plays a decisive function. The precision of classification influences directly the practical value and level of remote sensing technology. But because the visual interpretation has shortcomings such as time-consuming, more human resources and so on, people have paid more and more attentions to the study of computer classification of the remote sensing image. How to enhance the precision of land-use computer classification is also a topic getting more and more concerns. Because the images received by remote sensing satellite only present the collection of information on the two-dimensional plane in natural complex, it’s difficult to reach the ideal effect only with remote sensing image to classify these spectral messages. By joint Chinese-German technical project (2007DFB70200) and Shandong provincial natural science foundation (Y2008E10), I explain that using some integrated remote information or some non-spectral information to help computer make classifications, which can enhance the classification precision of one certain or several land categories. The main research contents and results are as follows:
(1)On the basis of seeing previous research results, I described the theoretical methods of remote sensing image classification, analyzed several common methods of computer classification, introduced the theory and methods of remote sensing image pretreatment, made the previous pretreatment of remote sensing image, and analyzed the features of remote sensing image and the best optimal band combinations in the study area.
(2)On the basis of material collection, I built the training samples and classified land-use of the study area by studying remote sensing image and field survey. By classification, I found remote sensing image easily produce two phenomena: one is called“same object with different spectra”, and the other is called“different objects with same spectra”. Therefore the essay classified the land category of the former phenomenon into many kinds, and then combined them. It’s not easy to distinguish the land category of the latter one, so combining them into one and then reclassifying with the help of non-spectral data.
(3)Having discussed the classification precision of remote sensing image supported by different non-spectral information and of different classification methods in Long Kou city. By researching the study area, I found the phenomenon of“different objects with same spectra”mainly is caused by topographic factors. So, this essay combines image with DEM, grade of slope, and changing slope, then supervise and classify them. The classification precision of image combined with DEM is higher, reaching 99.2 percent, and the index of kappa is 0.8692. The image with DEM is classified by decision-tree method and neural network method. The precision of neural network classification is relatively high, with whole precision of 93.4 percent. And the index of kappa is 0.9099. The classification results indicate that according to the factual conditions of the study area, introducing corresponding non-spectral information to change the original spectral information can enhance the classification precision relatively. It has more advancement than the traditional precision classified on spectral information. The neural network classification method is superior to the other two classification methods.
(1)在参阅前人的研究成果的基础的上,阐述了遥感影像分类的理论方法,分析了几种常见的计算机分类方法,介绍了遥感影像预处理的理论及方法,对遥感影像进行前期的预处理,并且分析了研究区遥感影像的特征及最佳波段组合。
(2)在资料收集的基础上,通过对研究区域遥感影像的研究以及实地考察,建立研究区域的训练样本,对其进行土地利用分类。通过分类发现遥感影像容易产生“同物异谱”或者“异物同谱”的现象,因此本文将“同物异谱”的地类分成多类,然后进行合并,对于“异物同谱”的地类并不好区分,因此合并成一类,利用非光谱数据再进行分类。
(3)探讨了龙口市遥感影像的在不同非光谱信息的支持的分类精度和不同的分类方法的精度。通过对研究区域的研究,发现龙口市遥感影像的“异物同谱”现象大多是由于地形因素的原因引起的,因此本文分别用DEM、坡度和坡度变化与影像进行融合,然后进行监督分类,发现融合DEM的影像分类精度较高,达到90.2%,kappa系数为0.8692;又对融合DEM的影像通过决策树法和神经网络法进行分类,发现神经网络分类的精度较高,总体精度为93.4%,kappa系数为0.9099。分类结果说明了根据研究区域的实际情况引入相应的非光谱信息改变其原来的光谱信息可以相应地提高分类精度,比传统的基于光谱信息进行分类的精度有很大的提高;神经网络分类方法要优于另外两种分类方法。
With the rapid development of remote sensing technology, remote sensing image plays a decisive function. The precision of classification influences directly the practical value and level of remote sensing technology. But because the visual interpretation has shortcomings such as time-consuming, more human resources and so on, people have paid more and more attentions to the study of computer classification of the remote sensing image. How to enhance the precision of land-use computer classification is also a topic getting more and more concerns. Because the images received by remote sensing satellite only present the collection of information on the two-dimensional plane in natural complex, it’s difficult to reach the ideal effect only with remote sensing image to classify these spectral messages. By joint Chinese-German technical project (2007DFB70200) and Shandong provincial natural science foundation (Y2008E10), I explain that using some integrated remote information or some non-spectral information to help computer make classifications, which can enhance the classification precision of one certain or several land categories. The main research contents and results are as follows:
(1)On the basis of seeing previous research results, I described the theoretical methods of remote sensing image classification, analyzed several common methods of computer classification, introduced the theory and methods of remote sensing image pretreatment, made the previous pretreatment of remote sensing image, and analyzed the features of remote sensing image and the best optimal band combinations in the study area.
(2)On the basis of material collection, I built the training samples and classified land-use of the study area by studying remote sensing image and field survey. By classification, I found remote sensing image easily produce two phenomena: one is called“same object with different spectra”, and the other is called“different objects with same spectra”. Therefore the essay classified the land category of the former phenomenon into many kinds, and then combined them. It’s not easy to distinguish the land category of the latter one, so combining them into one and then reclassifying with the help of non-spectral data.
(3)Having discussed the classification precision of remote sensing image supported by different non-spectral information and of different classification methods in Long Kou city. By researching the study area, I found the phenomenon of“different objects with same spectra”mainly is caused by topographic factors. So, this essay combines image with DEM, grade of slope, and changing slope, then supervise and classify them. The classification precision of image combined with DEM is higher, reaching 99.2 percent, and the index of kappa is 0.8692. The image with DEM is classified by decision-tree method and neural network method. The precision of neural network classification is relatively high, with whole precision of 93.4 percent. And the index of kappa is 0.9099. The classification results indicate that according to the factual conditions of the study area, introducing corresponding non-spectral information to change the original spectral information can enhance the classification precision relatively. It has more advancement than the traditional precision classified on spectral information. The neural network classification method is superior to the other two classification methods.
引文
[1]蒋艳凰.遥感图像高精度并行监督分类技术研究[D].国防科学技术大学,2004:
[2]彭望碌等.遥感概论[M].北京:高等教育出版社,2002, 261.
[3]杨玉静.遥感影像的计算机分类技术研究[D].昆明理工大学,2006:
[4]章孝灿,黄智才,赵元洪.遥感数字图像处理[M].浙江:浙江大学出版社.1997:
[5]吴学军.城市TM遥感图像分类方法研究[D].广西师范大学,2007:
[6]郑明国.ERDAS软件支持下的土地利用/土地覆盖分类研究—以郑州市土地利用/土地覆盖制图为例[D].河南大学,2002:
[7]刘建平,赵英时.高光谱遥感数据解译的最佳波段选择方法研究[J].中国科学院研究生院学报,1999,16(10):153-161
[8]赵英时等.遥感应用分析原理与方法[M].北京:科学出版社,2003:259-264
[9]刘建平,赵英时,孙淑玲.高光谱遥感数据最佳波段选择方法实验研究[J].遥感技术与应用,2001,16(1):7-13
[10]赵新振,朱登强,黄家荣等.矿区植被遥感调查最佳波段组合的选择[J].山地农业生物学报,2008,27(5):464-466.
[11]濮静娟等.遥感图像目视解译原理与方法[M].北京:中国科学技术出版社,1990:
[12]孟治国.BP神经网络在土地利用分类中的应用分析[D].吉林大学,2004:
[13]方从刚.RBFNN算法在遥感土地分类中的应用—以内江市为例[D].四川农业大学,2007:
[14]曾志远.卫星遥感图像计算机分类与地学应用研究[M].北京:科学出版社,2004:
[15]李四海.提高遥感数据分类应用的有效途径[J].国土资源遥感,1995,15(4):252-255
[16]廖文峰.基于数据融合的土地遥感分类技术研究[D].新疆农业大学,2004:
[17]罗红霞.地学知识辅助遥感进行山地丘陵区基于系统分类标准的土壤自动分类方法研究[D].武汉大学,2005:
[18]杨龙士,周天颖.遥感探测理论与分析实务[D].逢甲大学,2000:
[19]帅美新.基于RS技术的鄱阳湖区土地利用的分类研究[D].南昌大学,2005:
[20]余建杰.基于TM遥感影像的鄱阳湖湿地分类研究[D].南昌大学,2005:
[21]吴健平,杨星卫.遥感数据监督分类中训练样本的纯化[J].国土资源遥感,1996,(1):36-41
[22]李宝华.基于GIS改进先验概率的遥感土地利用/土地覆盖分类研究—以开封市为例[D].河南大学,2006:
[23]许殿元,丁树柏.遥感影像信息处理[D].北京:宇航出版社,1991:
[24]方红亮,黄绚.地学应用中的遥感图像处理若干问题的分析[J].地理研究,1997年,16(2):96-104
[25]肖捷颖.基于GIS和Landsat TM/ETM+遥感资料分析的城市土地利用/覆盖变化研究—以石家庄市为例[D].河北师范大学,2002.
[26]杨玉静.遥感影像的计算机分类技术研究[D].昆明理工大学,2006:
[27]Congalton R.G.A review of assessing the accuracy of classification of remotely sensed data.Remote Sensing of Environment,1991,37:35-46
[28]周成虎等.遥感影像地学理解与分析[M].科学出版社,北京,1999:
[29]杜明义,武文波,郭达志.多源地学信息在土地荒漠化遥感分类中的应用研究[J].中国图象图形学报,2002,7(7):740-744
[30]段晓芹.对LANDSAT7 ETM+遥感图像分类识别技术的研究与应用[D].内蒙古大学,2006:
[31]马兰.基于1:5万地形图更新的Landsat-7遥感影像融合方法的研究[D].解放军信息工程大学,2005:
[32]王雷,冯学智,都金康.遥感影像分类与地学知识发现的集成研究[J].地理研究,2001,20(5):637-643
[33]陈鑫.基于决策树技术的遥感影像分类研究[D].北京林业大学,2006:
[34]刘伟.基于人工神经网络的遥感影像模式分类研究[D].鄱阳工业大学,2005:
[35]于君娜.基于神经网络的高光谱遥感图像分类研究[D].哈尔滨工业大学,2007:
[36]陈鑫.基于决策树技术的遥感影像分类研究[D].北京林业大学,2006:
[37]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京:清华大学出版社,2000:
[38]Stephon I Gallant.Neural Network Learning and Expert Systems.Gambrdge.Massachusetts: The MIT Press,1992:
[39]郭欣.遥感图像的分类[D].北京化工大学,2004:
[40]段新成.基于BP人工神经网络的土地利用分类遥感研究[D].中国地质大学,2008:
[2]彭望碌等.遥感概论[M].北京:高等教育出版社,2002, 261.
[3]杨玉静.遥感影像的计算机分类技术研究[D].昆明理工大学,2006:
[4]章孝灿,黄智才,赵元洪.遥感数字图像处理[M].浙江:浙江大学出版社.1997:
[5]吴学军.城市TM遥感图像分类方法研究[D].广西师范大学,2007:
[6]郑明国.ERDAS软件支持下的土地利用/土地覆盖分类研究—以郑州市土地利用/土地覆盖制图为例[D].河南大学,2002:
[7]刘建平,赵英时.高光谱遥感数据解译的最佳波段选择方法研究[J].中国科学院研究生院学报,1999,16(10):153-161
[8]赵英时等.遥感应用分析原理与方法[M].北京:科学出版社,2003:259-264
[9]刘建平,赵英时,孙淑玲.高光谱遥感数据最佳波段选择方法实验研究[J].遥感技术与应用,2001,16(1):7-13
[10]赵新振,朱登强,黄家荣等.矿区植被遥感调查最佳波段组合的选择[J].山地农业生物学报,2008,27(5):464-466.
[11]濮静娟等.遥感图像目视解译原理与方法[M].北京:中国科学技术出版社,1990:
[12]孟治国.BP神经网络在土地利用分类中的应用分析[D].吉林大学,2004:
[13]方从刚.RBFNN算法在遥感土地分类中的应用—以内江市为例[D].四川农业大学,2007:
[14]曾志远.卫星遥感图像计算机分类与地学应用研究[M].北京:科学出版社,2004:
[15]李四海.提高遥感数据分类应用的有效途径[J].国土资源遥感,1995,15(4):252-255
[16]廖文峰.基于数据融合的土地遥感分类技术研究[D].新疆农业大学,2004:
[17]罗红霞.地学知识辅助遥感进行山地丘陵区基于系统分类标准的土壤自动分类方法研究[D].武汉大学,2005:
[18]杨龙士,周天颖.遥感探测理论与分析实务[D].逢甲大学,2000:
[19]帅美新.基于RS技术的鄱阳湖区土地利用的分类研究[D].南昌大学,2005:
[20]余建杰.基于TM遥感影像的鄱阳湖湿地分类研究[D].南昌大学,2005:
[21]吴健平,杨星卫.遥感数据监督分类中训练样本的纯化[J].国土资源遥感,1996,(1):36-41
[22]李宝华.基于GIS改进先验概率的遥感土地利用/土地覆盖分类研究—以开封市为例[D].河南大学,2006:
[23]许殿元,丁树柏.遥感影像信息处理[D].北京:宇航出版社,1991:
[24]方红亮,黄绚.地学应用中的遥感图像处理若干问题的分析[J].地理研究,1997年,16(2):96-104
[25]肖捷颖.基于GIS和Landsat TM/ETM+遥感资料分析的城市土地利用/覆盖变化研究—以石家庄市为例[D].河北师范大学,2002.
[26]杨玉静.遥感影像的计算机分类技术研究[D].昆明理工大学,2006:
[27]Congalton R.G.A review of assessing the accuracy of classification of remotely sensed data.Remote Sensing of Environment,1991,37:35-46
[28]周成虎等.遥感影像地学理解与分析[M].科学出版社,北京,1999:
[29]杜明义,武文波,郭达志.多源地学信息在土地荒漠化遥感分类中的应用研究[J].中国图象图形学报,2002,7(7):740-744
[30]段晓芹.对LANDSAT7 ETM+遥感图像分类识别技术的研究与应用[D].内蒙古大学,2006:
[31]马兰.基于1:5万地形图更新的Landsat-7遥感影像融合方法的研究[D].解放军信息工程大学,2005:
[32]王雷,冯学智,都金康.遥感影像分类与地学知识发现的集成研究[J].地理研究,2001,20(5):637-643
[33]陈鑫.基于决策树技术的遥感影像分类研究[D].北京林业大学,2006:
[34]刘伟.基于人工神经网络的遥感影像模式分类研究[D].鄱阳工业大学,2005:
[35]于君娜.基于神经网络的高光谱遥感图像分类研究[D].哈尔滨工业大学,2007:
[36]陈鑫.基于决策树技术的遥感影像分类研究[D].北京林业大学,2006:
[37]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京:清华大学出版社,2000:
[38]Stephon I Gallant.Neural Network Learning and Expert Systems.Gambrdge.Massachusetts: The MIT Press,1992:
[39]郭欣.遥感图像的分类[D].北京化工大学,2004:
[40]段新成.基于BP人工神经网络的土地利用分类遥感研究[D].中国地质大学,2008:
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |