竹资源专题信息提取纹理特征量构建研究

在"同物异谱""异物同谱"问题较为严重的区域,单纯利用光谱信息难以有效地提取出竹资源专题信息。研究以ALOS影像为数据源,以南方多山县顺昌为试验区域,在考虑海拔、坡向、坡度等地形因子的前提下,通过分析竹、马尾松、杉木和其他树种等典型植被的纹理特征差异,构建了最佳纹理量,在此基础上,分别利用基于像元与面向对象的监督分类方法进行竹资源专题信息提取。经精度验证,提取的总体精度分别为71.5%、74.0%,说明:单纯利用纹理特征量亦能使竹资源专题信息的提取精度达到与单纯利用光谱信息相当的水平;通过最佳纹理量的构建,可以较有效地利用地物的纹理信息,从而更有效地利用遥感影像丰富的信息;实践证明了纹理特征量用于竹资源专题信息提取的可行性,为综合利用光谱与纹理信息进行专题信息提取的研究提供了实例支撑和技术参考,可以预见,基于光谱、纹理特征和地形因子的竹资源专题信息提取精度将得到进一步提高;在纹理特征量构建基础上进行竹资源专题信息提取,面向对象的监督分类较之于传统基于像元的监督分类的精度稍高,连续性更好。

Study on Texture Characteristics Construction for Bamboo Resources Thematic Information Extraction

In the regions of the same surface feature with different spectrums or different surface features with the same spectrum existing seriously,it is hard to extract the bamboo resources thematic information effectively relying on the texture information only.The study takes ALOS images as data source,and Shunchang County where many mountains as the testing area.Taking the topographical factors of elevation,aspect and slope into consideration and construct the best texture characteristics by way of the analysis of the texture difference among bamboo,mansson pine,Chinese fir and other kinds of trees.On the basis of the texture characteristics construction,it uses the supervised classification methods of pixel-based and object-oriented respectively,in order to extract the bamboo resources thematic information.The accuracy verification shows that the total accuracy reaches 71.5% and 74.0%.Four results could be indicated.Firstly,as using spectrum information singly,the bamboo thematic information extraction accuracy could also reach an acceptable level.Secondly,according to the texture construction,it could make use of the texture information of the surface features more effectively,so as to use the abundant information of the remote sensing images more effectively.Thirdly,the feasibility of bamboo resources thematic information drawing on the texture has been proved,which will offer the case support and technique reference for the comprehensive use of spectrum and texture information.The bamboo thematic information extraction accuracy would be increased further based on the combination of spectrum,texture characteristics and topographical factors.Fourthly,to extract the bamboo resources thematic information on the basis of texture construction,object-oriented supervised classification offers a bit of higher accuracy and performs better continuity than the traditional supervised pixel-based classification.