基于支持向量机的高光谱遥感影像道路提取  

Hyperspectral RS image road feature extraction based on SVM

作　　者：沈照庆[1] 黄亮 陶建斌[3] 

机构地区：[1]长安大学公路学院,陕西西安710064 [2]深圳市中地软件工程有限公司,广东深圳518001 [3]武汉大学遥感信息工程学院,湖北武汉430079

出　　处：《长安大学学报（自然科学版）》2012年第5期34-38,共5页Journal of Chang’an University(Natural Science Edition)

基　　金：国家重点基础研究发展计划973项目(2006CB701303);长江学者和创新团队发展计划资助项目(NO.1050);中央高校基本科研业务费专项资金项目(CHD2011JC011)

摘　　要：为了从高光谱遥感影像中高精度提取各种线形道路,提出了基于支持向量机(SVM)的道路特征快速提取算法,首先利用PCA对高光谱影像进行合理压缩,由SVM模式识别理论推导出该算法具有快速精确提取道路网信息的能力,针对高光谱遥感影像高信息量和道路网复杂度高的特点,提出基于1Vm(一对多算法)的多种道路SVM一次性高精度提取的多分类策略,在提高精度的同时,兼顾了道路特征识别的效率。研究结果表明:SVM对线状道路模式判别能力比常规方法有更强的优势,对小样本的道路识别效果更加明显,从遥感影像中不仅能准确地辨别出道路的线形特征,还能识别出其材质和类型;该算法能同时识别出多种道路,执行效率更高。In order to accurately extract various types of linear road from hyperspectral remote sensing（RS） images,a road feature fast extraction algorithm was proposed based on support vector machine（SVM）.The reasonable image compression was conducted by PCA firstly.It was derived that SVM can extract road network information fast and accurately by SVM pattern recognition theory.Because hyperspectral RS image had a big amount of data and road network was very complex,we proposed the classification strategy based on 1Vm which could extract multi-class road fast and accurately at once.The new algorithm improved both efficiency and accuracy of road recognition.The experimental results show that the linear road feature recognition of SVM has better advantages than conventional method,especially for small sample road identification.The new algorithm can recognize not only linear feature of road but also its material and type.The algorithm multi-class strategy is constructed to recognize multi-class road with higher operation efficiency.

关 键 词：道路工程 高光谱遥感影像 支持向量机 特征提取 

分 类 号：U412.24[交通运输工程—道路与铁道工程]