基于时频特征的高分辨率遥感图像道路提取  

Road Extraction from High Resolution Remote Sensing Image Based on Time Frequency Features

作　　者：张曦[1] 胡根生[1,2] 梁栋[1,2] 鲍文霞[1,2] 

机构地区：[1]安徽大学电子信息工程学院,安徽合肥230601 [2]安徽大学计算智能与信号处理教育部重点实验室,安徽合肥230039

出　　处：《地理空间信息》2016年第6期18-21,24,共5页Geospatial Information

基　　金：国家自然科学基金资助项目(61401001);安徽省自然科学基金资助项目(1408085MF121)

摘　　要：针对传统的遥感图像道路提取算法存在错漏率较高的问题,提出一种基于时频特征提取和域自适应学习分类器的高分辨率航空遥感图像道路提取新算法。首先利用地统计学抽取道路的时域纹理特征和三维小波变换抽取道路的频域光谱特征,构成高分辨率航空图像中的道路特征。然后使用抽取的道路特征训练由迁移支持向量机模型构建的域自适应分类器,并对高分辨率航空遥感图像进行道路粗提取。最后,结合道路形态特征,用数学形态学方法处理粗分类后的道路,获得提取后的完整道路。仿真实验表明,所提方法能有效降低高分辨率航空遥感图像道路提取中非道路目标的干扰,提高道路提取的精确性。To improve the high mistake-omission rate of the widely used methods for roads classifier from high resolution remote sensing image, this paper proposed a region self-adaption algorithm which selecting time frequency features to extract roads. Firstly, we used the time domain textural feature extracted by geostatistical technique and the frequency domain spectral feature extracted by 3D wavelet transform to make up features of the road. And then, we used extracting features to train the region self-adaption classifier constructed by the transfer SVM. Finally, combining the road morphology characteristic, we used the mathematical morphology method to process the road images and extract the complete road. Simulation results show that the proposed method can enhance the precision and accuracy of road extraction from the high resolution remote sensing image.

关 键 词：高分辨率遥感影像 道路提取 地统计学 三维小波变换 迁移学习 

分 类 号：P237.3[天文地球—摄影测量与遥感]