基于NSCT及张量分解的遥感建筑边缘检测技术  

Remote sensing image building edge detection technology based on NSCT and tensor decomposition

作　　者：吴颖斌 周明全 耿生玲[1,3] 张丹 WU Yingbin;ZHOU Mingquan;GENG Shengling;ZHANG Dan(School of Computer Science,Qinghai Normal University,Xining 810008,China;School of Mathematics and Information Technology,Yuncheng University,Yuncheng 044000,China;State Key Laboratory of Tibetan Intelligent Information Processing and Application,Xining 810008,China)

机构地区：[1]青海师范大学计算机学院,西宁810008 [2]运城学院数学与信息技术学院,山西运城044000 [3]藏语智能信息处理及应用国家重点实验室,西宁810008

出　　处：《扬州大学学报（自然科学版）》2024年第4期56-64,共9页Journal of Yangzhou University：Natural Science Edition

基　　金：国家重点研发计划资助项目(2020YFC1523300);国家自然科学基金资助项目(62102213)。

摘　　要：常见的边缘检测及语义分割技术在遥感图像中获取的建筑物边缘不够精细,且不能反映出建筑物屋顶的细节信息。针对上述问题,提出基于非下采样轮廓波变换(nonsubsampled contourlet transform,NSCT)与张量分解相结合的边缘检测技术,以获取包括屋顶在内的建筑物细节边缘特征。首先,利用NSCT进行图像分解,得到不同尺度和角度的精细子带系数;其次,对子带系数进行位置编码,得到相应位置的二阶对称张量,再将同一位置的不同尺度、不同角度的张量进行加权求和完成特征融合;最后,根据谱理论进行张量分解,得到图像的边缘特征。实验结果表明:与双向级联网络(bi-directional cascade network,BDCN)等5种方法相比,所提方法的峰值信噪比(peak signal-to-noise ratio,PSNR)和结构相似性(structural similarity,SSIM)指标均优于其他方法,其中PSNR和SSIM较基于深度学习的BDCN方法分别提升1.20和0.03,且检测结果能够更准确细致地反映出建筑物的边界及屋顶的边缘信息,可对建筑物的类型和风格研判提供更好的支持。The edges of buildings obtained in remote sensing images by common edge detection and semantic segmentation techniques are not fine enough and can not reflect the details of the building roof.To solve these problems,an edge detection technique based on Non Subsampled Contourlet Transform(NSCT)combined with tensor decomposition is proposed to obtain the edge features of building detaileds including roof.Firstly,NSCT is used to decompose the image to obtain the subband frequency information of different scales and angles.Secondly,these subband coefficients are encoded to obtain second-order symmetric tensors of corresponding positions,and then the tensors of different scales and angles at the same position are weighted and summed to complete the feature fusion.Finally,the edge features of the image are obtained by tensor decomposition according to spectral theory.The experimental results show that compared with other five detection algorithms such as BDCN(Bi-Directional Cascade Network),the index of peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)of the proposed method is superior to other methods.Compared with BDCN method based on deep learning,PSNR and SSIM increase by 1.20 and 0.03 respectively,and the detection results can more accurately and meticulously reflect the boundary information of buildings and the edge information of roofs,which can provide better support for the classification and style analysis of buildings.

关 键 词：遥感图像 建筑物 边缘检测 非下采样轮廓波变换 张量分解 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]