基于特征点和SAR-SIFT算法的林权测绘地形图像几何配准  

Geometric registration of terrain images for forest right surveying and mapping based on feature points and SAR-SIFT algorithm

作　　者：舒斯红 SHU Sihong(Jiangxi Nuclear Industry Surveying and Mapping Institute Group Company Limited,Nanchang,Jiangxi 330038,China)

机构地区：[1]江西核工业测绘院集团有限公司,江西南昌330038

出　　处：《北京测绘》2024年第5期750-757,共8页Beijing Surveying and Mapping

基　　金：江西省地质局青年科学技术带头人培养计划(2022JXDZKJRC06)。

摘　　要：目前常规的测绘地形图像几何配准主要通过对单应性矩阵进行求解,从而实现图像配准,由于对待配准图像的灰度化处理程度较低,导致配准精度较差。对此,本文研究基于特征点和合成孔径雷达(SAR)-尺度不变特征变换(SIFT)算法的林权测绘地形图像几何配准方法。首先对SAR图像成像机理进行分析,在此基础上采用滤波算法对图像进行处理,并采用加权平均法对红、绿、蓝三原色(RGB)通道分量进行灰度化处理。然后结合SAR-SIFT算法,构建多层次的尺度空间,并结合海森矩阵对像素梯度值进行计算,生成特征描述子。最后通过对特征描述子的相似度进行计算,结合匹配阈值,实现图像特征点的匹配。实验结果表明:采用基于特征点和SAR-SIFT算法的林权测绘地形图像几何配准方法对图像进行几何配准时,算法的均方根误差值较小,具备较为理想的几何配准精度。At present,the conventional geometric registration methods for terrain images of surveying and mapping mainly rely on solving the homography matrix to achieve image registration.However,due to the low grayscale processing of the image to be registered,the registration accuracy is poor.Therefore,a geometric registration method for terrain images for forest right surveying and mapping based on feature points and synthetic aperture radar(SAR)-scale-invariant feature transform(SIFT)algorithm was studied.Firstly,the imaging mechanism of SAR images was analyzed.On this basis,filtering algorithms were used to process the images,and the weighted average method was used to perform grayscale processing on the red,green,and blue(RGB)channel components.Then,combined with the SAR-SIFT algorithm,a multi-level scale space was constructed,and pixel gradient values were calculated using the Hessian matrix to generate feature descriptors.Finally,by calculating the similarity of the feature descriptors and combining it with the matching threshold,image feature points were matched.The experimental results show that the geometric registration method for terrain images for forest right surveying and mapping based on feature points and the SAR-SIFT algorithm has a low root mean square error value and ideal geometric registration accuracy while performing geometric registration.

关 键 词：尺度不变特征变换(SIFT)算法 遥感图像 几何配准 配准精度 

分 类 号：P231[天文地球—摄影测量与遥感] TP231[天文地球—测绘科学与技术] TP75[自动化与计算机技术—检测技术与自动化装置]