基于奇异值分解的航空遥感图像小目标提取方法  

Small Target Extraction Method for Aerial Remote Sensing Images Based on Singular Value Decomposition

作　　者：申晓平 SHEN Xiaoping(Changzhou Vocational Institute of Industry Technology,Changzhou 213000,China)

机构地区：[1]常州工业职业技术学院,江苏常州213000

出　　处：《计算机测量与控制》2024年第9期262-268,共7页Computer Measurement &Control

摘　　要：针对航空遥感图像中小目标提取受低秩噪声干扰导致的精度下降和漏检问题,提出了一种基于奇异值分解的提取方法;该方法利用奇异值分解准则,结合不均匀变化奇异值特征向量,有效提取小目标并计算奇异值变换能量增益;在此基础上,构建信号空间杂波的协方差矩阵,以反映信号分布及信号间联系;通过奇异值分解矩阵,避免计算杂波协方差矩阵的影响,准确反映小目标的形状、大小、纹理等信息;进一步分解图像矩阵,获取行列像素强度信息,并通过正交矩阵分解和重建图像矩阵,实现图像压缩;将图像分为分散、完全叠加和部分叠加目标三部分,计算其能量衰减倍数,完成小目标提取;实验结果显示,该技术具有高召回率和准确率,船类小目标最大漏检量为4只,验证了其精准高效的提取效果。To address the accuracy reduction and miss detection caused by low-rank noise interference in the small target extraction of airborne remote sensing images,a extraction method based on singular value decomposition is proposed.This method utilizes singular value decomposition criteria,combined with unevenly changing singular value eigenvectors,to effectively extract small targets and calculate the energy gain of singular value transformation.Based on this,the covariance matrix of clutter in signal space is constructed to reflect the distribution of signals and the connection between signals.Through the singular value decomposition of the matrix,the impact of calculating the clutter covariance matrix is avoided,accurately reflecting the shape,size,texture,and other information of small targets.Furthermore,the image matrix is decomposed to obtain the intensity information of row and column pixels,and the image matrix is compressed through the orthogonal matrix decomposition and reconstruction.The image is divided into three parts:dispersed,completely superimposed,and partially superimposed targets,and the energy attenuation factor is calculated to achieve the extraction of small targets.Experimental results show that this technique has high recall rate and accuracy,with a maximum miss detection of 4 for ship-like small targets,validating its precise and efficient extraction.

关 键 词：奇异值分解 航空遥感图像 小目标提取 能量衰减倍数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]