高分辨率遥感图像感兴趣目标的提取算法  

A Extracting Scheme for Targets of Interest in High Resolution Remote Sensing Imagery

作　　者：周伟[1,2] 关键[1,2] 张国华[2] 

机构地区：[1]海军航空工程学院电子信息工程系,山东烟台264001 [2]中国航空工业洛阳电光设备研究所火力控制技术国防科技重点实验室,河南洛阳471009

出　　处：《光电工程》2011年第2期115-121,共7页Opto-Electronic Engineering

基　　金：国家自然科学基金项目(60672140;60802088);航空科学基金项目(20095184004)

摘　　要：在遥感图像的判读和解译过程中,快速准确地提取感兴趣的目标可以极大地改善整个处理系统的负担。借鉴人类的注意机制,提出了一种基于视觉注意的感兴趣目标提取算法。该方法首先通过分析图像的谱残差求取图像显著图,再利用全局竞争机制和返回抑制策略控制注意焦点快速搜索显著图,实现无先验知识的显著区域提取;处于焦点的大部分显著区域利用感兴趣目标可能尺寸和形状等信息就可以剔除,否则,再利用感兴趣目标和非感兴趣目标在目标特征子空间上投影系数的差异,对其作进一步鉴别,最终实现感兴趣目标的检测。利用大量高分辨率遥感图像对本文方法进行了实验分析,结果表明了算法的有效性。During the analysis and interpretation of remote sensing image, it can greatly relieve the processing system of the heavy computing burden to extract targets of interest rapidly and correctly. According to the attention mechanism of human, a detection algorithm for targets of interest based on visual attention mechanism was proposed. Firstly a saliency map was computed by analyzing the spectral residual, and then salient target can be extracted without a priori knowledge by searching through the saliency map with a control mechanism of win-take-all competing and inhibition-of-return. A majority of salient regions were discarded by using the possible bound of size and shape for target of interest. Otherwise the focused regions were projected to the target feature subspace, and the difference in the projection coefficients can be used to determine whether the region is of interest or not. The proposed scheme is tested using an amount of high resolution remote sensing images, and the results show that it is effective.

关 键 词：目标检测 视觉注意 显著图 特征子空间 高分辨率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN911.73[自动化与计算机技术—计算机科学与技术]