基于形态学的两种高光谱目标探测改进算法

发布时间：2018-03-06 07:28

本文选题：高光谱目标探测　切入点：形态学　出处：《中山大学学报(自然科学版)》2017年01期 　论文类型：期刊论文

【摘要】：基于高光谱图像的目标探测主要利用高光谱图像具有连续光谱信息的优势,通过目标与背景的光谱信息差异进行探测,空间形态信息则很少使用。然而在一些光谱分析算法的基础上充分考虑图像像元之间的空间关系能够克服这些算法存在的不足。本文针对约束能量最小化(CEM)算法采用全图信息估计背景信息、正交子空间投影(OSP)算法难以准确构造背景子空间的不足,引入目标的空间尺寸信息,然后通过形态学开运算滤掉可能存在的目标以此构造一个精确的背景,进而在统计精确背景信息的基础上使用OSP、CEM两种算法进行目标探测。最后本文采用三维受试者操作特性(3D ROC)曲线进行探测精度的评价,克服了传统2D ROC曲线评价探测精度的不足。实验结果表明,基于形态学的CEM和OSP算法有效地降低了虚警率,提升了探测效率。
[Abstract]:The main use of hyperspectral images with continuous spectral information of hyperspectral image target detection based on the advantages of detection by spectral information of target and background difference, spatial information is rarely used. However, in some spectral analysis algorithm based on considering the spatial relationship between the image pixel can overcome the shortcomings of these algorithms are based on. A constrained energy minimization (CEM) algorithm with full information estimation of background information, orthogonal subspace projection (OSP) algorithm is difficult to accurately enough tectonic background subspace, space size information into a target, and then through the morphological opening filter out the possible target to construct an accurate background, then use the OSP based on statistics precise background information on the target detection of CEM two algorithms. Finally this paper adopts three-dimensional receiver operating characteristic (3D ROC Qu Xianjin) The evaluation of line detection accuracy overcomes the shortcomings of the traditional 2D ROC curve evaluation. The experimental results show that the CEM and OSP algorithm based on morphology can effectively reduce the false alarm rate and improve the detection efficiency.

【作者单位】：成都理工大学地球物理学院;内蒙古农业大学水利与土木建筑工程学院;中国科学院遥感与数字地球研究所∥遥感科学国家重点实验室;
【基金】：国家自然科学基金(61071121) 中国科学院遥感与数字地球研究所开放基金(2013LDE006)
【分类号】：TP751

【相似文献】