高光谱影像亚像元目标检测方法研究
发布时间：2018-11-07 19:25
【摘要】：近年来，随着高光谱成像技术的发展，高光谱影像的目标检测算法日益受到关注。高光谱影像光谱分辨率较高，且具有图谱合一的特点，每一个波段形成一幅图像，同一像元的不同波段连起来又形成一条连续的光谱曲线，这使得其在目标检测中有独特的优势。另一方面，由于传感器空间分辨率有限，加上地物分布复杂，有时会出现目标的大小小于一个像元，和多种地物混合在一起的情况，这种情况下的目标称为亚像元目标。混合像元的光谱与目标差别较大，难以识别。这使得亚像元目标检测成为高光谱信息处理的难点问题，加上大气、光照等条件的影响，更加大了目标检测的难度。 在高光谱影像亚像元检测时，现有的方法多假设影像的噪声服从多元正态分布。然而，该假设并不完全符合高光谱影像中的实际情况。在对高光谱图像中噪声分布模型进行了仔细研究后，我们发现噪声梯度的分布同样符合多元正态分布。为了提高亚像元检测的检测效果，有必要将该先验引入目标检测算法。本文针对高光谱亚像元检测问题展开研究，提出了新的目标检测算法。本文的主要研究内容如下：(1)根据噪声梯度分布的先验，对现有的线性混合模型进行改进，提出了混合梯度模型；(2)将本文提出的混合梯度模型引入统计假设检验框架，提出两种混合梯度探测器，分别对应结构背景假设和非结构背景假设。与传统的高光谱亚像元目标检测算法相比，，本文提出的算法有如下优点：(1)对噪声分布的描述更符合实际情况，因此抗干扰性较强；(2)改进了像元混合模型，对亚像元目标有更好的检测效果。
[Abstract]:In recent years, with the development of hyperspectral imaging technology, the target detection algorithm of hyperspectral image has been paid more and more attention. The spectral resolution of hyperspectral image is high, and it has the characteristic of unifying the spectrum. Each band forms an image, and the different bands of the same pixel form a continuous spectral curve. This makes it have a unique advantage in target detection. On the other hand, due to the limited spatial resolution of sensors and the complex distribution of ground objects, sometimes the target size is smaller than one pixel and mixed with a variety of ground objects. In this case, the target is called sub-pixel target. The spectrum of the mixed pixel is different from that of the target, so it is difficult to recognize. This makes subpixel target detection become a difficult problem in hyperspectral information processing, combined with the influence of atmosphere, illumination and other conditions, which increases the difficulty of target detection. In hyperspectral image subpixel detection, the existing methods assume that the noise of the image is multivariate normal distribution. However, this assumption does not fully conform to the actual situation in hyperspectral images. After a careful study of the noise distribution model in hyperspectral images, we find that the noise gradient distribution also accords with the multivariate normal distribution. In order to improve the detection effect of subpixel detection, it is necessary to introduce the priori into the target detection algorithm. In this paper, a new target detection algorithm is proposed to solve the problem of hyperspectral subpixel detection. The main contents of this paper are as follows: (1) according to the prior noise gradient distribution, the existing linear mixed model is improved and the mixed gradient model is proposed. (2) the mixed gradient model proposed in this paper is introduced into the statistical hypothesis test framework, and two kinds of hybrid gradient detectors are proposed, which correspond to the structural background hypothesis and the non-structural background hypothesis, respectively. Compared with the traditional hyperspectral sub-pixel target detection algorithm, the proposed algorithm has the following advantages: (1) the description of the noise distribution is more in line with the actual situation, so the anti-jamming is stronger; (2) the mixed pixel model is improved, and the detection effect of sub-pixel target is better.
【学位授予单位】：中国科学院研究生院（西安光学精密机械研究所）
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41
本文编号：2317342
[Abstract]:In recent years, with the development of hyperspectral imaging technology, the target detection algorithm of hyperspectral image has been paid more and more attention. The spectral resolution of hyperspectral image is high, and it has the characteristic of unifying the spectrum. Each band forms an image, and the different bands of the same pixel form a continuous spectral curve. This makes it have a unique advantage in target detection. On the other hand, due to the limited spatial resolution of sensors and the complex distribution of ground objects, sometimes the target size is smaller than one pixel and mixed with a variety of ground objects. In this case, the target is called sub-pixel target. The spectrum of the mixed pixel is different from that of the target, so it is difficult to recognize. This makes subpixel target detection become a difficult problem in hyperspectral information processing, combined with the influence of atmosphere, illumination and other conditions, which increases the difficulty of target detection. In hyperspectral image subpixel detection, the existing methods assume that the noise of the image is multivariate normal distribution. However, this assumption does not fully conform to the actual situation in hyperspectral images. After a careful study of the noise distribution model in hyperspectral images, we find that the noise gradient distribution also accords with the multivariate normal distribution. In order to improve the detection effect of subpixel detection, it is necessary to introduce the priori into the target detection algorithm. In this paper, a new target detection algorithm is proposed to solve the problem of hyperspectral subpixel detection. The main contents of this paper are as follows: (1) according to the prior noise gradient distribution, the existing linear mixed model is improved and the mixed gradient model is proposed. (2) the mixed gradient model proposed in this paper is introduced into the statistical hypothesis test framework, and two kinds of hybrid gradient detectors are proposed, which correspond to the structural background hypothesis and the non-structural background hypothesis, respectively. Compared with the traditional hyperspectral sub-pixel target detection algorithm, the proposed algorithm has the following advantages: (1) the description of the noise distribution is more in line with the actual situation, so the anti-jamming is stronger; (2) the mixed pixel model is improved, and the detection effect of sub-pixel target is better.
【学位授予单位】：中国科学院研究生院（西安光学精密机械研究所）
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41
【参考文献】
相关期刊论文 前1条
1 杨哲海,韩建峰,宫大鹏,李之歆;高光谱遥感技术的发展与应用[J];海洋测绘;2003年06期
相关博士学位论文 前1条
1 杜博;高光谱遥感影像亚像元小目标探测研究[D];武汉大学;2010年
本文编号：2317342
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2317342.html
