遥感图像亚像元定位方法的研究

发布时间：2019-07-07 20:42

【摘要】：随着遥感技术的迅猛发展,遥感图像已被广泛的应用在环境/资源管理、自然灾害监测、农业/植被规划、公共安全等技术领域。然而,遥感图像在数据获取过程中受环境参数和传感器分辨等因素影响,使得混合像元不可避免的存在。混合像元的存在限制了遥感图像的空间分辨率。空间分辨率受限为土地覆盖信息的提取带来了极大的困难。因此,如何提高遥感图像的空间分辨率成为遥感领域学者研究的热点问题之一。光谱解混技术虽然能够获得混合像元内部各地物类别所占的比例,但却无法知道各地物类别在混合像元内具体的空间分布情况。而亚像元定位正是一项确定混合像元内各地物类别具体空间分布的技术,它使得地物空间分部信息在更高尺度下显示。基于上述内容,本文在悉心学习并总结现有研究成果的基础上,对遥感图像亚像元定位展开了深入的研究,主要研究内容如下:1、在现有亚像元/亚像元空间引力模型的亚像元定位(SSSAM)的基础上对模型参数进行分析。通过对已有的SSSAM的学习与分析可知,在SSSAM方法中,距离权值函数是描述空间相关性的关键所在。不同的距离权值函数将从不同的角度对空间相关性进行诠释,进而影响着亚像元定位的精度。文中分别将常见的三种距离权值函数:距离倒数模型、指数模型和高斯模型应用到SSSAM中,并通过两组实验对采用不同距离权值函数的SSSAM方法的定位效果进行分析与比较,从而选择出最优的距离权值函数。实验结果表明:距离倒数模型定位效果最差,而指数模型与高斯模型相比,高斯模型定位精度略高,但其对模型参数变化较敏感。2、提出了一种基于立方卷积插值算法的亚像元定位方法。亚像元定位过程可以描述成以下两个步骤:第一,对低分辨率图像进行超分辨率,得到具有概率隶属度信息的高分辨率图像;第二,按照概率值大小,来确定亚像元定位的最终结果,即亚像元尺度下的“硬分类”结果。基于这一思想,文中利用立方卷积插值算法对低分辨率图像进行超分辨率,再对超分辨率结果进行亚像元级别上的“硬分类”,从而得到亚像元尺度上的地物分类图。实验表明:该方法具有较好的定位精度,同时,无需训练样本和迭代计算,是一种简单易实现的方法。3、提出了一种基于混合插值算法的亚像元定位方法。利用传统经典的插值算法进行亚像元定位的方法虽然可行,但插值算法本身的边缘模糊效应不可避免的存在。为了避免图像插值过程的边缘模糊效应对亚像元定位精度的制约,设计了一种新的插值算法,即对偶插值,并将其分别与双线性插值算法和反距离权值插值算法结合,形成两种不同的混合插值算法。利用两种不同的混合插值算法分别对低分辨率分量图进行超分辨率,得到高空间分辨率下各亚像元属于某一地物类别的概率值,再根据分量值约束信息进行亚像元级别上的“硬分类”,获得亚像元定位的最终结果。实验结果表明:与应用单一插值算法相比,利用混合插值算法进行亚像元定位能够更好地保持图像的边缘特性,获得更高的定位精度。本文主要对三种亚像元定位方法进行研究。首先,将不同的距离权值函数应用到SSSAM中,通过实验为SSSAM选择最佳的距离权值函数;然后,将传统的图像插值算法应用于亚像元定位中,为亚像元定位技术的研究开辟了一个可行的新途径,该方法在不需要训练样本和迭代计算的情况下实现亚像元定位;而基于混合插值算法的亚像元定位是在上一种新方法的基础上提出的,该方法利用高低分辨率图像之间的空间对偶特性来克服传统插值算法所固有的边缘模糊效应,进一步提高了亚像元定位的精度。三种方法从不同的角度改善亚像元定位精度,对遥感图像的应用有着极其重要的意义。
文内图片：

图片说明：图１．１遥感图像中混合像元的处理流程逡逑目前为止，科研工作者己尝试了多种光谱解混的方法，，如线性光谱解混模型［３］、基逡逑
[Abstract]:With the rapid development of remote sensing technology, remote sensing images have been widely used in the fields of environment/ resource management, natural disaster monitoring, agriculture/ vegetation planning and public safety. However, the remote sensing image is affected by environmental parameters and sensor resolution in the data acquisition process, so that the mixed image element is inevitable. The existence of the mixed image element limits the spatial resolution of the remote sensing image. The limitation of spatial resolution brings great difficulty to the extraction of land cover information. Therefore, how to improve the spatial resolution of the remote sensing image is one of the hot issues in the field of remote sensing. The spectral unmixing technique can obtain the proportion of the object categories in the mixed image elements, but it is not possible to know the specific spatial distribution of the object categories in the mixed image elements. And the sub-pixel location is a technique for determining the specific spatial distribution of the object class in the mixed image element, which makes the space segment information of the figure space be displayed at a higher scale. Based on the above-mentioned contents, this paper, on the basis of learning and summarizing the existing research results, has carried out an in-depth study on the location of the sub-image elements of the remote sensing image. The main contents of this study are as follows:1. The model parameters are analyzed on the basis of the sub-pixel location (SSSAM) of the existing sub-pixel/ sub-pixel space gravity model. It can be seen from the study and analysis of the existing SSSAM that the distance weight function is the key to describing the spatial correlation in the SSSAM method. The different distance weights function will interpret the spatial correlation from different angles, thus affecting the accuracy of the sub-pixel location. In this paper, the common three distance weight functions: the distance reciprocal model, the exponential model and the Gaussian model are applied to the SSSAM, and the positioning effect of the SSSAM method with different distance weight functions is analyzed and compared by two groups of experiments, so that the optimal distance weight function is selected. The experimental results show that the positioning effect of the distance reciprocal model is the worst, and the exponential model is slightly higher than that of the Gaussian model, but it is sensitive to the variation of the model parameters. the sub-pixel positioning process can be described as two steps: first, the super-resolution of the low-resolution image is performed to obtain a high-resolution image with the probability membership information; secondly, the final result of the sub-pixel positioning is determined according to the probability value size, I. e., the "hard classification" results at the sub-pixel scale. Based on this idea, the super-resolution of the low-resolution image is carried out by the cubic convolution interpolation algorithm, and the "hard classification" of the sub-image element level is carried out on the super-resolution result, so as to obtain the ground object classification map on the sub-image element scale. The experiment shows that the method has better positioning accuracy, and is a simple and easy-to-implement method without training samples and iterative calculation. Although the traditional interpolation algorithm is feasible, the edge-fuzzy effect of the interpolation algorithm is inevitable. In order to avoid the restriction of the edge-fuzzy effect of the image interpolation process on the positioning accuracy of the sub-image element, a new interpolation algorithm, that is, the dual interpolation, is designed and combined with the bilinear interpolation algorithm and the inverse distance weight interpolation algorithm to form two different mixed interpolation algorithms. the method comprises the following steps of: respectively carrying out super-resolution on a low-resolution component image by using two different mixed interpolation algorithms to obtain a probability value of each sub-image element belonging to a certain property class at high spatial resolution, and then carrying out "hard classification" at the sub-image element level according to the component value constraint information, The final result of the sub-pixel location is obtained. The experimental results show that the edge characteristics of the image can be better preserved by using the mixed interpolation algorithm compared with the application of a single interpolation algorithm, and the higher positioning accuracy is obtained. In this paper, three kinds of sub-pixel location methods are studied. First, the different distance weight functions are applied to the SSSAM, and the best distance weight function is selected for the SSSAM through the experiment; then, the traditional image interpolation algorithm is applied to the sub-image element positioning, and a feasible new way is opened for the research of the sub-image element positioning technology, the method realizes the sub-pixel location without the need of training samples and iterative calculation, and the sub-pixel location based on the mixed interpolation algorithm is proposed on the basis of the previous novel method, The method uses the space dual property between high and low resolution images to overcome the edge fuzzy effect inherent in the traditional interpolation algorithm, and further improves the accuracy of the sub-pixel positioning. It is of great significance for the application of remote sensing image to improve the positioning accuracy of the sub-image element from different angles.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】