遥感图像森林植被纹理自适应滤波技术研究

发布时间：2018-02-10 02:26

本文关键词： 纹理滤波图像分割灰度共生矩阵局部谱直方图蓝噪声　出处：《辽宁师范大学》2016年硕士论文　论文类型：学位论文

【摘要】：森林资源是社会发展的重要物质基础,其可再生性对经济发展、社会进步和环境的稳定具有重要意义,是其它资源所不能代替的。随着遥感技术的快速发展及广泛应用,地物的纹理、形状等信息十分丰富。近年来,使用滤波技术对图像纹理特征进行纹理捕捉,已成为提高图像分割精度研究的主要方向之一。综合运用光谱信息和图像纹理信息提取森林植被纹理结构特征,已经成为当前遥感图像分割技术研究的一个重要方向。由于,遥感图像中森林植被树种多样,受地区与气候的影响,导致森林植被分割难度大。使得固定参数的纹理滤波方法不能准确表达纹理的特征,影响分割精度。因此,本文提出了一种在谱直方图统计模型下的自适应纹理滤波算法,该算法是根据分割对象的特征属性来设置纹理滤波的参数。在基于谱直方图的纹理统计模型下,先从获取典型森林植被区域入手,用基于蓝噪声特征理论探测典型森林植被区域。对典型森林植被区域进行灰度共生矩阵的纹理特征统计,得到典型森林植被纹理的基础属性,主要包括森林植被纹理基元的尺度、方向和灰度分布等。谱直方图的纹理特征提取通过多个滤波函数相结合,对图像纹理特征有选择性增强的作用。根据先验知识和典型森林植被获取到的森林植被纹理特征选择滤波函数,并自适应设置滤波参数,包括滤波器的窗口尺寸、方向、频率和强度等。对比本文方法与固定参数设置滤波参数方法的滤波效果图可以明显看出本文方法提取对象纹理的优势。图像分割的效果依赖于特征提取阶段的准确性,因此,本文方法对最后分割效果有很大的提升。此外,谱直方图的统计中积分窗口的尺度设置对分割结果也具有一定的影响,本文也通过实验证明了以分割对象的纹理基元作为积分尺度,分割过程中尽可能的避免了其它地物纹理的干扰,对提高分割精确度具有重要意义。进一步的研究工作将对典型森林植被的获取方法上进行改进,对于图像中森林植被稀疏以及森林植被面积过小的情况,对确保森林植被纹理特征更精确,采取拼接等方法来获取典型森林植被区域。
[Abstract]:Forest resources are the important material foundation of social development, and their reproducibility is of great significance to economic development, social progress and environmental stability, which can not be replaced by other resources. With the rapid development and wide application of remote sensing technology, Texture, shape and other information of ground objects are very rich. In recent years, filtering technology is used to capture texture features of images. Comprehensive use of spectral information and image texture information to extract texture features of forest vegetation has become an important research direction of remote sensing image segmentation technology. The diversity of forest vegetation species in remote sensing images results in the difficulty of forest vegetation segmentation due to the influence of region and climate. Therefore, the texture filtering method with fixed parameters can not accurately express the texture features and affect the segmentation accuracy. This paper presents an adaptive texture filtering algorithm based on the statistical model of spectral histogram, which sets the parameters of texture filtering according to the characteristic attributes of the segmented object. Firstly, the typical forest vegetation region is obtained, and the typical forest vegetation area is explored by using the blue noise feature theory. The texture characteristics of the typical forest vegetation region are analyzed by gray-scale co-occurrence matrix, and the basic properties of the typical forest vegetation texture are obtained. It mainly includes the scale, direction and grayscale distribution of forest vegetation texture elements. The texture feature extraction of spectral histogram is combined with several filtering functions. The filter function is selected according to the prior knowledge and the typical forest vegetation texture features, and the filter parameters are set adaptively, including the window size and direction of the filter. Compared with the filter effect diagram of the fixed parameter method and the fixed parameter filter parameter method, we can clearly see the advantage of this method in extracting object texture. The effect of image segmentation depends on the accuracy of feature extraction stage, therefore, the effect of image segmentation depends on the accuracy of feature extraction stage. In addition, the scale setting of the integral window in the statistics of spectral histogram also has a certain influence on the segmentation results. In this paper, it is also proved by experiments that the texture primitive of the segmentation object is taken as the integral scale, and that the texture of other objects is avoided as far as possible in the course of segmentation. It is of great significance to improve the segmentation accuracy. Further research work will improve the acquisition method of typical forest vegetation, in the case of sparse forest vegetation and small forest vegetation area in the image, To ensure that the texture features of forest vegetation are more accurate, splicing is adopted to obtain typical forest vegetation areas.
【学位授予单位】：辽宁师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP751

【相似文献】