遥感图像区域变化检测技术的研究

发布时间：2018-04-10 19:50

本文选题：遥感图像 + 图像去噪　；参考：《北京交通大学》2014年硕士论文

【摘要】：近年来,遥感图像的区域变化检测技术在国民经济和国防建设中发挥了重要的作用。本文围绕着变化检测技术中一些关键技术进行了研究,主要涉及：1)遥感图像去噪；2)彩色遥感图像分割和3)遥感图像分类识别。由于气象的原因,在遥感图像成像过程中有时会混有云雾噪声。区域性云雾的存在会严重影响遥感图像的判读和分析。本文在考察含云遥感图像中地物信息和噪声信息的频率分布特点的基础上,提出了基于小波变换及HSI颜色空间的遥感图像云雾噪声去除的算法。此算法在较大程度上保留遥感图像中有用信息的基础上,较好的去除了遥感图像中存在的云雾噪声。遥感图像分割是实现区域变化检测的前提条件之一。分割质量的好坏,决定着区域变化检测的成败。本文基于统计学原理利用统计区域合并算法对彩色遥感图像进行了多尺度分割。针对分割过程中存在的“过分割”问题,本文结合分割后的区域的LBP特征及边缘特征,提出了一种基于阈值的区域合并算法。良好的图像分类识别技术,是区域变化检测结果准确的保障。文章对传统的视觉词袋模型(Bag-Of-Visual-Words, BOVW)算法中“单词”分配步骤进行了分析研究。针对传统的BOVW算法中“单词”在“硬分配”过程中存在的问题,采用了具有鲁棒性的“软分配”方法。利用此方法,可以提高遥感图像的分类识别率。传统的BOVW算法在计算时间上消耗巨大。本文在对传统的BOVW算法进行较为深入研究的基础上,提出的Fast BOVW算法。此算法主要通过改进视觉词典建立的方法,来达到提高计算速度的目的。通过在实际的数据库上的实验表明,本文提出的Fast BOVW算法在保证分类精度前提下能够将计算速度提高L(本文为20～30倍)左右。本项研究取得的相关结果对提高遥感图像区域发生变化的处理能力有一定的指导和借鉴意义。
[Abstract]:In recent years, regional change detection technology of remote sensing images has played an important role in national economy and national defense construction.This paper focuses on some key technologies of change detection, including: 1) denoising of remote sensing image (2) segmentation of color remote sensing image and 3) classification and recognition of remote sensing image.Due to meteorological reasons, cloud noise is sometimes mixed in remote sensing image imaging.The existence of regional cloud and fog will seriously affect the interpretation and analysis of remote sensing images.Based on the investigation of the characteristics of frequency distribution of ground object information and noise information in remote sensing images containing cloud, an algorithm for removing cloud noise from remote sensing images based on wavelet transform and HSI color space is proposed in this paper.On the basis of preserving the useful information in remote sensing image to a large extent, this algorithm can remove the cloud and fog noise in remote sensing image.Remote sensing image segmentation is one of the prerequisites for regional change detection.The quality of segmentation determines the success or failure of regional change detection.Based on the principle of statistics, the multi-scale segmentation of color remote sensing images is carried out by using the statistical region merging algorithm.Aiming at the problem of "over-segmentation" in the segmentation process, this paper proposes a threshold-based region merging algorithm combining the LBP features and edge features of the segmented regions.Good image classification and recognition technology is the guarantee of accurate regional change detection results.In this paper, the steps of word allocation in the traditional visual word bag model (Bag-Of-Visual-WordsBOVW) are analyzed and studied.Aiming at the problem of "word" in traditional BOVW algorithm in "hard assignment", a robust "soft assignment" method is adopted.By using this method, the classification and recognition rate of remote sensing images can be improved.The traditional BOVW algorithm consumes a lot of computation time.Based on the deep research of the traditional BOVW algorithm, this paper proposes the Fast BOVW algorithm.This algorithm can improve the speed of calculation by improving the method of visual dictionary building.The experiments on the actual database show that the proposed Fast BOVW algorithm can improve the computing speed by about 30 times (20 ~ 30 times in this paper) on the premise of ensuring the classification accuracy.The results obtained in this study are helpful to improve the processing ability of remote sensing images.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】