基于自适应权值差异图融合和聚类的SAR图像变化检测方法研究

发布时间：2018-05-11 14:49

本文选题：变化检测 + 图像融合　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：研究同一地区不同时相的两种或多种SAR(Synthetic Aperture Radar,SAR)图像之间发生的变化叫做SAR图像变化检测,可以被应用于社会、经济、军事等许多方面。近几年,由于SAR图像具备全天候、全天时的工作能力以及对地物具有一定的穿透能力和成像覆盖面积大等一系列优点,使得SAR图像成为变化检测的主要数据来源。本论文对SAR图像变化检测中差异图的分析和差异图的构造等方面进行了探索,所取得的主要研究成果为:1.提出了一种基于自适应权值的差异图融合策略,并用于SAR图像变化检测。利用所提出的基于自适应权值的差异图融合策略,将对数比值差异图与直接比值差异图进行融合,再对融合后的差异图进行kmeans聚类,获得最后的变化检测结果。该方法不仅使得融合参数能够根据不同差异图的特性自适应调整,而且考虑了邻域灰度信息与空间信息,提高了变化检测精度和算法的抗噪性能。2.提出了一种基于自适应权值图像融合及PCA特征提取的SAR图像变化检测方法。根据自适应权值,将直接比值差异图和对数比值差异图进行图像融合。再对融合后的差异图用PCA方法进行特征提取,从而获得每个像素对应的特征向量,并组成特征空间矩阵。将k-means聚类方法对于PCA提取的特征,对原始数据进行特征提取,获得特征空间矩阵,再进行聚类,该方法通过非线性映射可以较好地辨别、提取并放大有用的特征,获得更为精准的聚类,使变化检测的错误率进一步降低。使得抗噪能力增强,总错误数减少,提高了变化检测的精度。3.提出了一种基于马尔科夫随机场模糊聚类的SAR图像变化检测方法。该方法将马尔科夫随机场模型应用到模糊c均值聚类中,提高了在马尔科夫随机场模型中所用的能量函数的最小二乘模型方法拟合的精度,并改进了模糊聚类的目标函数,在原始目标函数中加入了KL(Kullback-Leibler)信息正则化,使聚类更为准确,从而提高了算法的变化检测精度与抗噪性能。本论文工作得到国家自然科学基金(No.61003199)、中央高校基本科研业务费专项资金(Nos.JB140216和K5051202019)资助。
[Abstract]:The study of the changes between two or more SAR(Synthetic Aperture Radarar images in the same area is called SAR image change detection, which can be applied to many aspects, such as society, economy, military and so on. In recent years, SAR images have become the main data source for change detection because of their all-weather, all-day working ability and a series of advantages such as the ability to penetrate objects and the large imaging coverage. In this paper, the analysis of the difference map and the construction of the difference map in the SAR image change detection are explored. The main research results are: 1: 1. A differential image fusion strategy based on adaptive weights is proposed and applied to SAR image change detection. Using the adaptive weight based differential map fusion strategy, the logarithmic ratio difference map and the direct ratio difference map are fused, and then the difference map after fusion is clustered by kmeans, and the final change detection results are obtained. This method not only adaptively adjusts the fusion parameters according to the characteristics of different difference images, but also takes into account the neighborhood gray level information and spatial information, and improves the accuracy of change detection and the anti-noise performance of the algorithm. A change detection method for SAR images based on adaptive weight image fusion and PCA feature extraction is proposed. According to the adaptive weights, the direct ratio difference map and the logarithmic ratio difference map are fused. Then the fused difference map is extracted by PCA method, and the corresponding feature vectors of each pixel are obtained, and the feature space matrix is formed. The k-means clustering method is used to extract the features extracted from the PCA, the original data is extracted, the feature space matrix is obtained, and then the clustering is carried out. The method can identify, extract and enlarge the useful features by nonlinear mapping. More accurate clustering was obtained, and the error rate of change detection was further reduced. The ability to resist noise is enhanced, the total number of errors is reduced, and the accuracy of change detection is improved by 3. 3. A change detection method for SAR images based on Markov random field fuzzy clustering is proposed. In this method, Markov random field model is applied to fuzzy c-means clustering, which improves the fitting accuracy of the least square method of energy function used in Markov random field model, and improves the objective function of fuzzy clustering. In the original objective function, KLU Kullback-Leibler information regularization is added to make the clustering more accurate, thus improving the change detection accuracy and anti-noise performance of the algorithm. The work of this paper is supported by the National Natural Science Foundation No. 61003199, and the special funds for basic scientific research operating expenses of central colleges and universities, namely, Nos.JB140216 and K5051202019.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】