多时相遥感图像变化检测技术研究

发布时间：2018-10-23 16:35

【摘要】：变化检测作为遥感图像分析中的一项重要应用,为环境监测、资源勘探、灾害救援与治理提供了有效的技术手段。近二十年来,遥感图像的变化检测方法虽然不断更新,但变化检测仍然受到不同因素的影响。大量的研究试图寻找各种新的遥感图像变化检测方法,但目前为止,并没有一种通用方法能够对不同条件、不同情况的应用给出完全满意的结果。针对目前现有的两时相变化检测方法存在的问题和局限性,本文进行了以下研究：首先,针对标准马尔科夫随机场(MRF)方法存在的先验能量和似然能量之间采用同样不变权重的问题,提出了一种基于自适应权重MRF模型的变化检测方法。该方法首先对于图像进行细节特征的提取,将图像中的细节特征位置进行判别。将不属于图像细节特征的位置赋予较大权重于先验能量,而将属于图像细节特征的位置赋以较小权重于先验能量。该方法首先基于8邻域线过程提取边缘像素点；然后规定了自适应权重函数(AWF)的条件,并列举了8个AWF的例子；最后对于多时相遥感图像进行了实验以验证本方法的有效性。针对标准EM参数估计方法不考虑像素间的邻域信息,易受噪声影响,参数估计不精确的问题,提出了一种基于证据理论的EM参数估计方法并将其应用于变化检测。为了能够在参数估计的过程中利用邻域信息,本文将Dempster-Shafer证据理论(DST)集成于文献[14]中的EM算法中,使每一步参数的迭代更新不仅取决于当前中心像素的亮度,还取决于其邻域像素的亮度。从而得到一种基于DST的EM方法(EEM)。为进一步提高变化检测精度,本文采用最大后验估计(MAP)标记方法对于EEM算法的结果进行MAP标记。假设差值图像的类别标记满足局部光滑条件,根据EEM算法得到的参数和初始标记,经过迭代MAP标记更新得到最终结果。实验结果表明,MAP标记方法的噪声抑制能力强于EEM。针对标准ACM模型不适用于合成孔径雷达(SAR)图像变化检测的问题,提出了一种基于广义高斯分布和活动轮廓模型的SAR图像变化检测方法。由于SAR图像通常受乘性斑点噪声的影响,传统的C-V活动轮廓模型假设图像为分段光滑,这与SAR图像数据性质相违背,因此不能直接应用于SAR图像变化检测。本文将C-V活动轮廓模型推广到广义高斯混合模型假设下,得到一种基于广义高斯分布和ACM模型的SAR图像变化检测方法,并验证了其有效性。针对通常采用的CVA方法丢失光谱特征空间信息的问题,提出了一种基于平稳小波和集成活动轮廓模型的多谱多时相遥感图像变化检测方法。该方法将光谱变化矢量特征空间看做2维黎曼流形嵌入到2+B维高维流形,其中B是光谱波段数。变化矢量图像的分割通过流形上的曲线演化,即IAC完成。IAC模型结合了测地活动轮廓模型(GAC)和无边缘活动轮廓模型(C-V模型)两者的优点,提高了检测精度。以上方法均使用模拟数据集或真实多时相遥感图像数据集进行了实验验证。实验结果表明本文提出方法的检测精度与其他主流方法相当,大部分结果优于其他主流方法。
[Abstract]:Change detection is an important application in remote sensing image analysis, which provides effective technical means for environmental monitoring, resource exploration, disaster rescue and treatment. In recent twenty years, the change detection methods of remote sensing images have been continuously updated, but change detection is still affected by different factors. A lot of research attempts to find a variety of new methods for change detection of remote sensing images, but so far, none of the common methods can give full satisfaction to different conditions and applications. In view of the existing problems and limitations of the existing two-time phase change detection method, the following research has been carried out: First, the same constant weight is used between the apriori energy and the quasi-random energy present in the standard Markov random field (MRF) method. In this paper, a change detection method based on adaptive weight MRF model is proposed The method comprises the following steps: firstly, extracting the detail feature of the image, and carrying out the detail feature position in the image; a position assigned to the image detail feature is assigned a smaller weight to a priori, The method comprises the following steps: firstly, extracting edge pixel points based on the 8-neighborhood line process; then defining the conditions of the adaptive weight function (AWF) and enumerating eight AWF examples; and finally, carrying out experiments on the multi-time-phase remote sensing image to verify the method In this paper, we propose an EM parameter estimation method based on evidence theory and its application to the problem of neighborhood information between pixels, which is susceptible to noise and inaccurate parameter estimation. In order to be able to utilize neighborhood information in the process of parameter estimation, Dempster-Shafer Evidence Theory (DST) is integrated in the EM algorithm in the literature[14], so that the iterative updating of each step depends not only on the brightness of the current center pixel, but also on its neighborhood. The brightness of the pixels, and thus a DST-based EM method is obtained. (EEM). In order to further improve the accuracy of change detection, the maximum post-test (MAP) marking method is used in this paper. MAP flag. If the category flag of the difference image satisfies the local smooth condition, the parameter and the initial mark obtained according to the EEM algorithm are updated by the iterative MAP mark. Finally, the experimental results show that the noise suppression ability of MAP marking method It is stronger than EEM. Aiming at the problem that the standard ACM model is not applicable to the image change detection of Synthetic Aperture Radar (SAR), a SAR based on generalized Gaussian distribution and active contour model is proposed. Because the SAR image is usually influenced by multiplicative speckle noise, the traditional C-V active contour model assumes that the image is piecewise smooth, which is contrary to the SAR image data property, and therefore cannot be directly applied to S. In this paper, we generalize the C-V active contour model to the generalized Gaussian mixture model, and obtain a SAR image change detection method based on the generalized Gaussian distribution and ACM model. In order to solve the problem of loss of spectral characteristic spatial information by CVA method, a multi-spectral multi-spectral model based on stationary small wave and integrated active contour model is proposed. In this method, the spectral change vector feature space is regarded as a 2 + B-dimensional high-dimensional stream. and B is the number of spectral bands. The curve evolution, i.e. the completion of the curve, incorporates both the geodynamic contour model (GAC) and the borderless active contour model (C-V model). The method has the advantages of improving the detection precision, The experimental results show that the detection precision of the method is equivalent to that of other methods.
【学位授予单位】：华中科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】