多尺度概率图模型与SAR图像分类的研究

项目名称： 多尺度概率图模型与SAR图像分类的研究

项目编号： No.60872064

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 温显斌

作者单位： 天津理工大学

项目金额： 25万元

中文摘要： 本项目研究重点之一是如何直接对含噪SAR图像建模与分类问题。针对不同参数的含噪SAR图像分别建立了多尺度Bayes网络和多尺度Bayes相关图模型；并在完全数据下，研究这两个模型的结构学习方法、参数学习方法、推断理论和算法；研究了基于模型和算法的SAR图像分类方法和评价方法。 另一个重点研究问题是如何解决SAR图像数据量大对分类等理解带来的困难问题。研究了基于不完全数据下上述两种多尺度概率图模型结构学习和推断算法；并解决海量SAR图像处理速度和充分利用图像信息之间的矛盾。 此外，还研究了一些与上述研究相关的基础或扩展问题，如图像的去噪、配准和识别等问题。 研究小组按计划完成了规定的研究内容，达到了预期目标。

中文关键词： 多尺度Bayesian网络；多尺度Bayes相关图模型；SAR图像分类。

英文摘要： One of the importment problems we investigated in the project is modeling and classification for thr noisy SAR image. For thr different parameters SAR images, multi-scale Bayes networks and multi-scale correlation graphical model were constructed; and under complete data, structure learning method, parameter learning, inference theory and algorithms were reserched for these two models; SAR image classification and evaluation methods are gived based on these two models and algorithms. The other important problem we investigated is how to solve the difficulties caused by the large amount of classified SAR image data. The structure learning and inference algorithms for the two multi-scale probabilistic graphical model were studied based on incomplete data; and the contradiction between the massive SAR image processing speed and take advantage of the image information is solved. We also studied some the foundation or extended problems associated with research studies such as image denoising, registration and identification issues. In conclusion,our researching team has accomplished the tasks specified by the original plan and the anticipated goal has been reached.

英文关键词： Multiscale Bayes Model; Multiscale Bayes Correlation Graphical Model; SAR image Classification.