多时相遥感图像融合去噪方法研究

发布时间：2018-08-07 19:34

【摘要】：遥感技术的出现,使我们能不与研究对象直接接触,通过传感设备来获取观察对象的基本信息。这就避免了一些偏远或险峻的地区信息无法取得的情况,成为至今为止全球范围内动态观测数据的唯一方式,被广泛应用到多个领域,对经济的增长和社会的发展起着很大的催化作用。然而,由于受天气、遥感设备及传输介质的影响,遥感图像在成像和传输的过程中,往往会受到很多噪声的影响,其中最为常见的噪声为高斯噪声、云噪声和雾噪声等。这些噪声的存在,将直接影响遥感图像的进一步处理、分析及应用,影响数据的使用价值。遥感图像去噪的目标在于在保护图像细节信息的前提下,最大限度地去除噪声,提高数据的可读性与有效性。目前,对于热噪声、散粒噪声等高斯噪声的处理,主要是针对单幅遥感图像,利用噪声在空间域或频域的特征,对遥感图像进行降噪处理。但这类去噪方法存在一个问题,即保留图像边缘与去除噪声的矛盾,往往会出现图像边缘信息被过度扼杀,造成边缘模糊或去除噪声不理想现象。针对云噪声,对于薄云,由于它不仅包含了与云相关的信息,还包含了地物等有效信息,对它的研究也比较多,常用的处理方式是削弱云信息,同时增强地物信息,使地物清晰。而对于厚云,由于地物信息被完全遮盖,几乎不含有用信息,使用单幅遥感图像去除厚云往往会引起信息空洞。这说明单幅遥感图像的信息量不足,需要将不同时间同一地区具有互补信息的多时相遥感数据根据一定的方法,有效的结合起来,得到一幅信息量更多的遥感图像。针对以上分析,本文研究了基于DS(Dempster-Shafer)证据理论的多时相遥感图像融合去噪方法,主要从以下3个方面展开:(1)分析了遥感图像中多类噪声的特点与研究现状,并分析了DS证据理论在多时相遥感图像融合去噪的可行性:DS证据理论作为一种推理理论,属于人工智能的范畴,它能融合多个证据并做出决策,对推理给出合理的阐释,可以有效解决由于对研究对象认知的不准确或认知缺失所造成的不确定性问题。遥感图像中,噪声具有随机性与不确定性,而DS证据理论能综合考虑来自多源的不确定信息,同样适合用在多时相遥感图像融合去噪过程中。(2)提出了基于DS证据理论的多时相遥感图像融合去除高斯噪声的方法,根据DS证据理论的基本原理,为获取证据的基本概率分配,设计四个高斯噪声检测模型,即两状态高斯混合模型、均值检测模型、中值检测模型、边缘分析模型,用于分析每个灰度值与噪声相关还是与地物相关。然后根据DS证据理论融合规则,将各幅遥感图像四个证据融合成一个整体,得到每幅遥感图像各像素与噪声相关或与地物相关总的证据。接着利用DS证据理论将多时相遥感图像的多个证据合成,得到最终结论。最后根据所得的结论与决策规则,对遥感图像进行去噪处理。实验结果表明,该算法在高斯噪声去除、图像边缘保持等方面优于传统的单幅遥感图像去噪算法,图像方差、信噪比和视觉效果方面都有所改进。(3)提出了基于DS证据理论的多时相遥感图像融合去除云噪声的方法,根据DS证据理论的基本原理,为获取证据的基本概率分配,设计两个云噪声检测模型,分别依据灰度统计值变化和频域信息变化。首先将多时相遥感图像按同样的标准分割成若干小区域,每个小区域按照以上两个模型,判断每个区域与云相关还是与地物相关。然后根据DS证据理论合成规则,将各幅遥感图像两个证据融合成一个整体,得到每幅遥感图像各小区域与云相关或与地物相关总的证据。接着利用DS证据理论将多时相遥感图像的多个证据合并,得到最终结论。最后根据所得的结论与决策规则,对遥感图像进行融合去云。实验结果表明,该算法在云噪声去除方面,通过利用有效互补信息,得到了信息更加丰富的图像。
[Abstract]:The emergence of remote sensing technology makes it possible for us to get the basic information of the observation objects without direct contact with the research objects. This avoids the information that the remote or steep regional information can't obtain. It has become the only way to date the dynamic observation data in the world so far, and has been widely used in many fields. However, because of the influence of weather, remote sensing equipment and transmission medium, remote sensing images are often affected by a lot of noise in the process of imaging and transmission. The most common noise is Gauss noise, cloud noise and fog noise. The existence of these noises will be direct. The further processing, analysis and application of remote sensing images affect the use value of the data. The target of remote sensing image denoising is to remove the noise and improve the readability and effectiveness of the data on the premise of protecting the details of the image. At present, the processing of Gauss noise, such as thermal noise and granular noise, is mainly aimed at the single. The remote sensing image is used to denoise the remote sensing image by using the characteristics of noise in space or frequency domain. However, there is a problem in this kind of denoising method, that is, to retain the edge of the image and to remove the noise, the edge information of the image is often excessively stifled, causing edge paste or removing noise is not ideal. Because it contains not only the information related to the cloud, but also the effective information such as ground objects, it also has more research on it. The common processing method is to weaken the cloud information, and to enhance the information of the ground, and make the ground objects clear. For thick clouds, because the information of the ground is completely covered, almost no useful information is contained, and a single remote sensing image is used. In addition to the thick cloud, it often causes information void. This shows that the information of single remote sensing images is insufficient. It is necessary to combine the multi temporal remote sensing data with complementary information at different time and the same area according to a certain method to get a more remote sensing image. Afer) the multi temporal remote sensing image fusion denoising method of evidence theory is mainly carried out from the following 3 aspects: (1) analyzing the characteristics and research status of multi class noise in remote sensing images, and analyzing the feasibility of DS evidence theory in multi temporal remote sensing image fusion denoising: DS evidence theory is a kind of reasoning theory, which belongs to the category of artificial intelligence. It can integrate a number of evidence and make decisions and give a reasonable explanation to the reasoning, which can effectively solve the uncertainty caused by the inaccuracy or lack of cognition of the research objects. In remote sensing images, the noise is random and uncertain, and the DS evidence theory can consider the uncertain information from multiple sources. In the process of multi temporal remote sensing image fusion de-noising. (2) a method of multi temporal remote sensing image fusion based on DS evidence theory is proposed to remove Gauss noise. According to the basic principle of DS evidence theory, the basic probability distribution of evidence is obtained, and four Gauss noise detection models are designed, that is, the two state Gauss mixture model and the mean detection model. Type, median detection model, edge analysis model, which are used to analyze the correlation of each gray value to noise or related to ground objects. Then, according to the fusion rules of DS evidence theory, four evidence of each remote sensing image is fused into a whole, and the total evidence of each pixel and noise related to or related to the ground objects is obtained in each remote sensing image. Then the DS evidence is used. In the theory, the multi temporal remote sensing images are synthesized and the final conclusion is obtained. Finally, the remote sensing image is de-noised according to the conclusions and the decision rules. The experimental results show that the algorithm is superior to the traditional single amplitude remote sensing image denoising algorithm, image variance, signal to noise ratio and view in Gauss noise removal and image edge preservation. The sense effect has been improved. (3) a method of multi temporal remote sensing image fusion based on DS evidence theory is proposed to remove cloud noise. According to the basic principle of DS evidence theory, the basic probability distribution of evidence is obtained, and two cloud noise detection models are designed, according to the change of gray level statistics and the change of frequency domain information respectively. Remote sensing images are divided into small areas according to the same standard. Each area is based on the above two models to determine whether each region is related to the cloud or the ground objects. Then, according to the DS evidence theory, the rules are synthesized and the two evidence of each remote sensing image is fused into a whole, and each area of the remote sensing image is related to or with the cloud. General evidence related to things. Then, using the DS evidence theory, multiple evidence of multi phase remote sensing image is merged and the final conclusion is obtained. Finally, the remote sensing image is fused to cloud based on the conclusions and the decision rules. The experimental results show that the algorithm can get more information through the use of effective complementary information in the removal of cloud noise, and the information is more abundant. A rich image.
【学位授予单位】：上海海洋大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】