高空间分辨率卫星图像的薄云去除研究

发布时间：2018-01-23 02:40

本文关键词： 卫星图像 Mallat算法薄云多尺度分析非线性增强中值滤波　出处：《中国科学技术大学》2017年硕士论文　论文类型：学位论文

【摘要】：高分卫星自投入使用以来,被广泛的应用于灾害监测、资源勘查以及环境保护等许多领域。而高分辨率卫星数据使得在小的空间尺度上面进行地表细节变化的观察以及完成人为活动对环境影响的检测等变为可能,具有重要的意义。由于搭载在卫星上的高空间分辨率成像设备获取到的遥感图像数据会受到云的干扰,数据质量存在不同程度的下降。当天空中存在厚云遮挡的时候,下垫面信息会完全丢失;而在薄云覆盖区,图像的质量虽然会退化,但仍有可供利用的下垫面信息。为了提高图像定量解译的水平和图像信息的利用率,有效地去除高分辨卫星图像中薄云的影响,本文针对图像的薄云去除进行了研究。论文的主要内容以及结论如下:(1)首先对我们要处理图像的类型以及表示方法进行了介绍,然后对常规遥感图像的退化模型以及受薄云影响的图像的成像模型进行了总结,并从空间特征和频率特征这两个方面对遥感图像云区的特征进行了分析;(2)提出本文所用的方法,对图像作Mallat小波分解得到高频细节部分和低频近似部分,依据云噪声在分解系数中处于低频部分而地物信息占据相对高频部分的特点,在多尺度分析的基础上,算法在最大尺度低频图像上按照云厚度掩模值对云区进行线性处理;对于高频子带图像根据尺度的不同运用非线性增强算子进行不同程度的增强,从而提高图像的清晰度,最后对经过重构后的图像作中值滤波以减少高频云的影响。针对高分一号卫星图像进行了试验,试验证明该方法能够取得比较好的效果。(3)设计薄云去除软件,在功能上实现了 TIFF图像的读取、显示、保存、子图像截取、小波分解与重构、2种薄云去除的方法以及退出程序等操作并给出了相应的处理结果图。(4)以高分一号卫星数据为例进行试验并将该方法与传统小波变换法进行比较分析,,该方法在除薄云的同时很好的保持了图像细节信息,去薄云效果优于传统小波变换法,论证了本文方法的有效性。
[Abstract]:Since the high score satellite put into use, is widely used in many fields of disaster monitoring, resource exploration and environmental protection. The high resolution satellite data that were observed in the surface details of the changes above small spatial scale and influence of human activities on the environment to complete the detection of possible, is of great significance. Because the remote sensing image the data with high spatial resolution imaging equipment on the satellite access to the cloud will be subject to interference, has the low quality of the data. In the sky when there is a thick cloud cover, surface information will be completely lost; and covered in thin cloud area, although the image quality will be degraded, but there are still under the mat the information available. In order to improve utilization level of image information and image quantitative interpretation of the rate, effectively remove the influence of high resolution satellite images in the cloud, according to the image Thin cloud removal was studied. The main contents and conclusions are as follows: (1) first said methods are introduced for us to deal with the types of images and then the degradation model of conventional remote sensing images and imaging model image by thin cloud effects are summarized, and the characteristics of the remote sensing image from the cloud area these two aspects of spatial features and frequency characteristics are analyzed; (2) the method used in this paper, the Mallat wavelet decomposition to obtain high frequency details and low frequency approximation part of the image, based on the cloud noise in the low frequency part of the wavelet coefficient and the feature information occupy a relatively high frequency part characteristics, based on multiscale analysis on the algorithm in maximum scale low frequency image according to the thickness of cloud mask value the cloud region linear processing; the high-frequency subband images according to different nonlinear scale enhancement operator is not With the enhancement degree, so as to improve the clarity of the image, finally through the reconstructed image median filtering to reduce the influence of high-frequency cloud. For a high score satellite images of the test, the experiment proves that the method can achieve better results. (3) software to remove thin cloud design, realizes the function of reading TIFF, image display, storage, image capture, wavelet decomposition and reconstruction, 2 methods of thin cloud removal and exit procedures and gives the processing results of the corresponding graph. (4) to score the 1st satellite data as an example of the test and the method with the traditional wavelet transform method were compared and analyzed. This method, in addition to the cloud while preserving the image details, to better than the thin cloud effect of traditional wavelet transform method, demonstrate the effectiveness of the proposed method.

【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】