自适应光学系统性能分析及湍流退化图像复原方法研究

发布时间：2018-02-02 07:19

本文关键词： 自适应光学大气湍流波前复原小波域的Curvelet变换极大似然估计　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：大气湍流效应引起的大气折射率随机起伏,会导致光波波前扭曲,严重影响光学系统成像性能,降低观测目标图像分辨率。自适应光学技术是克服大气湍流最有效的办法之一,但由于自适应光学系统自身性能及噪声等因素的影响,其只能对波前误差进行部分校正,所观测到的目标图像分辨率依然与系统衍射极限相差很远,因此,对湍流退化图像进行后续复原处理是至关重要的。为了提高湍流图像分辨率,本文重点研究了湍流退化图像去噪及复原算法:首先,研究自适应光学系统。介绍自适应光学系统结构、工作及成像原理,分析了云南天文台1.2米望远镜的61单元自适应光学系统参数及性能,并设计了双校正器自适应光学系统光路图。通过分析后已知,自适应光学系统对波前畸变的校正是不充分的,需对观测到的湍流图像进行后续复原处理。其次,研究波前复原技术及湍流波前相位屏模拟。研究Zernike多项式法波前复原技术,利用Zernike多项式法模拟了不同项数下大气湍流随机相位屏的湍流分布情况。从模拟结果可知,Zernike多项式项数较低时,湍流屏上缺乏高频成分,而低频成分突出,相位相对平稳;随着Zernike多项式项数的增加,其湍流屏上的湍流分布逐渐变缓,相位平稳度下降。再次,研究基于小波域的Curvelet变换法的湍流退化图像去噪算法。将多尺度几何变换的思想引入到湍流图像去噪中,首先介绍小波域的Curvelet变化原理,然后根据贝叶斯准则,改进了阈值的自适应选取方法,提取二维小波变换高频系数并对其进行基于Wrapping的Curvelet变换,最终实现了小波域的Curvelet变换法对湍流退化图像的去噪处理。实验结果表明,该算法处理后的湍流图像视觉效果清晰、相比于原始图像PSNR明显提高,MSE明显降低。最后,研究改进极大似然估计法的大气湍流图像复原算法。首先对湍流图像进行了去噪处理,然后利用Zernike多项式对湍流退化图像进行波前复原,根据波前信息建立了点扩散函数初值估计模型,接着结合图像先验信息,引入目标边缘保持约束及点扩散函数约束改进了ML算法,在迭代求解目标图像估计和点扩散函数F估计过程中,对点扩散函数进行了频域带宽约束,最终利用该算法对模拟湍流图像及实际观测湍流图像进行了复原处理。实验结果表明,该算法可以有效提高湍流复原图像分辨率,复原后图像的灰度平均梯度有明显提高,图像能量更加集中。
[Abstract]:The random fluctuation of atmospheric refractive index caused by atmospheric turbulence effect will lead to the distortion of optical wavefront and seriously affect the imaging performance of optical system. Adaptive optics is one of the most effective methods to overcome atmospheric turbulence, but due to the influence of the performance of adaptive optical system and noise and other factors. It can only partially correct the wavefront error and the resolution of the observed image is still far from the diffraction limit of the system. In order to improve the resolution of turbulent images, the de-noising and restoration algorithms of turbulent degraded images are studied in this paper. The structure, operation and imaging principle of adaptive optics system are introduced. The parameters and performance of the 61 unit adaptive optics system of the 1.2-meter telescope of Yunnan Observatory are analyzed. The optical circuit diagram of adaptive optical system with double corrector is designed. It is known by analysis that the correction of wavefront distortion by adaptive optics system is not sufficient, and it is necessary to carry out follow-up restoration of the observed turbulence image. Secondly. The wave front restoration technique and turbulent wavefront phase screen simulation are studied. The Zernike polynomial wave front restoration technique is studied. The Zernike polynomial method is used to simulate the turbulent distribution of atmospheric turbulent random phase screen under different term numbers. From the simulation results, it can be seen that the number of Zernike polynomials is low. There is no high frequency component on the turbulence screen, but the low frequency component is prominent and the phase is relatively stable. With the increase of Zernike polynomial terms, the turbulence distribution on the turbulent screen becomes slower and the phase stability decreases. The denoising algorithm of turbulent degraded image based on Curvelet transform in wavelet domain is studied, and the idea of multi-scale geometric transformation is introduced into the denoising of turbulent image. Firstly, the principle of Curvelet variation in wavelet domain is introduced, and then the adaptive threshold selection method is improved according to Bayesian criterion. The high frequency coefficients of two-dimensional wavelet transform are extracted and the Curvelet transform based on Wrapping is carried out. Finally, the Curvelet transform in wavelet domain is implemented to remove the noise of the turbulent degraded image. The experimental results show that the visual effect of the proposed algorithm is clear. Compared with the original image, the PSNR is significantly improved. Finally, the improved maximum likelihood estimation (MLE) algorithm for atmospheric turbulence image restoration is studied. Firstly, the turbulent image is de-noised. Then the Zernike polynomial is used to restore the wavefront of the turbulent degenerate image, and the initial value estimation model of the point diffusion function is established according to the wavefront information, and then the prior information of the image is combined. The ML algorithm is improved by introducing the edge preserving constraint and the point diffusion function constraint. In the iterative solution of the target image estimation and the point diffusion function F estimation, the point diffusion function is constrained in frequency domain bandwidth. Finally, the proposed algorithm is used to reconstruct the simulated and observed turbulent images. The experimental results show that the proposed algorithm can effectively improve the resolution of turbulent restored images. After restoration, the average grayscale gradient of the reconstructed image is obviously improved, and the image energy is more concentrated.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;O439

【相似文献】