【作者】 刘娜；

【导师】 乔双；

【作者基本信息】 东北师范大学 ， 电路与系统， 2017， 硕士

【摘要】 中子成像技术是一种重要的无损检测技术,这种技术是利用中子穿透物质后所形成的图像来显示物体的内部细节以及结构等信息。鉴于中子的独特性质,中子照相技术与传统的无损检测相比具有无可替代的作用,尤其是在检测较厚金属壳内部包含的氢、锂等轻元素方面具有更加突出的优势。在中子成像过程中将会受到由于环境因素或自身因素等原因造成的噪声污染。主要包括中子束存在微观的量子特性以及随机性,CCD摄像机自身受到环境的影响,γ射线的污染以及暗电流产生的噪声。这些噪声的统计特征近似于泊松-高斯混合分布,严重影响了中子图像质量以及中子照相技术在无损检测方面的准确度。因此中子图像去噪具有十分重要的研究价值。为了去除中子图像中的这类混合噪声,本文提出将Anscombe非线性方差稳定化变换引入PPB加权极大似然估计进行去噪处理。首先,对图像数据进行Anscombe非线性方差稳定化变换,使符合泊松-高斯分布的混合噪声转化为单一的近似高斯分布的噪声。经过这种变换,能够降低去除中子图像中混合噪声的难度。其次,利用PPB加权极大似然估计对变换后的图像进行去噪,这种方法的优点在于不但可以有效的去除高斯噪声,还能克服现有去噪算法对图像带来的伪影现象,有助于改善中子图像质量并提高中子照相在无损检测中的准确度。最后利用Anscombe无偏逆变换对数据进行还原,得到去噪后的中子图像。利用PSNR对模拟图像实验结果进行客观评价,结合真实中子图像实验进一步验证了本文提出的Anscombe非线性方差稳定化变换与PPB加权极大似然估计相结合的去噪方法,可以有效的去除中子图像中含有的泊松-高斯混合噪声。更多还原

【Abstract】 Neutron radiography technology is an important nondestructive testing technology,which uses the image of neutron penetrating material to display the internal details and structure information of an object.In view of neutron’s unique properties,neutron radiography technology has the irreplaceable function compared with traditional nondestructive testing,especially to detect internal light elements of thicker metal casing like hydrogen,lithium,etc.Neutron imaging will be affected by noise pollution caused by various factors.It mainly includes the quantum properties and randomness of neutron beam,CCD camera influenced by environment,the gamma ray pollution and the dark current noise.The statistical characteristics of these noises are similar to mixture distribution of Poisson and Gaussian,which seriously affects neutron image quality and the accuracy of neutron radiography technique in nondestructive testing.Therefore,it is very important to research neutron image denoising.In order to remove the mixed noises in neutron image,in this paper,we introduce the Anscombe nonlinear variance stabilizing transformation into PPB weighted maximum likelihood estimation for denoising.Firstly,transforming the image data through Anscombe nonlinear variance stabilizing transformation,the mixed noises submitted to Poisson-Gaussian distribution are transformed into a single noise obeyed Gaussian distribution.Through this transformation,it reduces the difficulty of removing Poisson noise in neutron image.Secondly,PPB weighted maximum likelihood estimation is used to filter the transformed image,which not only is able to remove the Gaussian noise effectively,but also overcome artifacts in existing denoising algorithms,It is helpful to improve neutron image quality and increase the accuracy of neutron radiography in nondestructive testing.Finally,the data is restored by the Anscombe unbiased inverse transform to obtain the restored neutron image.The peak signal to noise ratio(PSNR)is used to evaluate denoising results in simulated image,with the real neutron image,further to verify the denoising method which combining the Anscombe nonlinear variance stabilizing transformation with PPB weighted maximum likelihood estimation that it is effective to remove mixed noise of Poisson and Gaussian in neutron image.更多还原