基于远场焦斑测量数学模型改进的CNN去噪方法研究
Research on CNN Denoising Algorithm Based on an Improved Mathematical Model for the Measurement of Far -field Focal Spot
查看参考文献17篇
|
文摘
|
针对基于纹影的高动态范围远场焦斑测量数学模型没有考虑噪声对测量结果影响的缺点,本文对基于纹影的远场焦斑测量方法从三个方面进行优化.首先,改进基于纹影的远场焦斑测量数学模型,将噪声作为影响实验结果的重要因素引入数学模型中,使该数学模型和真实的实验环境相匹配,提高了该数学模型的实用性和理论支撑作用;其次,将基于卷积神经网络的去噪算法(DnCNN)引入主瓣和旁瓣CCD图像去噪处理中,并改进该去噪算法存在的不足,使得能够有效去除主瓣和旁瓣12位图像、不同级别(0~75 dB)的噪声;最后,完整仿真了远场焦斑测量实验的全过程,包括分光、衰减、加噪声、纹影小球遮挡、去噪、衰减倍率放大、焦斑重构等,获得了有效的焦斑重构实验结果,其中重构焦斑图像和理论焦斑图像的相关系数为0.998 9,重构焦斑动态范围与理论焦斑动态范围之间误差为3.22%.仿真实验结果表明,通过该数学模型和DnCNN去噪算法的改进措施,验证了改进的数学模型必要性和DnCNN去噪算法在提高重构焦斑二维分布和动态范围精度方面的优越性能,提高了基于纹影的高动态范围远场焦斑测量的可信度,满足了高动态范围远场焦斑测量对于精度和效率的要求. |
|
其他语种文摘
|
Aim at the shortcomings that the mathematical model for the measurement of far-field focal spot with high dynamic range does not consider the influence of noise on the measurement results,this paper optimizes the measurement method of far-field focal spot based on schlieren from three aspects. Firstly,the mathematical model for the measurement of far-field focal spot based on schlieren is improved,and the noise is added to the mathematical model,which makes the mathematical model match with the real experimental environment,and improves the practicability and theoretical support of the mathematical model;Secondly,the denoising algorithm based on Convolution Neural Network(DnCNN)is used in the de-noise processing of the main lobe and side lobe CCD image,and the original denoising algorithm is improved effectively in this paper,which can remove the noise of different levels (0~75 dB) of the mainlobe and sidelobe 12-bit images;Finally,the whole experimental process of far-field focal spot measurement is simulated,including light splitting,attenuation,adding noise,schlieren sphere occlusion, denoising,attenuation magnification,focal spot reconstruction,etc.,and the effective experimental results of reconstructed focal spot is obtained,which the correlation coefficient between the reconstructed and theoretical focal spot images is 0.998 9,and the error of dynamic range between the reconstructed and theoretical focal spot is 3.22%. The simulation results show that through the improvement of the mathematical model and the DnCNN denoising algorithm,the necessity of the improved mathematical model and the superior performance of the DnCNN denoising algorithm in improving the accuracy of the two-dimensional distribution and dynamic range of reconstructed focal spot are verified. The reliability of the measurement of far-field focal spot with high dynamic range based on schlieren is improved,and the accuracy and efficiency of the measurement of far-field focal spot in high dynamic range is met in the end. |
|
来源
|
光子学报
,2020,49(12):1212001 【核心库】
|
|
DOI
|
10.3788/gzxb20204912.1212001
|
|
关键词
|
远场焦斑测量
;
纹影法
;
焦斑重构
;
DnCNN
;
去噪方法
|
|
地址
|
中国科学院西安光学精密机械研究所, 西安, 710119
|
|
语种
|
中文 |
|
文献类型
|
研究性论文 |
|
ISSN
|
1004-4213 |
|
学科
|
自动化技术、计算机技术 |
|
基金
|
国家自然科学基金
;
陕西省重点研发计划
|
|
文献收藏号
|
CSCD:6878106
|
参考文献 共
17
共1页
|
|
1.
刘维宝. 神光-Ⅲ激光装置的总体集成及探索.
国防科技,2014,34(12):30-36
|
CSCD被引
1
次
|
|
|
|
|
2.
翟学锋. 神光-Ⅲ3ω光路自动准直系统设计.
应用光学,2007,28(2):177-180
|
CSCD被引
8
次
|
|
|
|
|
3.
程娟. 纹影法测量远场焦斑实验研究.
强激光与粒子束,2006,18(4):612-614
|
CSCD被引
9
次
|
|
|
|
|
4.
王姮. 一种强辐射环境监测下图像降噪的新方法.
传感器与微系统,2011,30(11):59-61
|
CSCD被引
6
次
|
|
|
|
|
5.
王拯洲. 高动态范围激光焦斑测量数学模型研究.
光子学报,2014,43(10):1010002
|
CSCD被引
5
次
|
|
|
|
|
6.
孙跃文. 基于卷积神经网络的辐射图像降噪方法研究.
原子能科学技术,2017,51(9):1678-1682
|
CSCD被引
5
次
|
|
|
|
|
7.
Gain V. Natural Image Denoising with Convolutional Networks.
Advances in Neural Information Processing Systems
|
CSCD被引
1
次
|
|
|
|
|
8.
Divakar N. Image denoising via CNNs:an adversarial approach.
Computer Vision and Pattern Recognition Workshops,2017:1076-1083
|
CSCD被引
3
次
|
|
|
|
|
9.
Jiang J D. RedNet:residual encoder-decoder network for indoor RGB-D semantic segmentation.
Computer Vision and Pattern Recognition,2018:1806-1813
|
CSCD被引
1
次
|
|
|
|
|
10.
Xu C. MemNet:enhancing throughput and energy efficiency for hybrid workloads via para-virtualized memory sharing.
2016 IEEE 9th International Conference on Cloud Computing,2016
|
CSCD被引
1
次
|
|
|
|
|
11.
He K M. Deep residual learning for image recognition.
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2016:770-778
|
CSCD被引
681
次
|
|
|
|
|
12.
Zhang K. Beyond a Gaussian denoiser:residual learning of deep CNN for image denoising.
IEEE Transactions on Image Processing,2017,26(7):3142-3155
|
CSCD被引
390
次
|
|
|
|
|
13.
Faraji H. CCD noise removal in digital images.
IEEE Transactions on Image Processing,2006,15(9):2676-2685
|
CSCD被引
17
次
|
|
|
|
|
14.
牛志光. 近场强度分布对激光束光束质量的影响.
应用激光,2011,31(6):200-203
|
CSCD被引
1
次
|
|
|
|
|
15.
Wang Z Z. An improved schlieren method for measurement and automatic reconstruction of the farfield focal spot.
Plos One,2017,12(2):e0171415
|
CSCD被引
3
次
|
|
|
|
|
16.
秦瑜.
基于卷积神经网络的低照度图像去噪与增强研究,2019
|
CSCD被引
1
次
|
|
|
|
|
17.
王学伟. 基于疑似目标区域提取的空间图像降噪.
电光与控制,2008,15(9):52-56
|
CSCD被引
1
次
|
|
|
|
|
了解气象卫星更多信息，请访问