Research on algorithm for restoration of large aperture and thick pinhole imaging based on neural network
-
摘要: 为了更好地获取低强度辐射源空间分布图像，提出一种使用神经网络算法将大孔径厚针孔退化图像复原的方法。建立了孔径5 mm、10 mm、15 mm的厚针孔模型，获得了3600个汉字形状辐射源的厚针孔退化图像集。基于DnCNN神经网络模型，建立了大孔径厚针孔退化图像复原神经网络，并与维纳滤波、Lucy-Richardson这些传统算法进行了比较。在考虑噪声影响后，利用迁移学习理论，对原神经网络模型进行迁移训练，再对含噪大孔径厚针孔退化图像进行复原。神经网络算法复原的RMSE明显低于传统方法，迁移学习显著减小了噪声的影响。证明了神经网络算法在大孔径厚针孔退化图像复原领域的优越性，并验证了神经网络方法复原含噪大孔径厚针孔退化图像的可行性。Abstract: To obtain the spatial distribution image of low intensity radiation source better, a method is proposed to restore large aperture thick pinhole degraded image using neural network algorithm. The thick pinhole model of 5 mm, 10 mm and 15 mm apertures is established, and the degenerate image sets of thick pinhole for the shape radiation source of 3600 Chinese characters are obtained. Based on the DnCNN neural network model, the neural network for image restoration with large aperture and thick pinhole is obtained, and compared with traditional algorithms such as Wiener filter and Lucy Richardson. After considering the influence of noise, the original neural network model is trained by means of transfer learning theory, and then the degraded image of large aperture pinhole with noise is restored. The RMSE of neural network algorithm is significantly lower than that of the traditional one, and the effect of noise is greatly improved by transfer learning. This paper proves the superiority of neural network algorithm in the field of image restoration with large aperture and thick pinhole, and verifies the feasibility of neural network method to restore the large aperture thick pinhole degraded image with noise.
-
Key words:
- large thick aperture /
- neural network /
- image reconstruction /
- transfer learning
-
图 7 10mm孔径厚针孔退化图像添加噪声
Figure 7. 10 mm aperture thick pinhole degraded image adds noise
表 1 测试数据平均RMSE
Table 1. RMSE of test data
aperture/mm average RMSE 5 16.6067 10 30.4662 15 35.3384 
下载: 导出CSV
表 2 测试数据平均RMSE
Table 2. Comparison of average RMSE of test data
aperture/mm average RMSE Wiener filtering Lucy-Richardson neural network 5 46.654 6 48.316 8 16.781 9 10 48.873 6 50.161 2 31.144 2 15 50.613 4 50.845 5 36.294 7 注：加粗字体为每行最优值。
下载: 导出CSV
-
[1] Berzins G J, Han K S. Pinhole imaging of a test fuel element at the Transient Reactor Test Facility[J]. Nucl Sci Eng, 1978, 65(1): 29-40. [2] Bohlin H, Brack F E, M Erveňák, et al. Radiative characterization of supersonic jets and shocks in a laser-plasma experiment[J]. Plasma Physics and Controlled Fusion, 2021, 63: 045026. doi: 10.1088/1361-6587/abe526 [3] 李勤, 王毅, 刘云龙, 等. 针孔法测量X光源焦斑尺寸[J]. 强激光与粒子束, 2021, 33:044007. (Li Qin, Wang Yi, Liu Yunlong, et al. X-ray spot size measurement with pinhole[J]. High Power Laser and Particle Beams, 2021, 33: 044007 doi: 10.11884/HPLPB202133.200132 [4] 宋顾周, 朱宏权, 韩长材, 等. 杆箍缩二极管X射线焦斑的测量[C]//中国核科学技术进展报告——中国核学会2009年学术年会论文集. 2009: 7Song Guzhou, Zhu Hongquan, Han Changcai, et al. X-ray spot measurement for rod-pinch diode radiographic source[C]//Progress Report on China Nuclear Science & Technology — Proceedings of the 2009 Annual Conference of the Chinese Nuclear Society. 2009: 3519-3525 [5] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014. [6] He K , Zhang X , Ren S , et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016: 770-778. [7] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[M]. IEEE Computer Society, 2014. [8] Kai Z, Zuo W, Chen Y, et al. Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising[J]. IEEE Transactions on Image Processing, 2016, 26(7): 3142-3155. [9] 姚志明, 段宝军, 马继明, 等. 大孔径厚针孔成像数值模拟研究[J]. 原子能科学技术, 2019, 53(2):379-384. (Yao Zhiming, Duan Baojun, Ma Jiming, et al. Numerical simulation of large thick aperture imaging[J]. Atomic Energy Science and Technology, 2019, 53(2): 379-384 doi: 10.7538/yzk.2018.youxian.0294 [10] Pan S J, Qiang Y. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10): 1345-1359. doi: 10.1109/TKDE.2009.191 [11] 付晓峰, 吴俊, 牛力. 小数据样本深度迁移网络自发表情分类[J]. 中国图象图形学报, 2019, 24(5):93-101. (Fu Xiaofeng, Wu Jun, Niu Li. Classification of small spontaneous expression database based on deep transfer learning network[J]. Journal of Image and Graphics, 2019, 24(5): 93-101 [12] 汤鹏杰, 谭云兰, 李金忠. 融合图像场景及物体先验知识的图像描述生成模型[J]. 中国图象图形学报, 2017, 22(9):1251-1260. (Tang Pengjie, Tan Yunlan, Li Jinzhong. Image description based on the fusion of scene and object category prior knowledge[J]. Journal of Image and Graphics, 2017, 22(9): 1251-1260 doi: 10.11834/jig.170052 期刊类型引用(5)
1. 董大鹏，于信，汪逸群，潘国涛，白素平. 0.05～0.1 MPa宽气压环境下校准光源光学系统设计. 光学精密工程. 2023(08): 1124-1135 . 
百度学术2. 王华，王玲维，黄汉云. 基于大数据的低照度微弱点目标激光成像方法. 激光杂志. 2023(09): 182-187 . 百度学术3. 陶宗慧，刘唯奇，陈亚楠，倪小龙，娄岩，刘显著，姜会林. 大气信道激光通信系统光束偏振特性. 兵工学报. 2022(03): 481-488 . 百度学术4. 王姝懿，刘智，林鹏，刘树通，刘艺. 基于液晶空间光调制器的激光束散角控制技术. 液晶与显示. 2022(11): 1430-1438 . 百度学术5. 李伟，朱敏，左常玲. 低信噪比环境下光斑图像信号的定位优化研究. 重庆科技学院学报(自然科学版). 2022(05): 70-74 . 百度学术其他类型引用(3)
-
点击查看大图
图(8) / 表(2)
计量
- 文章访问数: 843
- HTML全文浏览量: 400
- PDF下载量: 49
- 被引次数: 8
