量子电子学报2024，Vol.41Issue(5)：738-751,14.DOI:10.3969/j.issn.1007-5461.2024.05.004
一种利用光子图像解决深度学习中子CT缺乏成对数据的方法
A method to solve lack of paired data in neutron computed tomography for deep learning by using photon images
摘要
Abstract
Due to the lack of high-quality paired datasets,the application and development of deep learning in neutron computed tomography(CT)reconstruction are severely hindered.Although the imaging principles of neutron CT and photon CT are both based on the Radon transform,the imaging characteristics of the two processes during particle transport are different,so the network trained for photon CT cannot be directly used to solve the reconstruction problem of neutron CT.Therefore,in this work,an unsupervised domain adaptive network is proposed that can solve the probability distribution difference problem in the migration process from photon tomography to neutron tomography.In the proposed method,the maximum mean difference is introduced to reduce the distribution difference between photon and neutron tomography image features,and furthermore,wavelet transform and convolution neural network are combined to enhance the effective features of reconstruction.The comparison experiments with other algorithms show that the proposed method can reconstruct high-quality neutron tomography images from low-flux neutron tomography results,effectively alleviating the degradation of low-flux neutron tomography.关键词
图像处理/中子CT重建/域适应迁移学习/稀疏层析Key words
image processing/neutron computer tomography reconstruction/domain adaptive transfer learning/sparse tomography分类
信息技术与安全科学引用本文复制引用
郭虎,陈帅,杨明翰,张子恒,邵慧,汪建业..一种利用光子图像解决深度学习中子CT缺乏成对数据的方法[J].量子电子学报,2024,41(5):738-751,14.基金项目
安徽省自然科学基金(2108085QF285),合肥市自然科学基金(2021003),安徽省古建筑智能感知与高维建模国际联合研究中心开放课题(GJZZX2021KF04) （2108085QF285）
