摘要
发展了一种基于逆卷积神经网络的图像级重建方法用于聚变等离子体辐射分布的断层反演。通过引入结构相似度(SSIM)作为损失函数,该方法在模拟数据实验中表现出了较好的重建效果。模拟实验结果表明,在弦积分信号噪声强度为10%、15%及20%时,该方法的重建结果依然具有良好的精确度和鲁棒性。
A deconvolutional neural network(deCNN)which allows to implement a pixel-wise emission distribution reconstruction is developed for plasma radiation distribution tomographic reconstruction.By adopting the structural similarity(SSIM)as the loss function,the method reveals better reconstruction accuracy in phantom data experiments.The simulated experiments results show that this method could still achieve an admissible reconstruction precision and is noise robustness when the line-integrated noise magnitude is 10%,15%and 20%.
作者
王朝辉
肖建元
秦宏
WANG Zhao-hui;XIAO Jian-yuan;QIN Hong(University of Science and Technology of China Department Physics,Hefei 230026)
出处
《核聚变与等离子体物理》
CAS
CSCD
北大核心
2021年第3期207-214,共8页
Nuclear Fusion and Plasma Physics
基金
国家自然科学基金(11775219,11575186,11805273)
中国博士后科学基金资助项目(2017LH002)。
关键词
等离子体辐射分布断层重建
逆卷积神经网络
弦积分诊断系统
图像重建
Plasma radiation distribution tomography
Deconvolutional neural networks
Line-integrated diagnostic system
Image reconstruction
