报告题目：Joint reconstruction and deformation correction in 4D X-ray spectro-tomography

报告人：南方科技大学 王超助理教授

报告时间：2026年8月10日11：00-12：00

报告地点：星空体育106会议室

报告摘要：X-ray spectro-tomography is a high-dimensional imaging technique frequently compromised by non-rigid sample deformations, such as thermal drift or mechanical instabilities, which introduce severe artifacts and degrade spectral accuracy. We propose a data-driven variational framework that addresses this ill-posed inverse problem by modeling the object and its motion through implicit neural networks. By representing the continuously varying attenuation coefficients as a coordinate-based function and the motion as a differentiable warp field, we transform the traditional grid-based reconstruction into a joint optimization over network weights. This approach utilizes the physics of the Radon transform as a self-supervised regularizer, providing a continuous inductive bias that enables high-fidelity, motion-corrected spectral recovery without the need for external training data.

报告人简介：王超，南方科技大学统计与数据科学系副研究员、博士生导师，主要研究方向为图像处理、科学计算与交叉学科的数据科学。以第一作者或通讯作者身份在Cell子刊、SIAM系列、IEEE汇刊等权威期刊及CCF-A会议发表论文30余篇。入选广东省青年珠江学者、深圳市鹏城孔雀计划特聘岗位，获CVPR研讨会最佳论文奖、CSIAM年会学生论文奖，以及2次SIAM差旅奖。主持国家自然科学基金项目2项、省部级基金1项、深圳市科研项目1项。担任期刊J. Math. Imaging Vision（中国数学会T2期刊）特刊客座主编。