论文标题
部分可观测时空混沌系统的无模型预测
A non-convex variational model for joint polyenergetic CT reconstruction, sensor denoising and material decomposition
论文作者
论文摘要
储层计算是预测湍流的有力工具，其简单的架构具有处理大型系统的计算效率。然而，其实现通常需要完整的状态向量测量和系统非线性知识。我们使用非线性投影函数将系统测量扩展到高维空间，然后将其输入到储层中以获得预测。我们展示了这种储层计算网络在时空混沌系统上的应用，该系统模拟了湍流的若干特征。我们表明，使用径向基函数作为非线性投影器，即使只有部分观测并且不知道控制方程，也能稳健地捕捉复杂的系统非线性。最后，我们表明，当测量稀疏、不完整且带有噪声，甚至控制方程变得不准确时，我们的网络仍然可以产生相当准确的预测，从而为实际湍流系统的无模型预测铺平了道路。
Computed Tomography (CT) is widely used in engineering and medicine for imaging the interior of objects, patients, or animals. If the employed X-ray source is monoenergetic, image reconstruction essentially means the inversion of a ray transform. Typical X-ray sources are however polyenergetic (i.e. emit multiple wavelengths, each with different attenuation behaviour), and ignoring this fact may lead to artefacts such as beam hardening. An additional difficulty in some settings represents the occurrence of two different types of noise, the photon counting effect on the detector and the electronic noise generated e.g. by CCD cameras. We propose a novel variational image reconstruction model that takes both noise types and the polyenergetic source into account and moreover decomposes the reconstruction into different materials based on their different attenuation behaviour. In addition to a detailed mathematical analysis of the model we put forward a corresponding iterative algorithm including its convergence analysis. Numerical reconstructions of phantom data illustrate the feasibility of the approach.
