ISANet: Deep Neural Network Approximating Image Sequence Assimilation for Tracking Fluid Flows

Current machine learning methods make a positive difference in the outlook on data assimilation. In this article, an efficient assimilation of image sequences that are incorporated in a deep neural network (DNN) framework is put forward. To tackle the motion estimation of fluid, the image model characterizing both the evolution of the tracers and the velocity was considered. Owing to the ability to describe nonlinear physical processes, the output of DNN substitutes deterministic information from physical laws. It will be consistent with observed images by optimizing the network parameters rather than the variable itself. The dimension of the variable to be determined will be reduced. The regularized DNN that accounts for the characteristics of the flows was employed to improve the ability of structure preservation. Realistic assimilation of tracer image sequence from the perspective of the oceanic applications shows that the proposed method is robust for extracting velocity field with vortex structures.