Leverage Gaussian Interpolation Features Extraction and Advanced Encode–Decode Models for 2D Steady State Flows Estimation

Accurate estimation of two-dimensional (2D) multi-obstacle steady-state flows is crucial in various scientific and engineering disciplines, yet conventional methods often fall short in precision and computational efficiency. This paper introduces a novel approach that combines Gaussian interpolation-based feature extraction with advanced encode-decode models, including AutoEncoder, UNet, and DeepLabV3+. By leveraging Gaussian interpolation and advanced Encoder-Decoder models, we effectively capture the smooth, continuous nature of steady-state flows while mitigating noise and small-scale fluctuations that may obscure global flow patterns. We evaluate our method on a private dataset of 2D multi-obstacle steady-state flows, comparing the performance of each model based on accuracy, computational efficiency, and robustness. Our results demonstrate that the encode-decode architectures, particularly UNet and DeepLabV3+, significantly improve flow estimation accuracy and feature retention. This study suggests that integrating Gaussian interpolation with advanced encode-decode models offers an efficient and effective solution for complex flow estimation tasks, opening new avenues for research in fluid dynamics applications.