Deep learning-based deflection angle regression network for measurement of the atmospheric refractive index structure constant

Le résumé fourni par la source

The measurement of the refractive index structure constant plays a crucial role in the study of atmospheric turbulence. This paper proposes a method based on deflection angle regression network (DARN) for measuring the atmospheric refractive index structure constant. First, the feasibility and superiority of this method are verified through numerical simulation. The dataset used for deep learning training consists of 3600 frames of moiré fringes and a label dataset composed of 3600 frames of deflection angle information in one-to-one correspondence with them. Then, the trained model is used to predict the deflection angle, based on which the refractive index structure constant can be obtained. The simulation results show that the relative error of this method is about 0.3%, while that of the traditional method (TM) is about 2.1%. Finally, it is applied to the measurement of real atmosphere flow fields by moiré deflectometry. The experimental results indicate that the deep learning model not only achieves minimal training loss and rapid convergence of validation loss but also maintains a negligible prediction error, confirming its excellent performance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.