Anisotropic turbulence identification from classifying extreme morphological changes in targets with deep-learning

In the last decade, a nascent trend of characterizing turbulence from observing features of distant targets through ground-layer turbulence have been relentless growing. Either from observing regular geometrical features of buildings or arrays of LEDs, it is possible to retrieve the structure constant of the refractive index fluctuations. On the other hand, because of the lack of a definitive theoretical model describing anisotropic or inhomogeneous turbulence, most experimental observations have been reduced to mere descriptions in the event of deviations from expected Obukhov-Kolmogorov predictions. Our group has been able to retrieve power-spectrum exponents, without a prior knowledge of a subjacent model, and henceforth determine anisotropic behavior in controlled optical turbulence; furthermore, under convective turbulence, an exponent can be obtained from time series of the occurrence of power drops in optical communication links: extreme events. In this manuscript, we present a technique identifying as extreme events suden changes in morphological characteristics of an array of point sources observed through real controlled anisotropic turbulence assisted by a deep-learnig ad-hoc. This approach provides an effective approach to reduce high-volume data from imaging targets into a real-time stream of parameters to fully characterize optical turbulence.