Imaging through turbulence with improved point spread model in a compressive sensing system

Unmanned systems are becoming the preferred choice in sensing the environment, both in air and underwater. Strict requirement associated with these platforms calls for more effective use of power, sensing elements, control and data storage, in order to reduce the size and weight. Compressive sensing approach is one of the emerging methods that has the potential to gain in many of aforementioned elements. Previous work has shown that a compressive line sensing imager can perform efficiently in underwater hybrid scattering environment, where both particle and turbulence influence are present. In previous study, generic turbulence properties were used to model the light scattering by turbulence in a similar fashion to those in the atmosphere. In this study, we examine a more direct approach where both the turbulent kinetic energy dissipation (TKED) and temperature variance dissipation (TD) rate are taken into account in formulating the point spread function. The Simulated Turbulence and Turbidity Environment (SiTTE) facility at NRL has been utilized for the test. The unique convective turbulence tank is capable of generating and maintaining mixing intensities at desired values mimicking the inertial subrange of the Kolmogorov spectrum. The 5m long path length allows flexible adjustment to the turbidity and turbulence conditions necessary for imaging and communication studies. To avoid large fluctuations at the walls and top/bottom heat exchangers, only the central regions (0.2x0.2m) of the tank cross-section (0.5x0.5m) are used for image acquisition. Both measurements from the physical tank and integrated values from high-resolution solution of the numerical tank are used in estimation of the dissipation rates. Various turbulence intensities were examined, with corresponding TKED and TD ranging from 10^-7 to 10^-8 W/kg and 10^-4 to 10^-5 C²/s, respectively. These parameters are used in the simple underwater imaging model (SUIM) to derive the corresponding spread functions, which is then incorporated into the reconstruction of compressed line sensing imagery. We present the comparison between these different choices and discuss the dependencies of each parameter, and sensitivities to the imaging outcome.