SBIR/STTR Award attributes
The Submarine Force is phasing out conventional optical periscopes in favor of digital sensor masts that electronically transmit video to the Control Center. Digital imagery still suffers from distortions due to atmospheric turbulence caused by humidity and temperature gradients near the ocean surface. While systems for turbulence correction exist, they have largely focused on static scenes and rely on precise image alignment of multiple frames to produce clean images. However, submarine imagery comes from a moving camera capturing a moving target, in a moving scene making alignment infeasible, and results must be available quickly (i.e., the system cannot wait for multiple frames before producing output). Therefore, existing mechanisms for correcting blur and loss of contrast cannot be effectively applied to turbulence correction for submarine imagery. To address these shortcomings, we propose Turbulence Mitigation, Error Reduction, and Increased Contrast (TURMERIC), a software system that uses deep learning to correct for atmospheric turbulence in imagery such that watchstanders and automated detection algorithms can optimally detect surface contacts. Specifically, TURMERIC uses a generative adversarial network (GAN), a powerful new deep learning architecture, to intelligently de-blur and enhance the contrast of photonics mast imagery using only a single frame.
