Machine Learning Algorithm for Calculating Wavefronts from a Single Coherent Beam Return

The accuracy of future military electro-optical (EO) sensors for intelligence, reconnaissance, and surveillance (ISR), laser beam control, seeker systems and space surveillance, that operate along turbulent atmospheric lines-of-sight, can be enhanced using coherent wavefront sensing. The intervening atmospheric turbulence and vehicle wake distort the sensor observation, reducing target contrast and image resolution. A promising approach to real-time detection is single-shot digital holography (SSDH), in which a laser beam illuminates the target, and the return is processed to produce a reconstruction of the wavefront and a turbulence corrected image. The processing includes optical interference with a reference laser, retrieving the wavefront, and applying the wavefront to the return image to mitigate both atmospheric and laser speckle effects. It uses signal estimation methods to remove turbulence and speckle effects and correct the image. One of the current limitations to advancement of SSDH is the computational burden of the current physics-based Bayesian estimation approach to processing, which would require that the vehicle carry a massive on-board computer capability to produce real-time results. Spectral Sciences, Inc., in collaboration with researchers at New Mexico State University, propose to replace current computational approaches with a machine learning algorithm to speed up the processing by orders of magnitude. Our algorithm training will be informed by modeling of the sensor, vehicle wake, atmosphere, and potential targets. In Phase I, we will construct the algorithm, test it to assess its speed and performance with modeled inputs, and consider scalability. In Phase II, we will further develop the algorithm, demonstrate its performance in an optical breadboard SSDH system, and define a computational system for on-board operations.