Deep Learning Enabled Sensor Protection

Abstract

Optical phase masks will be designed for the purpose of laser sensor protection by making use of end-to-end physics-based deep learning principals to significantly enhance image performance metrics that include image quality, degree of laser suppression, and fabrication constraints. The neural net will be trained against realistic variables such as noise, aberrations, broadband illumination,and laser characteristics. Both experimental and simulated images will be used in validation procedures, whereupon data, methods, and/or phase masks will be shared between RIT and collaborators at the US Naval Research Laboratory (NRL). The proposed approach isexpected provide a significantly degree of protection from laser damage compared to traditional filtering based on linear systems theory or nonlinear optical material response. This advantage is afforded by the ability of a trained neural network to reconstruct a high-fidelity image from one that is severely blurred by a purposely engineered phase mask. The proposed work will not only provide the Navy with a sophisticated solution of the 65-year-old sensor protection problem, it will also provide a general framework forapproaching the class of imaging reconstruction problems that suffer from an extremely low Strehl ratio owing to the optical system, and/or an extremely high contrast owing to illumination conditions.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 24, 2023
Source ID
N000142312513

Entities

People

  • Grover Swartzlander

Organizations

  • Office of Naval Research
  • Rochester Institute of Technology
  • United States Navy

Tags

Fields of Study

  • Physics

Readers

  • Image Processing and Computer Vision.
  • Operations Research
  • Research Science/Academic Research

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks
  • Directed Energy