Deep Learning Approaches to Hyperspectral and Polarimetric Scene Generation for Remote Sensing Applications
Abstract
Machine learning has demonstrated powerful solutions to many image processing and computer vision problems that cannot be easily solved with traditional linear approaches. The recent success of these nonlinear algorithms has in part been due to new and faster processing capabilities, as well as the widespread availability of image data via the internet. Machine learning techniques have been slower to proliferate into more specialized imaging modalities, such as hyperspectral imaging (HSI) and polarimetric imaging (PI), largely due to insufficient availability of application-specific data within these domains [1]. HSI and PI data demonstrating significant variability across operating conditions and target classes is difficult to acquire, particularly from aerial and spacebased platforms, due to the rigorous planning and high costs associated with such experiments. Detailed physics-based models for simulation of HSI and PI are available to some extent but impose a steep learning curve upon the user and often require substantial time investments to create detailed 3D scene models [2]. Instead, we will perform fundamental research to develop an intuitive HSI and PI 3D scene generation framework leveraging conditional generative adversarial networks (CGAN). CGANs can learn the complex statistical distributions present in scene data and use them to produce realistic synthetic data – directly impacting algorithm development, testing, and validation activities. The intention of such a framework would not be to replace real data collections or first principles modeling tools, but rather would be complementary to both. Our framework will augment real data collections by producing additional datasets that exhibit realistic scene correlation and statistics.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502110234
Entities
People
- Bradley Ratliff
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Dayton