21-000000238, Machine Learning-based Turbulence Analysis and Mitigation for Hyperspectral Imaging
Abstract
Hyperspectral imaging (HSI) has significant potential for defense tasks that include target detection, recognition and identification, shape extraction, classification and material characterization. For remote sensing tasks, HSI is commonly applied in nadir (downlooking) ground survey applications with aircraft or satellite. However, our interest is imaging along horizontal or slant paths near the ground for applications that require both high spectral and spatial resolution. In this situation, atmospheric turbulence is a significant degrading factor for spatial and spectral resolution as it induces image blur and spectral mixing between objects. Increasingly, machine learning (ML) algorithms have become essential signal analysis tools for HSI image cube datasets due to their prediction capabilities, ability to estimate the statistics of classes of interest, and speed of operation. However, only a few ML algorithms have been proposed in the literature for addressing both blur and spectral mixing in HSI data. Our intent is to apply ML algorithms to exploit the diversity provided by both the spectral and spatial data to aide in image deblurring and unmixing.The objectives of the proposed two-year base effort are to develop ML algorithms to support the analysis of atmospheric turbulence effects on HSI and advance these tools for the mitigation of turbulence effects in these images. The scenarios of interest involve a wide spectral sensing range (e.g. 300 nm up to 10 m wavelength) and imaging over horizontal or slant paths of a few hundred meters to several kilometers that are relatively near the Earths surface.The base effort of our approach has three tasks: (1) acquisition of HSI datasets and generation of synthetic datasets via numerical wave optics simulation, (2) development of ML algorithms that are applied in a forward sense to model the effects of turbulence on the HSI imagery and (3) development of ML algorithms for estimation and mitigation of the spatially and spectrally varying blur field in HSI data. The first task provides the initial HSI data to support the two ML development tasks. The second task is the development of ML algorithms to synthesize new HSI data. This effort will provide timely generation of HSI data sets and the understanding gained from the work will assist in the development of the algorithms for thethird task. The third task is the development of ML algorithms for blind deblurring and un-mixing of the image without complete knowledge of the object or blurring parameters.The objectives for the three option years are to extend the temporal capabilities of theML algorithms and apply the algorithms to the analysis of future HSI datasets corresponding to the scenarios of interest. There is an expectation that a field data collection and experiment activity will be established through other programs and we propose to assist these efforts through consultation and onsite evaluation and, ultimately, integrate our ML algorithms into the field data processing pipeline.The anticipated outcome of this work is the development of highly efficient ML-based algorithms for synthesizing HSI data and for mitigating blur and spectral mixing in HSI due to atmospheric turbulence. The impact of our work will be to improve the performance and capabilities of future Navy HSI systems for applications such as small boat detection, detection and identification of UAVs, characterization of littoral regions, and identification tasks for directed energy systems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 05, 2021
- Source ID
- N000142112430
Entities
People
- David Voelz
Organizations
- New Mexico State University
- Office of Naval Research
- United States Navy