Longwave Infrared Hyperspectral Subpixel Target Detection with Machine Learning (Preprint)
Air Force Research Laboratory AFRL/RYMT Wright-Patterson AFB United States
Pagination or Media Count:
Hyperspectral imaging has been used to perform automated material detection and identification. However, traditional detection methods based on statistical data processing produce a higher than desired false alarm rate for subpixel targets due to violated assumptions. This paper compares performance of machine learning methods using neural networks in detecting subpixel targets with traditional statistical methods. The assessment will utilize airborne data collected by the SEBASS sensor under a variety of atmospheric conditions from a number of different altitudes. Various methods for atmospheric compensation and temperature-emissivity separation will be used as well to assess robustness of the detection approaches.