Impact of Electro-Optical Noises and Quality Hindrances in Satellite Imagery

The focus of this presentation is on various Electro-Optical and Infrared (EO/IR) noises and Quality Hindrances (QHs) that impact the detection of a slow-moving spatially-unresolved object captured by EO/IR sensor system. We have reserved the term, noises to mean degradation factors due to EO/IR system and QHs to mean degradations due to environmental factors outside EO/IR such as clouds, fixed pattern noise (FPN), etc. The goal is to apply various noises/QHs to a slow-moving spatially-unresolved object and measure/report the degree to which detection of the object is hindered. The measuring is accomplished through statistical methods. The data representative of images containing the object and various noises/QHs are generated with a simulated software that emulates Landsat8 EO/IR sensor system with the atmospheric conditions where a dim (low SNR) and spatially unresolved object moves in wide field of view (WFOV). The object detection and the accuracy prediction model is accomplished using a U-Net neural networks. The current work is the continuation of a previously published work that has used the same neural networks architecture with a relatively high accuracy albeit without considering noise/QHs. The intent of the spiral-up compendium works is to outline a sensitivity (robustness) analysis and their limits thereof.