Over the last 20 years, the cost of uncooled microbolometer-based imaging systems has drastically decreased while performance has increased. In the simplest terms, the figure of merit for these types of thermal detectors is given in terms of the τ-NETD product, the combination of the thermal time constant and the noise equivalent temperature difference. Considering these factors, optimal system design parameters are investigated to maximize visual information content. This dissertation focuses on improving scene information in the longwave infrared (LWIR) spectrum that has had its validity and quality degraded by noise, blur, and reflected radiance. Taken together, noise and blur degrade image quality, directly affecting system performance for object detectors trained with deep learning. Representing noise with NETD and blur in terms of equivalent angular resolution, this research provides a systematic method for relating design parameters to specific machine vision tasks that are difficult to define in a traditional imaging sense. This method provides for a system design approach based on information requirements rather than improvements to machine vision algorithms. As a machine vision function, automated target recognition (ATR) has improved with new technologies and the wide proliferation of infrared staring focal planes. Infrared search and track (IRST), which is target detection and localization at long ranges of unresolved targets, can be performed by both photon counting and microbolometer systems. The transition from broadband system design to one that involves spectral characterizations of components provides a better understanding of the performance and capabilities of new technologies. Unlike reflective bands such as visible and shortwave infrared (SWIR), reflected radiance reduces contrast in the LWIR, resulting in lost information. This research considers the sky path radiance contribution to the radiant exitance of a scene that reduces contrast, and consequently, information. Results show that reduced contrast can be overcome by utilizing multiband spectral imaging systems to remove the reflected component, thus increasing the scene information available. In addition, better scene consistency can be achieved between day and night when reflected radiance is removed. The multiband LWIR system designs presented take advantage of the low τ-NETD of modern microbolometers and demonstrate feasibility in future multiband applications.
If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu
Doctor of Philosophy (Ph.D.)
College of Optics and Photonics
Optics and Photonics
Optics and Photonics
Length of Campus-only Access
Doctoral Dissertation (Open Access)
Grimming, Robert, "Uncooled Microbolometer Imaging Systems for Machine Vision" (2022). Electronic Theses and Dissertations, 2020-. 1379.