Gas Detection using Hyperthermal Images
Abstract
The research has been divided into three parts. We started with gases being emitted from a smokestack. We developed a way to correlate known laboratory spectra with the unknown emitted spectra. We found the two main components: carbon dioxide and sulfur dioxide. Our method involved the careful selection of critical wavelengths for the analysis. While this enabled us to find the main concentrations of gas, we wanted to extend the physical extent of the cloud. We did this by shifting from the laboratory signatures to the actual averaged real data signatures of the gas cloud and then performing a matched filter. The image was then thresholded on the basis of the background statistics. This procedure was performed iteratively until a stable division of background/gas was achieved. The resulting signatures can be separated by energy into that which is due to the gases and that which due to the background; the percentages of each gas can be calculated. Finally, we successfully separated the temperature from the gas by recognizing the different temperature dependences of the gas components. In this report, we develop an algorithm for the detection, identification and relative quantification of effluent gases emitted from an industrial plume stacks using an LWIR hyperspectral remote sensing system. The algorithm is based on applying a stepwise regression process using known library gas signatures. The system first fits the spectra to a combination of all the possible gases; it then systematically reduces the number of gases while maintaining a good fit to the observed spectra. The system was tested on data cubes synthesized in the Digital Imaging and Remote Sensing (DIRS) Laboratory at the Rochester Institute of Technology and on actual gas cubes from the stack of a power plant; good results were obtained. In our new algorithm, a prior analysis shows which gases can be confused with each other; thresholds are set to avoid multiple gas identifications of a single gas.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 28, 2012
- Accession Number
- ADA558488
Entities
People
- Dan G. Blumberg
- Stanley Rotman