The U.S Environmental Protection Agency (USEPA) Airborne Spectral Photometric Environmental Collection Technology (ASPECT) program is designed in part to provide first responders with radiological mapping of potentially hazardous locations. This program utilizes an aircraft fitted with a gamma-ray spectrometer capable of remote detection of radioisotopes. The challenges present in detecting a radioisotope signal remotely are strongly tied to the signal-to-noise ratio of the collected gamma-ray spectra and the specific signal processing and pattern recognition methods used in the data analysis. Depending on the distance from the detector to the radioisotope source, Compton scattering can significantly reduce the analyte signal, and weakened signals pose a significant challenge when attempting to design an effective classifier for detecting radioisotopes of interest.
In this research, a basic methodology has been developed for the detection of cesium-137 (¹³⁷Cs) and cobalt-60(⁶⁰Co) utilizing only laboratory collected spectra and backgrounds from the field collected only once. The presented classifier methodology has been proven to provide a fundamental structure for which more advanced algorithms can be developed. Furthermore, this methodology has demonstrated the ability to strongly associate a level of confidence in a detection which allows for intelligent decision making. From this basic methodology, more sensitive and selective algorithms can be designed.
The Compton effect has previously been problematic in the development of gamma-ray pattern recognition systems. In this research, a background suppression methodology utilizing linear regression has been implemented to enhance the basic pattern recognition methodology. This background correction strategy has led to significant improvements in the remote detection of radioisotopes and enables the classification of more complex radioisotopes such as europium-152 (¹⁵²Eu). This research demonstrates not only the capabilities of the pattern recognition methodology, but also the flexibility of the procedure.
In the remote detection of radioisotopes, false detections and low sensitivity are the key challenges when developing a classifier. While the background corrected methodology was shown to greatly enhance the classifier performance, further advances can be made into the methodology through the use of committee classifiers. Since every classifier utilizes a different dataspace, a standardization procedure has been developed from which the classifier result can be averaged and generate a committee classifier result. This classifier methodology has been demonstrated to further improve the radioisotope classifier performance without sacrificing either sensitivity or selectivity.
While the development of targeted radioisotope classifiers is invaluable to the first responder, developing a general gamma-ray anomaly classifier can handle those radioisotopes that have no dedicated classifier. From this objective, a unique anomaly classifier based on the Compton region of the gamma-ray spectrum has been developed and demonstrated to operate in the field. Utilizing all of the strategies and techniques developed for single radioisotope classifiers, the anomaly classifier has been proven to detect background and natural radioactive sources as well as controlled and man-made radioisotope targets.
From the developed remote detection classifiers via pattern recognition techniques, robust radiation detection classifiers have been developed. This pattern recognition methodology eschews the need for extensive field data collection for training the algorithms, while also removing the need for on-site calibrations. When used in conjunction with one another, the dedicated radioisotope and anomaly classifiers provide a thorough and rugged remote detection capability to the first responder. The presented methodology also demonstrates that any radioisotope classifier can be generated, implying that this method can be used for the detection of any radioisotope in the field.
Identifer | oai:union.ndltd.org:uiowa.edu/oai:ir.uiowa.edu:etd-6713 |
Date | 01 August 2016 |
Creators | Dess, Brian William |
Contributors | Small, Gary W. (Gary Wray), 1957- |
Publisher | University of Iowa |
Source Sets | University of Iowa |
Language | English |
Detected Language | English |
Type | dissertation |
Format | application/pdf |
Source | Theses and Dissertations |
Rights | Copyright 2016 Brian William Dess |
Page generated in 0.002 seconds