Application of support vector machine to rapid classification of uranium waste drums using low-resolution γ-ray spectra

Haruhi Hata; Kaoru Yokoyama; Yuu Ishimori; Yoshiyuki Ohara; Yoshio Tanaka; Noritake Sugitsue

doi:10.1016/j.apradiso.2015.06.030

Application of support vector machine to rapid classification of uranium waste drums using low-resolution γ-ray spectra

Appl Radiat Isot. 2015 Oct:104:143-6. doi: 10.1016/j.apradiso.2015.06.030. Epub 2015 Jun 24.

Authors

Haruhi Hata¹, Kaoru Yokoyama², Yuu Ishimori², Yoshiyuki Ohara², Yoshio Tanaka², Noritake Sugitsue²

Affiliations

¹ Ningyo-toge Environmental Engineering Center, Japan Atomic Energy Agency, 1550 Kamisaibara, Kagamino-cho, Tomata-gun, Okayama 708-0698, Japan; Graduate School of Natural Science and Technology, Okayama University, 3-1-1, Tsushima-naka kita-ku, Okayama 700-8530, Japan. Electronic address: hata.haruhi@s.okayama-u.ac.jp.
² Ningyo-toge Environmental Engineering Center, Japan Atomic Energy Agency, 1550 Kamisaibara, Kagamino-cho, Tomata-gun, Okayama 708-0698, Japan.

PMID: 26159663
DOI: 10.1016/j.apradiso.2015.06.030

Abstract

We investigated the feasibility of using support vector machine (SVM), a computer learning method, to classify uranium waste drums as natural uranium or reprocessed uranium based on their origins. The method was trained using 12 training datasets were used and tested on 955 datasets of γ-ray spectra obtained with NaI(Tl) scintillation detectors. The results showed that only 4 out of 955 test datasets were different from the original labels-one of them was mislabeled and the other three were misclassified by SVM. These findings suggest that SVM is an effective method to classify a large quantity of data within a short period of time. Consequently, SVM is a feasible method for supporting the scaling factor method and as a supplemental tool to check original labels.

Keywords: Support vector machine; Uranium waste; Uranium waste management; γ-Ray spectra.