nomaly Detection in Gamma-ray Radiation Spectra using Artificial Neural Network and Ant Colony Optimization

etection of threatening radioactive sources is a crucial task for homeland security. One of the major challenges in radiation detection is to distinguish between the background radiation and the anomalous source radiation from the measured radiation spectrum at low source to background ratio. In this paper, an anomaly detection technique is proposed to detect anomalous Gamma-ray radiation spectra using machine learning. The method is based on dividing the radiation spectrum into two sub-spectra, where the background part of the second sub-spectrum is predicted from the first sub-spectrum through a neural network model. Hence, an anomaly spectrum is detected according to the difference between the predicted background radiation data and the measured values of the second sub-spectrum. The ant colony optimization is utilized to select, from the radiation spectrum, the values assigned to the first and second sub-spectra, where optimum prediction accuracy can be provided. To present the effectiveness of the proposed work, a performance comparison is conducted with both benchmark and a recent neural network-based method. The performance is evaluated using real data that represent both background and radioactive source radiation, which are measured through a network of detectors. Experimental results show that the proposed method outperforms the other methods in terms of the detection capability even at low source to background radiation ratios.