Anomaly detection in gamma-ray radiation spectra using artificial neural network and ant colony optimization.

Accurate detection of anomalous radioactive sources in environmental monitoring systems is critical for both radiological protection and nuclear security. This study addresses the fundamental challenge of discriminating anomalous radiation signals from natural background fluctuations, particularly at low source to background ratios. We present a novel machine learning approach for anomaly detection in gamma-ray spectra that combines neural network modeling with bio-inspired optimization. The method innovatively partitions radiation spectra into two complementary sub-spectra, using a trained neural network to establish their background correlation. Anomalies are identified through significant deviations between measured values and neural network predictions. A key innovation is the integration of ant colony optimization to select spectral partitions that provide maximum accuracy. The system was rigorously evaluated using empirical data from distributed radiation detectors, incorporating both background measurements and spectra from common radioactive sources (Cs and Co). Comparative experiments demonstrate superior performance over existing benchmark methods, with particular advantage in low source to background ratios. The proposed technique advances radiation monitoring capabilities by providing enhanced sensitivity to weak anomalous signals and practical deployment potential using standard detector networks. These improvements are particularly relevant for environmental monitoring and security applications where early detection of radiation anomalies is critical.