Performance of a Fisher Linear Discriminant Analysis Gamma-Ray Identification Algorithm

The development of a gamma-ray identification algorithm based on a Fisher Linear Discriminant Analysis (FLDA) approach has previously shown high potential in count starved situations. This work uses 48,000 semi-empirical synthetic spectra to evaluate the algorithm performance over a broad range of dose rates and acquisition times. True positive identifications of 100% were typically seen for dose rates ≥ 0.05 μSv/h and acquisition times ≥10 s. For in air and shielded radionuclide spectra, no false alarms were observed for dose rates ≥ 0.01 μSv/h and acquisition times ≥10 s. The single radionuclide identification performance typically exceeded the ANSI 42.34 standard by a factor of 10 for dose rate and a factor of 10 for acquisition time. The evaluated shielded signatures had no detrimental effects to the identification performance. The correct identification of HEU masked by a 0.5 μSv/h ⁴⁰K or ⁶⁰Co source could typically be made for a masking ratio of 10:1 for acquisition times ≥ 1 s. For masking with a 0.5 μSv/h ¹³⁷Cs source, HEU could be identified at 7:1 ratios for times ≥10 s. The excellent results were obtained for a non-optimal identification threshold. Optimization of the threshold would lead to further performance improvements.