Radioactive nuclide identification model construction method and device

• Main Authors: LIAO PENG, NIU DEQING, YUAN MINJUAN, LIANG JIE, HOU XIN, MU XIANGFAN, YAO FEI, HAN QIANG
• Format: Patent
• Language: Chinese; English
• Published: 13.01.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTEDFOR SPECIFIC APPLICATION FIELDS; MEASUREMENT OF NUCLEAR OR X-RADIATION; MEASURING; PHYSICS; TESTING

The invention discloses a radionuclide identification model construction method and device, which apply singular spectrum analysis to energy spectrum noise reduction processing, effectively extract characteristic components in energy spectrum data and remove useless noise signals, and provide a high-precision data source for deep learning network modeling. A large sample database required by deep learning network training is established by adopting the simulated energy spectrum, the problem that the types of deep learning network training samples and laboratory radioactive sources are insufficient is solved, and the method can be used for stable identification of various nuclides (> = 7) in a medium. The deep learning nuclide identification model obtained through modeling has the advantages of being high in identification rate (larger than or equal to 99%) and high in response speed (larger than or equal to 1 s), and can be used for screening radionuclides in different media such as soil, water and coal mines