Nuclide Identification Model Training for Mixed Spectrum Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods require a large amount of radiation measurement data and prolonged time to accurately identify nuclides, especially when multiple nuclides are mixed, leading to low accuracy.
Innovation Solution
A method and apparatus for training a nuclide identification model that preprocesses nuclide data by classifying it into energy spectrums and generating training data based on the number of data in each classified characteristic, using an artificial neural network model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional radiation measurement methods are used to collect sufficient data for accurate nuclide identification, then identification accuracy improves, but measurement time increases significantly to several hours or more
Solution Approach 1:
The patent applies preliminary action by pre-processing radiation measurement data to extract and store characteristic features of nuclides before actual identification occurs. The system pre-calculates and stores energy spectrum characteristics, decay patterns, and other identifying features in a database, so that during actual measurement, the system only needs to compare new data against these pre-prepared templates rather than analyzing raw data from scratch, dramatically reducing identification time while maintaining accuracy
Solution Approach 2:
The patent uses copying by creating simplified representations or models of nuclide characteristics from extensive measurement data. Instead of storing and processing all raw measurement data, the system creates compressed feature vectors, characteristic curves, or template patterns that capture the essential identifying features of each nuclide. These copies enable rapid comparison and identification without requiring the full original datasets during measurement
2Loss of time
If the amount of radiation measurement data is reduced to shorten measurement time, then measurement time decreases, but nuclide identification accuracy becomes significantly low
Solution Approach 1:
The patent applies extraction by isolating and extracting the most discriminative features from radiation measurement data. The system identifies and extracts key characteristics such as peak energy positions, full width at half maximum (FWHM), area under curves, and specific spectral features that are most important for nuclide identification. By extracting only these critical features rather than using all measurement data, the system achieves accurate identification with minimal data
Solution Approach 2:
The patent applies local quality by focusing analysis on specific critical regions or features of the energy spectrum rather than treating all data equally. The system identifies particular energy ranges, spectral peaks, or characteristic patterns that are most informative for distinguishing between nuclides and concentrates processing resources on these locally important features, achieving high accuracy with reduced overall data requirements
3Measurement precision
If traditional methods are used for nuclide identification with limited data, then device complexity remains simple, but identification accuracy is insufficient when multiple nuclides are mixed
Solution Approach 1:
The patent applies segmentation by dividing the complex task of nuclide identification into separate processing stages and feature analysis components. The system segments the identification process into: (1) data pre-processing and feature extraction, (2) characteristic pattern matching, (3) comparison against reference templates, and (4) identification decision-making. This segmentation allows each component to be optimized independently and simplifies the overall system architecture while handling complex mixed nuclide scenarios
Data Source
AI summary
A method for training an apparatus for training a nuclide identification model is provided. In the method, nuclide data is classified into characteristics of energy spectrums for nuclides, training data is generated based on a number of data in each of the classified characteristics, and the nuclide identification model is trained by using the training data.