Fast nuclide recognition method based on feature transformation and neural network

Abstract

The invention discloses a fast nuclide recognition method based on feature transformation and a neural network. The method comprises the steps that a gamma detector is used to measure a natural background spectrum and a radionuclide spectrum; a filtering algorithm is used to smooth the natural background spectrum to acquire a standard background spectrum; the filtering algorithm is used to smooth the radionuclide spectrum to acquire a standard radionuclide spectrum; the standard background spectrum is subtracted from the standard radionuclide spectrum to acquire a net quantization spectrum; eigentransformation is carried out on the net quantization spectrum; a certain number of transform coefficients are sequentially extracted as energy spectrum feature vectors and are normalized; and the normalized feature vectors are input into the neural network for nuclide recognition. The nuclide recognition method provided by the invention is not affected by the measurement time, the detection distance and the nuclide activity, has the advantages of fast response speed, natural background radiation, small noise interference and the like, and can be used for rapid nuclide recognition of a portable radiation detector.

Description

technical field [0001] The invention relates to a nuclide identification method, in particular to a fast nuclide identification method based on feature transformation and neural network. Background technique [0002] In a short period of time, no matter whether it is a crystal with a large size or a high detection efficiency, no clearly distinguishable nuclide characteristic peaks can be formed, so the nuclide identification algorithm based on the peak-seeking theory is not suitable for the rapid nuclide identification task. With the rapid increase of computer computing speed and the development of neural network, people consider using the full spectrum as a feature vector and inputting it into the neural network for nuclide identification. However, full-spectrum input has disadvantages such as large data redundancy, large number of input layers, long training time, and high requirements for computer performance. Contents of the invention [0003] The technical problem to...

AI Technical Summary

Problems solved by technology

However, full-spectrum input has disadvantages such as large data redundancy, large number of input layers, long training time, and high requirements for computer performance.

Images