A novel method based on adaptive autoencoder and improved long short-term memory and gated recurrent unit for nuclear radiation measurement and monitoring

Abstract

One way to implement the defense against nuclear threat is based on the measurement and detection of radiation. To cope with the problems of low precision and slow warning speed in nuclear radiation monitoring and warning, a novel method based on adaptive autoencoder and improved Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) is proposed. Considering the large amount of redundant information in the nuclear radiation energy spectrum, an adaptive autoencoder network is designed to reduce the dimension of data. In order to overcome the problems of slow training speed of LSTM/GRU and insufficient utilization of internal features of the network, the polynomial combinations are increased, which will help to improve the accuracy on the task. The experiments conducted on data measured in real scenario show that the proposed method has high accuracy and reliability for nuclear radiation monitoring and early warning in the real world event.