Pulsar Candidate Sifting Using Multi-input Convolution Neural Networks

Pulsar candidate sifting is an essential process for discovering new pulsars. It aims to search for the most promising pulsar candidates from an all-sky survey, such as the High Time Resolution Universe (HTRU), Green Bank Northern Celestial Cap (GBNCC), Five-hundred-meter Aperture Spherical Radio Telescope, etc. Recently, machine learning (ML) has become a hot topic in investigations of pulsar candidate sifting. However, one typical challenge in ML for pulsar candidate sifting comes from the learning difficulty arising from the high class imbalance between the observed numbers of pulsars and non-pulsars. Therefore, this work proposes a novel framework for candidate sifting, named Multi-input Convolutional Neural Networks (MICNN). MICNN is an architecture of deep learning with four diagnostic plots of a pulsar candidate as its inputs. To train our MICNN on a highly class-imbalanced data set, a novel image augmentation technique is proposed, as well as a three-stage training strategy. Experiments on observations from HTRU and GBNCC show the effectiveness and robustness of these proposed techniques. In the experiments on HTRU, our MICNN model achieves a recall rate of 0.962 and a precision rate of 0.967 even in a highly class-imbalanced test data set.