Pulsar Candidates Classification with Deep Convolutional Neural Networks. (arXiv:1909.05301v1 [astro-ph.IM])
<a href="http://arxiv.org/find/astro-ph/1/au:+Wang_Y/0/1/0/all/0/1">Yuanchao Wang</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Li_M/0/1/0/all/0/1">Mingtao Li</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Pan_Z/0/1/0/all/0/1">Zhichen Pan</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Zheng_J/0/1/0/all/0/1">Jianhua Zheng</a>
As performance of dedicated facilities continually improved, massive pulsar
candidates are being received, which makes selecting valuable pulsar signals
from candidates challenging. In this paper, we designed a deep convolutional
neural network (CNN) with 11 layers for classifying pulsar candidates. Compared
to artificial designed features, CNN chose sub-integrations plot and sub-bands
plot in each candidate as inputs without carrying biases. To address the
imbalanced problem, data augmentation method based on synthetic minority
samples is proposed according to characteristics of pulsars. The maximum pulses
of pulsar candidates were first translated to the same position, then new
samples were generated by adding up multiple subplots of pulsars. The data
augmentation method is simple and effective for obtaining varied and
representative samples which keep pulsar characteristics. In the experiments on
HTRU 1 dataset, it shows that this model can achieve recall as 0.962 while
precision as 0.963.
As performance of dedicated facilities continually improved, massive pulsar
candidates are being received, which makes selecting valuable pulsar signals
from candidates challenging. In this paper, we designed a deep convolutional
neural network (CNN) with 11 layers for classifying pulsar candidates. Compared
to artificial designed features, CNN chose sub-integrations plot and sub-bands
plot in each candidate as inputs without carrying biases. To address the
imbalanced problem, data augmentation method based on synthetic minority
samples is proposed according to characteristics of pulsars. The maximum pulses
of pulsar candidates were first translated to the same position, then new
samples were generated by adding up multiple subplots of pulsars. The data
augmentation method is simple and effective for obtaining varied and
representative samples which keep pulsar characteristics. In the experiments on
HTRU 1 dataset, it shows that this model can achieve recall as 0.962 while
precision as 0.963.
http://arxiv.org/icons/sfx.gif
