Supernova Detection Based on Multi-scale Fusion Faster RCNN

Supernovae are of great significance in the study of life evolution and expansion history of the universe. In this paper, multi-scale fusion Faster RCNN model is used to realize automatic detection of supernovae. Firstly, the dataset was synthesized and rotated to enhance supernova features. Secondly, based on the Faster RCNN architecture, the network combined with Resnet50 and PAFPN is used to extract image features, so as to obtain multi-scale feature maps and improve the ability of the network to extract small target features. Finally, OHEM network is introduced into the network to balance the difference between positive and negative sample in number. The prepared dataset was used for the training of the multi-scale fusion Faster RCNN model, and the MACRO-F1 of the test set was up to 96.6%. The method proposed in this paper can realize the detection of supernovae from astronomical imagery with relatively high accuracy.