Deblending and classifying astronomical sources with Mask R-CNN deep learning

Abstract

We apply a new deep learning technique to detect, classify, and deblend sources in multi-band astronomical images. We train and evaluate the performance of an artificial neural network built on the Mask R-CNN image processing framework, a general code for efficient object detection, classification, and instance segmentation. After evaluating the performance of our network against simulated ground truth images for star and galaxy classes, we find a precision of 92% at 80% recall for stars and a precision of 98% at 80% recall for galaxies in a typical field with ∼ 30 galaxies/arcmin 2 . We investigate the deblending capability of our code, and find that clean deblends are handled robustly during object masking, even for significantly blended sources. This technique, or extensions using similar network architectures, may be applied to current and future deep imaging surveys such as LSST and WFIRST. Our code, Astro R-CNN, is publicly available at https://github.com/burke86/astro_rcnn.

Supernovae Detection with Fully Convolutional One-Stage Framework

“…If the candidate is located off-center in the image, the performance of the model will decrease. Burke et al [ 8 ] applied object detection on the astronomical source classification. Their object detection framework is based on the Mask-RCNN [ 21 ] and treat the classification task as a star/galaxy detection task and achieved performance with 92% precision and 80% recall on stars and 98% precision and 80% recall on galaxies.…”

“…Because most of the works are finished by neural networks, the detection speed and accuracy have been significantly improved. The Mask-RCNN [ 21 ], the segmentation version of the Faster-RCNN, is also used in Burke’s work [ 8 ] and has great performance.…”

“…Most of the data used by the existing works, as mentioned in Section 2 , are simulated data or partially generated by simulation. In [ 7 , 8 , 12 , 16 , 19 ], the positive candidates for training are simulated. In [ 13 , 14 ], the authors simulated observation fluctuations such as flux and SNR variation.…”

“…Muthukrishna et al [ 7 ] introduced the Real-time Automated Photometric Identification (RAPID) architecture and implemented fine-grade supernovae classification. Burke et al [ 8 ] applied object detection to the astronomical source classification.…”

Detection and Classification of Astronomical Targets with Deep Neural Networks in Wide-field Small Aperture Telescopes

“…Duev et al (2019b) proposes to use the machine learning algorithm to classify star-galaxy and separate Real/Bogus transients. Burke et al (2019) uses semantic segmentation model named Mask R-CNN ) for real-time astronomical targets detection and classification. These methods mentioned above are mainly used for general purpose sky survey telescopes and they require large amount of astronomical images which are labeled by human experts as training data.…”