# Deep Learning for Image Sequence Classification of Astronomical Events

**Authors:** Rodrigo Carrasco-Davis, Guillermo Cabrera-Vives, Francisco Förster, Pablo A. Estévez, Pablo Huijse, Pavlos Protopapas, Ignacio Reyes, Jorge Martínez, Cristóbal Donoso  
**arXiv:** [1807.03869](https://arxiv.org/abs/1807.03869)  
**Date:** 2018-07-10  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1807.03869v1/

## Key points

- Rodrigo Carrasco-Davis and colleagues propose a new sequential classification model for astronomical objects based on a recurrent convolutional neural network (RCNN) that uses sequences of images as inputs.
- This approach avoids the computation of light curves or difference images and is the first time that sequences of images are used directly for the classification of variable objects in astronomy.
- The authors transform synthetic light-curves to images in a realistic way, taking into account observational conditions and instrumental parameters to solve partially the problem of transfer learning from synthetic to real-world images.
- Using a simulated dataset is faster and more adaptable to different surveys and classification tasks compared to collecting real labeled image sequences.
- To test the RCNN classifier trained with a synthetic dataset, they used real-world data from the High cadence Transient Survey (HiTS) obtaining an average recall of $87%$ on four classes: supernovae, RR Lyrae, non-variables, and asteroids.
- Their RCNN model outperforms the light curve classifier due to the extra information contained on the images.
- The results obtained encourage them to use and continue developing this proposed method for astronomical alert brokers systems that will process alert streams generated by new telescopes such as the Large Synoptic Survey Telescope and the Zwicky Transient Facility.

## Summary

In their paper "Deep Learning for Image Sequence Classification of Astronomical Events," Rodrigo Carrasco-Davis and colleagues propose a new sequential classification model for astronomical objects based on a recurrent convolutional neural network (RCNN) that uses sequences of images as inputs. This approach avoids the computation of light curves or difference images and is the first time that sequences of images are used directly for the classification of variable objects in astronomy. The authors solve partially the problem of transfer learning from synthetic to real-world images by transforming synthetic light-curves to images in a realistic way, taking into account observational conditions and instrumental parameters. This approach allows them to generate datasets to train and test their RCNN model for different astronomical surveys and telescopes. Moreover, using a simulated dataset is faster and more adaptable to different surveys and classification tasks compared to collecting real labeled image sequences. To test the RCNN classifier trained with a synthetic dataset, they used real-world data from the High cadence Transient Survey (HiTS) obtaining an average recall of $87%$ on four classes: supernovae, RR Lyrae, non-variables, and asteroids. They compare the results of their model with those of a light curve classifier in both simulated and real data. Their RCNN model outperforms the light curve classifier due to the extra information contained on the images. The results obtained encourage them to use and continue developing this proposed method for astronomical alert brokers systems that will process alert streams generated by new telescopes such as the Large Synoptic Survey Telescope and the Zwicky Transient Facility. The authors' work represents an important step forward in automating astronomical object classification using deep learning techniques applied directly on image sequences without requiring intermediate processing steps such as light curves or difference imaging.