Temporal Stamp Classifier: Classifying Short Sequences of Astronomical Alerts

TL;DR

The paper addresses the need for fast, accurate classification of astronomical alerts from the Zwicky Transient Facility by developing a Temporal Stamp Classifier that processes short sequences (2–5 detections) of stamp images and metadata, augmented with AllWISE crossmatch features. The approach combines CNN-based per-detection feature extraction with recurrent networks to capture temporal information, and systematically evaluates multiple recurrent architectures, with and without AllWISE data. Key findings include near-98% test accuracy for 2–4 detections and substantial gains from AllWISE crossmatches, as well as an enhanced single-detection stamp classifier using random rotations. The work bridges the gap between first-detection stamp classification and longer time-series approaches, offering a practical, efficient solution for real-time alert brokers and follow-up prioritization, and outlines directions toward further improvements such as subtyping and transformer-based models.

Abstract

In this work, we propose a deep learning-based classification model of astronomical objects using alerts reported by the Zwicky Transient Facility (ZTF) survey. The model takes as inputs sequences of stamp images and metadata contained in each alert, as well as features from the All-WISE catalog. The proposed model, called temporal stamp classifier, is able to discriminate between three classes of astronomical objects: Active Galactic Nuclei (AGN), Super-Novae (SNe) and Variable Stars (VS), with an accuracy of approximately 98% in the test set, when using 2 to 5 detections. The results show that the model performance improves with the addition of more detections. Simple recurrence models obtain competitive results with those of more complex models such as LSTM.We also propose changes to the original stamp classifier model, which only uses the first detection. The performance of the latter model improves with changes in the architecture and the addition of random rotations, achieving a 1.46% increase in test accuracy.

Figures (9)

• Figure 1: Examples of ZTF alerts of five classes of astronomical objects: AGN, SN, VS, Asteroid and Bogus.
• Figure 2: Confusion matrix of the original Stamp Classifier stamp using our own implementation.
• Figure 3: Confusion matrix of the enhanced Stamp Classifier that includes an image size of $33 \times 33$ and 6 rotations with $0.08[rad]$ of rotation range.
• Figure 4: Test accuracy v/s inference time for 4 alerts. Diamonds represent the best configuration for a given model in terms of test accuracy.
• Figure 5: Average confusion matrix for the test set using 5 different realizations of the temporal stamp classifier for each number of alerts using the model configurations of Table \ref{['table:recurrent']}.