Image-Based Multi-Survey Classification of Light Curves with a Pre-Trained Vision Transformer

2025-07-15

AI Summary

Multi-survey light-curve classification faces significant challenges due to strong data heterogeneity and pronounced survey-specific characteristics. Method: We propose a joint modeling paradigm based on image-based representation and multi-source fusion: ZTF and ATLAS light curves are uniformly encoded as time-series images and fed into a pre-trained Swin Transformer V2; a dedicated multi-survey architecture explicitly models survey-specific biases and cross-survey interactions. Contribution/Results: This work pioneers the adaptation of vision Transformers to multi-survey time-domain astronomical classification, introducing an end-to-end framework for cross-survey feature alignment and collaborative learning. Experiments demonstrate state-of-the-art performance on joint multi-survey classification—substantially outperforming single-survey baselines and conventional sequence models—validating strong generalizability and scalability. The approach establishes a novel paradigm for intelligent, large-scale time-domain astronomical classification.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

We explore the use of Swin Transformer V2, a pre-trained vision Transformer, for photometric classification in a multi-survey setting by leveraging light curves from the Zwicky Transient Facility (ZTF) and the Asteroid Terrestrial-impact Last Alert System (ATLAS). We evaluate different strategies for integrating data from these surveys and find that a multi-survey architecture which processes them jointly achieves the best performance. These results highlight the importance of modeling survey-specific characteristics and cross-survey interactions, and provide guidance for building scalable classifiers for future time-domain astronomy.

Problem

• Classify light curves using pre-trained vision Transformer
• Integrate multi-survey data for improved classification performance
• Model survey-specific traits for scalable astronomy classifiers

Innovation

• Uses Swin Transformer V2 for classification
• Integrates multi-survey data jointly
• Models survey-specific and cross-survey interactions