Astronet-Triage-v2: Eclipsing Signal Classification from TESS Mission

Approximately 25,000 astronomical signals detected via the Box Least Squares method in the first three years of the TESS mission. The dataset was created by Evan Tey at MIT for training a CNN to classify eclipsing signals, such as exoplanetary transits or eclipsing binary stars, from other signal types. Each signal received one of five labels from at least three individuals, with most conflicts resolved into a consensus label.

Use Cases

• Training a convolutional neural network (CNN) for astronomical signal classification based on labeled light curve data.
• Classifying eclipsing binary star signals from other periodic variations in stellar brightness.
• Identifying potential exoplanetary transit candidates from threshold-crossing events (TCEs).
• Benchmarking new machine learning models against the Astronet-Triage-v2 architecture.
• Studying the characteristics of signals detected by the Box Least Squares method in TESS data.

Strengths

• Contains approximately 25,000 labeled signals, providing a substantial sample for model training.
• Each signal was labeled by at least three individuals, and a consensus label was established for most conflicts, suggesting a process for label reliability.
• Signals are derived from the first three years of the TESS mission, a major contemporary space telescope.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Massachusetts Institute of Technology (MIT), Evan Tey

Collection Method

Signals detected via Box Least Squares (Kovacs 2002) from TESS mission light curve data.

Time Range

First 3 years of the TESS mission