Research map / Time series / Foundation models for time series

Kepler Light Curve Classification Using Deep Learning and Markov Transition Field (Student Abstract)

AAAI 2024 · Time series › Foundation models for time series · approach: Satellite image time series · 2 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Converts Kepler light curves into Markov Transition Field images and classifies them as exoplanet vs false positive with a custom CNN, reporting 99.39% accuracy versus standard-plot and GADF baselines.

Problem

Manually classifying Kepler light curves as exoplanet or false positive is infeasible at large data volumes, so an automated classifier is needed.

Approach

Encode each light curve time series as a Markov Transition Field image so periodic dimming becomes visible as structure, then classify the image with a small custom CNN instead of feeding raw 1D light curves to models such as AstroNet.

Key results

• MTF input with the proposed CNN achieved 99.39% accuracy, 99.68% precision, 99.46% recall and 99.57% F-score.
• Standard light curve plot input reached 93.35% accuracy and GADF reached 94.06%, so MTF outperforms them by up to 6.04 points.
• The authors compare against reported AstroNet (95.8%) and modified AstroNet (98%) numbers from prior work, not on the same data.

Measured effects reported in the paper

Surprising findings

• The standard light curve plot gave almost 6% lower accuracy than the MTF input, which the authors describe as surprising.

Limitations

• Only long-exposure data was used; fast and short exposures were not studied.
• Baseline AstroNet results were not reproduced on the same dataset, so the comparison to prior work is indirect.
• Small single-split evaluation with no reported variance or cross-validation.

Experimental setup

Datasets: NASA Exoplanet Archive Kepler light curves (6552 KOIs)

Benchmarks: Kepler KOI light curve classification (held-out 10% split)

Metrics: accuracy precision recall F-score

Baselines: Standard light curve plot CNN GADF CNN AstroNet (reported 95.8%) Dattilo et al. 2019 modified AstroNet (reported 98%)

Builds on: AstroNet Markov Transition Field Gramian Angular Field

Closest papers by meaning

• A New Statistical Model of Star Speckles for Learning to Detect and Characterize Exoplanets in Direct Imaging Observations (CVPR 2025 · similarity 0.82)
• Probabilistic Data-Driven Modelling of Astrophysical Transients: The Neural Process Family for Ultrafast and Class-Agnostic Light Curve Reconstruction (NeurIPS 2026 · similarity 0.81)
• Kepler codebook (ICML 2024 · similarity 0.79)
• Unified Framework for Diffusion Generative Models in SO(3): Applications in Computer Vision and Astrophysics (AAAI 2024 · similarity 0.78)
• The Multimodal Universe: Enabling Large-Scale Machine Learning with 100 TB of Astronomical Scientific Data (NeurIPS 2024 · similarity 0.77)
• Improved Learning via k-DTW: A Novel Dissimilarity Measure for Curves (ICML 2025 · similarity 0.77)