Efficient labeling of solar flux evolution videos by a deep learning model

Machine learning is becoming a critical tool for the interrogation of large, complex data. Labelling, defined as the process of adding meaningful annotations, is a crucial step of supervised machine learning. However, labelling datasets is time consuming. Here we show that convolutional neural networks (CNNs) trained on crudely labelled astronomical videos can be leveraged to improve the quality of data labelling and reduce the need for human intervention. We use videos of the solar magnetic field that are divided into two classes—emergence or non-emergence of bipolar magnetic regions (BMRs)—on the basis of their first detection on the solar disk. We train CNNs using crude labels, manually verify, correct disagreements between the labelling and CNN, and repeat this process until convergence is reached. Traditionally, flux emergence labelling is done manually. We find that a high-quality labelled dataset derived through this iterative process reduces the necessary manual verification by 50%. Furthermore, by gradually masking the videos and looking for maximum changes in CNN inference, we locate BMR emergence time without retraining the CNN. This demonstrates the versatility of CNNs for simplifying the challenging task of labelling complex dynamic events. Big-data labelling is critical to harness the power of supervised machine learning in astronomy. Neural networks applied to the solar flux emergence problem considerably reduce the manual labelling burden and easily extract higher-level information.