Multi-channel coronal hole detection with convolutional neural networks (2104.14313v2)

Abstract: We develop a reliable, fully automatic method for the detection of coronal holes, that provides consistent full-disk segmentation maps over the full solar cycle and can perform in real-time. We use a convolutional neural network to identify the boundaries of coronal holes from the seven EUV channels of the Atmospheric Imaging Assembly (AIA) as well as from line-of-sight magnetograms from the Helioseismic and Magnetic Imager (HMI) onboard the Solar Dynamics Observatory (SDO). For our primary model (Coronal Hole RecOgnition Neural Network Over multi-Spectral-data; CHRONNOS) we use a progressively growing network approach that allows for efficient training, provides detailed segmentation maps and takes relations across the full solar-disk into account. We provide a thorough evaluation for performance, reliability and consistency by comparing the model results to an independent manually curated test set. Our model shows good agreement to the manual labels with an intersection-over-union (IoU) of 0.63. From the total of 261 coronal holes with an area $>1.5\cdot10^{10}$ km$^2$ identified during the time range 11/2010 - 12/2016, 98.1% were correctly detected by our model. The evaluation over almost the full solar cycle no. 24 shows that our model provides reliable coronal hole detections, independent of the level of solar activity. From the direct comparison over short time scales of days to weeks, we find that our model exceeds human performance in terms of consistency and reliability. In addition, we train our model to identify coronal holes from each channel separately and show that the neural network provides the best performance with the combined channel information, but that coronal hole segmentation maps can be also obtained solely from line-of-sight magnetograms.

• The paper introduces CHRONNOS, a CNN architecture that progressively grows to accurately segment coronal holes using multi-channel solar data.
• It integrates seven EUV channels and magnetograms to achieve 98.1% detection accuracy and an IoU of 0.63, outperforming manual methods.
• The framework shows temporal stability, correlating coronal hole areas with solar activity over an entire solar cycle.

Multi-Channel Coronal Hole Detection with Convolutional Neural Networks

Introduction

The paper "Multi-channel coronal hole detection with convolutional neural networks" outlines a novel approach to accurately and automatically detect coronal holes using a deep learning framework. Utilizing the Atmospheric Imaging Assembly (AIA) and Helioseismic and Magnetic Imager (HMI) data from the Solar Dynamics Observatory (SDO), the research successfully develops a Convolutional Neural Network (CNN) architecture known as CHRONNOS. This system demonstrates robust coronal hole segmentation across varied solar conditions and phases over an entire solar cycle.

Method

The primary method involves a progressively growing neural network architecture, which allows for efficient training and precise segmentation by gradually increasing the resolution level during training. This is implemented via the CHRONNOS model:

• Progressive Growth: The model begins with a low-resolution image of 8x8 pixels, incrementally scaling up to 512x512 pixels (Figure 1). This strategy ensures computational efficiency and promotes convergence to a stable solution by allowing deeper layers to learn global context before moving on to finer spatial details.

  Figure 1: Training procedure of the progressively growing architecture (CHRONNOS), showcasing both the incremental inclusion of new ConvBlocks and transitions through varying levels of resolution.

• Multi-Channel Input: CHRONNOS integrates data from seven EUV channels and LOS magnetograms to enhance the robustness of the segmentation maps, allowing for a nuanced understanding of the solar disk.

The Single Channel Analyzing Network (SCAN) complements this by evaluating each channel's individual contribution to the detection process, thereby offering insights into their relative importance.

Data Set and Preprocessing

Data for training and evaluation includes EUV images from AIA and magnetograms from HMI. The dataset is divided temporally to prevent overfitting, using the last two months of each year as a test set. The data undergoes rigorous preprocessing, including spatial normalization and exposure correction, ensuring consistent quality across observations.

Evaluation and Results

The model's efficacy is benchmarked against independent manually curated datasets, exhibiting high reliability with a pixel-wise Intersection-over-Union (IoU) of 0.63 and a True Skill Statistic (TSS) of 0.81. Remarkably, the CHRONNOS model detected 98.1% of coronal holes correctly, outperforming even human classifications and other semi-automatic labeling methods.

Figure 2: The IoU calculation demonstrates overlapping area determination between segmentation maps, validating model accuracy.

The temporal stability of the CHRONNOS detection is asserted through its strong anti-correlation with solar activity cycles (coronal hole area vs. sunspot number correlation coefficient $r = -0.88$), confirming its consistency over solar cycle no. 24.

Channel Analysis and Interpretations

The SCAN models provide detailed analysis on the information contribution of each spectral channel. The 193 Å channel shows superior performance among individual channels, aligning with its design to capture quiet corona conditions effectively. Conversely, the magnetogram analysis, though less precise, offers promising pathways for detecting coronal structures using magnetic field information alone, facilitating ground-based solar monitoring.

Figure 3: Example of progressive resolution increases demonstrating improvement in boundary precision and identifying previously undetectable regions.

Conclusion

The development of the CHRONNOS model represents a significant step in solar coronal observations, combining deep learning techniques with multi-channel solar image data to offer a fast, reliable, and autonomous method for coronal hole detection. This framework not only enhances operational space weather forecasting but also supports longitudinal studies of solar cycle dependencies, offering potential advancements in solar physics and coronal research. Future directions involve refining the model for ground-based data and extending its application scope in varied astronomical settings.