Chromosphere bright points (CBPs) are small and bright magnetic structures, which are the reflection of the cross sections of the magnetic-flux tubes crossing the chromosphere. Accurate detection and segmentation of CBPs enable large-scale data acquisition and then feature extraction. In this paper, we propose the CLPNet model based on the LPNet architecture. By restructuring the global feature extractor and patch network module, we improve the segmentation accuracy and small-object detection. The Ca II H image-series of the quiet regions from the Solar Optical Telescope (SOT) on board Hinode are used to construct a training set containing of 448 images and approximately 4000 CBPs were constructed, and two test sets, which contain 40 images including 600 CBPs and 50 images including 700 CBPs respectively. The average precision, recall, F1-score of test sets 1 and 2 are 0.846, 0.847 and 0.846, respectively. For the segmentation effect at the pixel level, the average pixel precision, pixel recall and pixel F1-score and mIoU value of test sets 1 and 2 are 0.722, 0.766, 0.742 and 0.640, respectively. This indicates that CLPNet demonstrates a commendable level of efficiency and accuracy in both detection and segmentation tasks.The compactness is used for classifying the morphology of CBPs. Specifically, CBPs exhibiting compactness less than or equal to 1.13 are regarded as point-like CBPs, which could correspond to a single slender flux tube. Otherwise, CBPs are regarded as non-point-like CBPs, which could correspond to the interaction of several slender flux tubes. There are differences in features between point-like and non-point-like CBPs. The mean compactnesses are 1.06 ± 0.04 and 1.20 ± 0.07, respectively. The mean equivalent diameters are 201 ± 50 and 279 ± 60 km, respectively. The mean values of the maximum intensity contrast are 1.14 ± 0.47 and 1.54 ± 0.66 <InlineEquation ID="IEq1"> <EquationSource Format="MATHML"><math> <mo stretchy="false">〈</mo> <msub> <mi>I</mi> <mrow> <mtext>QS</mtext> <mi mathvariant="normal">_</mi> <mtext>Ca</mtext> </mrow> </msub> <mo stretchy="false">〉</mo> </math></EquationSource> <EquationSource Format="TEX">$\langle I_{\text{QS}\_\text{Ca}} \rangle $</EquationSource> </InlineEquation>, respectively. The mean values of eccentricity are 0.60 ± 0.14 and 0.78 ± 0.11, respectively. This research provides a more precise methodology for the comprehensive study of CBPs, offering novel insights into the physical phenomena occurring on the chromosphere and then establishing a quantitative foundation for three-dimensional modeling of magnetic-flux tubes and investigations into coronal heating mechanisms.

Detection and Segmentation for Chromosphere Bright Points by CLPNet

期刊论文
发表日期：

2026年3月24日

查看全文

Xichen Zhang,
Yunfei Yang,
Xiaoli Zhang,
Song Feng,
Wei Dai,
Bo Liang,
Jianping Xiong

摘要

Chromosphere bright points (CBPs) are small and bright magnetic structures, which are the reflection of the cross sections of the magnetic-flux tubes crossing the chromosphere. Accurate detection and segmentation of CBPs enable large-scale data acquisition and then feature extraction. In this paper, we propose the CLPNet model based on the LPNet architecture. By restructuring the global feature extractor and patch network module, we improve the segmentation accuracy and small-object detection. The Ca II H image-series of the quiet regions from the Solar Optical Telescope (SOT) on board Hinode are used to construct a training set containing of 448 images and approximately 4000 CBPs were constructed, and two test sets, which contain 40 images including 600 CBPs and 50 images including 700 CBPs respectively. The average precision, recall, F1-score of test sets 1 and 2 are 0.846, 0.847 and 0.846, respectively. For the segmentation effect at the pixel level, the average pixel precision, pixel recall and pixel F1-score and mIoU value of test sets 1 and 2 are 0.722, 0.766, 0.742 and 0.640, respectively. This indicates that CLPNet demonstrates a commendable level of efficiency and accuracy in both detection and segmentation tasks.

The compactness is used for classifying the morphology of CBPs. Specifically, CBPs exhibiting compactness less than or equal to 1.13 are regarded as point-like CBPs, which could correspond to a single slender flux tube. Otherwise, CBPs are regarded as non-point-like CBPs, which could correspond to the interaction of several slender flux tubes. There are differences in features between point-like and non-point-like CBPs. The mean compactnesses are 1.06 ± 0.04 and 1.20 ± 0.07, respectively. The mean equivalent diameters are 201 ± 50 and 279 ± 60 km, respectively. The mean values of the maximum intensity contrast are 1.14 ± 0.47 and 1.54 ± 0.66 $〈 I_{QS_Ca} 〉$ $\langle I_{\text{QS}\_\text{Ca}} \rangle $ , respectively. The mean values of eccentricity are 0.60 ± 0.14 and 0.78 ± 0.11, respectively. This research provides a more precise methodology for the comprehensive study of CBPs, offering novel insights into the physical phenomena occurring on the chromosphere and then establishing a quantitative foundation for three-dimensional modeling of magnetic-flux tubes and investigations into coronal heating mechanisms.