| |
 |
| Paper: |
Pixel-Level Solar Filament Segmentation Based on Deep Learning |
| Volume: |
527, Astronomical Data Analysis Software and Systems XXIX |
| Page: |
159 |
| Authors: |
Zhu, G.; Lin, G.; Wang, D.; Liu, S.; Yang, X. |
| Abstract: |
In previous research work, we have used the improved U-Net network to recognize solar filaments in Hα full-disk images. However, there are still some limitations in the previous improved network. It has the risk of falling into a local optimum, which will lead to retraining the model. In this paper we introduce batch normalization layers into the network. At the same time the learning rate decline strategy is applied to the network. Our method will greatly reduce the possibility of falling into a local optimum during network training. The loss of the network is reduced to 0.0017, and the accuracy was improved to 0.9993. |
|
|
|
|
|
