在本研究中，提出了一種基於Spatial Attention U-Net (SA U-Net)神經網路從電離圖中提取電離層回波訊號的方法。電離圖資料由位於台灣花蓮(北緯 23.99°，東經 131.61°)的垂直入射脈衝電離層雷達(Vertical Incidence Pulsed Ionospheric Radar，VIPIR)提供，每天可產生288張電離圖。SA U-Net是一種以圖像分割為目的而開發的卷積神經網路。目前，我們收集了2013年8月至2014年6月的常態數據，使用SA U-Net訓練以及自動處理這些數據，並以各訊號的臨界頻率以及虛擬高度準確度作為判斷SA U-Net性能的方式。最後，我們還對訓練資料進行分類處理，嘗試使用分類後的資料訓練神經網路模型，期望能提高模型對該類電離圖訊號的判斷成功率。

In this study, we propose a method for extracting ionospheric echo signals from ionograms based on Spatial Attention U-Net (SA U-Net) neural network. Ionogram data are provided by the Vertical Incidence Pulsed Ionospheric Radar (VIPIR) located in Hualien, Taiwan (23.99° north latitude, 131.61° east longitude), which can produce 288 ionization maps every day. SA U-Net is a convolutional neural network developed for the purpose of image segmentation. Currently, we have collected normal data from August 2013 to June 2014, used SA U-Net to train and automatically process the data, and used the critical frequency and virtual height accuracy of each signal as a criterion to judge the performance of SA U-Net. Finally, we classify the training data and try to use the classified data to train the neural network model, hoping to improve the judgment success rate of the model.

[1] 周政宏，神經網路-理論與實務，初版，臺北市，1995
[2] 柳銘哲，電磁波在電漿中的傳播，第四十四期，元培科技大學通識中心，PP.59-69，2013
[3] 劉正彥，地球的大氣層與電離層，電離層電波科學實驗室，2018
[4] David K. Cheng, Addison Wesley. Field and Wave Electromagnetics. Pearson Education Limited, 2014.
[5] Kelley, M. C. The Earth's Ionosphere: Plasma Physics and Electrodynamics (2nd ed.). Academic Press. Published 1989.
[6] Antonio Gulli, Sujit Pal. Deep Learning with Keras. Packt Publishing. Released April 2017.
[7] Harris, T. J., & Pederick, L. H. (2017). A robust automatic ionospheric O/X mode separation technique for vertical incidence sounders. Radio Science, 52, 1534–1543.
[8] Guan-Han Huang, Alexei V. Dmitriev, Lung-Chih Tsai, Enkhtuya Tsogtbaatar, Yu-Chi Chang, Mon-Chai Hsieh, Merlin M. Mendoza, Yu-Chiang Lin, Hao-Wei Hsu, and Chia-Hsien Lin. (2022). SA-UNet for The Recovery of Ionograms [Data set]. Kaggle. https://doi.org/10.34740/KAGGLE/DS/1933301 [last accessed July, 2024]
[9] Chang, Y.-C.; Lin, C.-H.; Dmitriev, A.V.; Hsieh, M.-C.; Hsu, H.-W.; Lin, Y.-C.; Mendoza, M.M.; Huang, G.-H.; Tsai, L.-C.; Li, Y.-H.; et al. State-of-the-Art Capability of Convolutional Neural Networks to Distinguish the Signal in the Ionosphere. Sensors 2022, 22, 2758.
[10] Mendoza, M.M.; Chang, Y.-C.; Dmitriev, A.V.; Lin, C.-H.; Tsai, L.-C.; Li, Y.-H.; Hsieh, M.-C.; Hsu, H.-W.; Huang, G.-H.; Lin, Y.-C.; et al. Recovery of Ionospheric Signals Using Fully Convolutional DenseNet and Its Challenges. Sensors 2021, 21, 6482.
[11] Wada, K. (2016). Labelme: Image polygonal annotation with Python. Retrieved from https://github.com/wkentaro/labelme
[12] Golnaz Ghiasi, Tsung-Yi Lin, Quoc V. Le. DropBlock: A regularization method for convolutional networks. Accepted at NIPS 2018. Computer Vision and Pattern Recognition.
[13] Shibani Santurkar, Dimitris Tsipras, Andrew Ilyas, Aleksander Madry. How Does Batch Normalization Help Optimization? NeurIPS'18, Statistics, Machine Learning (stat.ML).
[14] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin. Attention Is All You Need. Computer Science, Computation and Language. (2017)
[15] Meng-Hao Guo, Tian-Xing Xu, Jiang-Jiang Liu, Zheng-Ning Liu, Peng-Tao Jiang, Tai-Jiang Mu, Song-Hai Zhang, Ralph R. Martin, Ming-Ming Cheng, Shi-Min Hu. Attention Mechanisms in Computer Vision: A Survey. Computational Visual Media, 2022.
[16] David E. Rumelhart, Geoffrey E. Hinton & Ronald J. Williams. Learning representations by back-propagating errors. Nature volume 323, pages533–536 (1986).
[17] Changlu Guo, Márton Szemenyei, Yugen Yi, Wenle Wang, Buer Chen, Changqi Fan. SA-UNet: Spatial Attention U-Net for Retinal Vessel Segmentation. ICPR 2020.
[18] Atlas, Homma, and Marks. An Artificial Neural Network for Spatio-Temporal Bipolar Patterns: Application to Phoneme Classification. Neural Information Processing Systems (NIPS 1987).
[19] Chervyakov, N.I.; Lyakhov, P.A.; Deryabin, M.A.; Nagornov, N.N.; Valueva, M.V.; Valuev, G.V. Residue Number System-Based Solution for Reducing the Hardware Cost of a Convolutional Neural Network. Neurocomputing. September 2020.
[20] Sergey Ioffe, Christian Szegedy. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. 11 Feb 2015.
[21] Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-Net: Convolutional Networks for Biomedical Image Segmentation. Computer Science Department and BIOSS Centre for Biological Signalling Studies, University of Freiburg, Germany. 18 May 2015.
[22] Evan Shelhamer, Jonathan Long, Trevor Darrell. Fully Convolutional Networks for Semantic Segmentation. CVPR 2015.
[23] Andrew Lavin, Scott Gray. Fast Algorithms for Convolutional Neural Networks. CVPR 2016.
[24] Hendrycks, Dan; Gimpel, Kevin. "Gaussian Error Linear Units (GELUs)". Trimmed version of 2016 draft. arXiv:1606.08415
[25] Prajit Ramachandran, Barret Zoph, Quoc V. Le. Searching for Activation Functions. ICLR 2018.