喉区极光的机器识别

喉区极光是一种发生在电离层对流喉区附近的极光现象，是极光卵向低纬侧延伸出的南北向分立结构，其可能对应由磁鞘高速流与磁层顶作用引发的磁层顶重联过程.喉区极光研究对深入理解太阳风—磁层—电离层耦合过程具有重要意义.从长期观测所积累的大量全天空极光观测数据中准确高效识别出喉区极光结构，是开展喉区极光统计研究的基础.本文利用北极黄河站2003—2017年全天空成像仪的极光观测数据，建立了喉区极光图像标注数据集；基于密集连接卷积神经网络（DenseNet）对极光图像全局高维表征的自动学习，首次实现了喉区极光结构的机器识别.算法对喉区极光识别准确率达96%，且具有良好的泛化性能.研究表明基于深度学习的图像识别方法可用于从海量极光观测数据中自动识别喉区极光事件. 

Machine Identification of Throat Aurora

Throat aurora is an auroral form frequently observed nearby the ionospheric convection throat region. Extending from the equatorward edge of the dayside auroral oval, and thus appears to be a north-south aligned discrete auroral structure. It is suggested to be the projection of the magnetopause reconnection process caused by the interaction between magnetopause and high-speed jets in magnetosheath. The investigation on throat aurora is meaningful for understanding solar wind-magnetosphere coupling process. It is the fundamental work of statistic study that identifying accurately and efficiently these special auroral structures from plenty of ASI images. Through preprocessing and labeling ASI images from Yellow River Station during 2003 and 2017, the throat auroral dataset is established, and the feature space of the auroral data is explored by using DenseNet and classified simultaneously based on whether an observation image has a structure of throat aurora, through which machine identification of throat auroral is achieved for the first time. The accuracy of the identification model is close to 96% and it has good generalization performance.