A Machine Learning Approach to Classifying MESSENGER FIPS Proton Spectra
A Machine Learning Approach to Classifying MESSENGER FIPS Proton Spectra
复制标题
DOI：
10.1029/2019ja027352
复制
发表时间：
2020-06
期刊：
影响因子：
--
通讯作者：
M. K. James;S. Imber;J. Raines;T. Yeoman;E. Bunce
中科院分区：
文献类型：
--
作者：
M. K. James;S. Imber;J. Raines;T. Yeoman;E. Bunce
将 κ 分布函数拟合到水星表面、空间环境、地球化学和测距 (MESSENGER) 1 分钟快速成像等离子体光谱仪 (FIPS Andrews et al., 2007, https://doi.org/10.1007/s11214-007-9272-5) 质子光谱的整个数据集，然后使用人工神经网络 (ANN) 进行评估数据拟合的质量。使用下坡单纯形法将 κ 分布函数拟合到每个质子谱，提供密度、n、温度、T 和 κ 参数的估计，该参数控制分布的形状。最终训练的神经网络实现了 96% 的分类准确率，并已用于对 MESSENGER 在水星轨道运行约 4 年期间收集的 1 分钟质子数据集进行分类。在 223,282 个光谱中，约 160,000 个被归类为具有“良好”拟合 κ 分布，其中约 133,000 个是从磁层内获得的测量值，约 18,000 个来自磁鞘。
The κ distribution function is fitted to the entire data set of MErcury Surface, Space ENvironment, GEochemistry and Ranging's (MESSENGER) 1‐min Fast Imaging Plasma Spectrometer (FIPS Andrews et al., 2007, https://doi.org/10.1007/s11214‐007‐9272‐5) proton spectra, and then artificial neural networks (ANNs) are used to assess the quality of this fit to the data. The κ distribution function is fitted to each proton spectrum using the downhill‐simplex method, providing an estimate for density, n, temperature, T, and the κ parameter, which controls the shape of the distribution. The final trained neural network achieved classification accuracy of 96% and has been used to classify the 1‐min proton data set collected during MESSENGER's ∼4 years in orbit of Mercury. Of the 223,282 spectra, ∼160,000 were classified as having “good” fitting κ distributions, ∼133,000 of which were measurements obtained from within the magnetosphere, and ∼18,000 were from the magnetosheath.