Machine learning for inversing FeO and TiO2 content on the Moon: Method and comparison

1Denggao Qiu,1Fei Li,1Jianguo Yan,1Wutong Gao,1Zheng Chong
Icarus (in Press) Link to Article []
1State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430070, China
Copyright Elsevier

The FeO and TiO2 contents are critical for distinguishing petrological properties of the Moon and for studying the distribution of the lunar maria and its multi-period volcanic activity. Traditional methods used the ratio between spectral reflectances to estimate FeO and TiO2 contents, which are empirical models. The development of machine learning algorithms offered new ideas for solving inversion problems, and these algorithms can automatically mine the data for potential correlations wherever possible. In this work, by using the Kaguya Multiband Imager data, we construct an optimized spectral inversion model using the Convolutional Neural Network (CNN) algorithm to produce a map of the FeO and TiO2 content on the lunar surface. The CNN models were compared with the traditional linear model and the Random Forest (RF) model. The results were indicated that the CNN models had higher accuracy and the CNN model eliminated the shortcoming of the RF model that the inversion results were limited by the training data, and certainly optimizes the impact of data striping. The CNN models can better describe the nonlinear relationship between spectral reflectance and oxide content. This also provides the basis for the inversion of the other oxides (e.g., MgO, Al2O3, CaO and SiO2). These new maps from the CNN model provide reference information for further studies of the geological evolution of the Moon.


Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s