Hybridization of partial least squares and neural network models for quantifying lunar surface minerals

► A hybridized PLS-BPNN model is developed for quantifying lunar surface minerals. ► The model is validated with the LSCC dataset and additional 12 Apollo samples. ► The model can assess minerals and chemistry for Apollo soils non-destructively. ► The model is suitable for hyper- and multi-spectral data to quantify minerals. ► We try to understand the physical meaning of BPNN by analyzing loading curves.