Rapid authentication of soil heterogeneity at different depths via laser-induced breakdown spectroscopy

Laser-induced breakdown spectroscopy (LIBS), as a localized remote sensing technique, is an emerging method for analyzing the composition of soil samples. However, studies leveraging LIBS spectral data to explore soil heterogeneity across different depths remain limited. In this study, we applied LIBS combined with machine learning to estimate the presence of some chemical elements at varying depths. The analysis detected significant concentrations of elements such as Ca, Mg, and N, revealing spatial heterogeneity in nitrogen content across ten soil layers. Back propagation (BP) neural network, random forest algorithm, and convolutional neural network were applied for soil depth classification and regression analysis. BP achieved the highest F1 score of 0.9125 in classification and the lowest root mean squared error (RMSE) of 9.8663 cm in regression, demonstrating its superiority in soil spectral analysis. Dimensionality reduction was performed using principal component analysis, linear discriminant analysis (LDA), and SelectKBest, followed by regression with BP. LDA combined with BP achieved the best performance, with an RMSE of 8.7742 cm. The integration of LIBS, LDA, and BP provides an efficient solution to contribute to a rapid and precise identification of different abundant chemical elements in the soil at different depths.