Determination of soil source using laser induced breakdown spectroscopy combined with feature selection

Abstract

Determining the soil source is crucial for agricultural planning, forensic case analysis, and archaeological site research. This study used laser induced breakdown spectroscopy (LIBS) technology combined with convolutional neural network (CNN) algorithm to determine the soil source. The experiment collected ten soil samples from different regions and extracted soil spectrum data using LIBS technology. In this study, CNN and random forest (RF) algorithms are used to analyze the data. To improve the accuracy of the model, the mean decrease accuracy (MDA) and mean decrease impurity (MDI) feature selection methods of RF are used to filter the data. Four models were constructed using CNN and RF: MDA-CNN, MDA-RF, MDI-CNN, and MDI-RF, and applied to predict soil sources. The experimental results revealed that the MDA-CNN model performs the best with the accuracy of 94.61%, precision rate of 0.9512, recall rate of 0.9461, and F1 score of 0.9487. The experimental results indicate that this analysis method can effectively determine the soil source, which holds significant implications for the development of soil source determination technology.