Rapid identification of soil types in agricultural land based on laser induced breakdown spectroscopy (LIBS)

Abstract

The difference in the nutrient content and harmful heavy metal content of different regions and types of soils can have a certain impact on the crops grown on them. Different crop varieties have their own ecologically suitable areas for planting. Identification of the types of agricultural soil from different regions, can quickly providing references for the selection and cultivation of crops on different types/regions of soil, which is conducive to ensuring the safety of agricultural products. This study is carried out using a laser-induced breakdown spectra (LIBS) device with a frequency-doubled 532 nm excitation wavelength, a laser pulse width of 8 ns, a repetition frequency of 5 Hz, and a pulsed laser energy of 15 mJ. The LIBS data of nine standard agricultural soils collected from different regions. Each sample is collected 60 sets of spectra and a total of 540 sets of data are obtained. At first, the all obtained spectral data are normalized to compensate for the spectral changes in the measurement process. Then the spectral data in the 200–800 nm band are analyzed. Randomly select 30 %, 50 %, and 70 % of the spectral data according to the Kennard-Stone (KS) classification as the training set, and the remaining spectral data as the test set. The Support Vector Machine (SVM) model is used to classify and identify the soil samples of agricultural soil in different areas. The results show that when 30 % of the spectral data are selected as the training set, the accuracy of identifying different agricultural soil types is 88.6 %, and the accuracy of identifying all nine types of agricultural soils is above 76.7 %. The accuracy of soil identification for the two types of agricultural soil is 100 %, and the identification accuracy for the seven types of soil is above 86 %. When 50 % spectral data are selected as training set, the overall recognition accuracy can reach 95.9 %, the recognition accuracy of 8 types of agricultural soil samples is above 90.5 %, and the recognition accuracy of 4 types of soils can reach 100 %, The recognition accuracy of only one type of soil is less than 90 %, which is 86.7 %. When the selected training set contains 70 % of spectral data, the overall recognition accuracy can reach 96.3 %, the recognition accuracy of 8 types of agricultural soil samples is above 90 %, and the accuracy of 1 type of agricultural soil recognition is less than 90 %, which is 89.19 %. The results show that LIBS technology can be applied to the rapid identification of soil types in agricultural soils.