Adaptive Particle Swarm Optimization Radial Basis Neural Network (APSO-RBF)-Based Method for Classifying Soils by Laser-Induced Breakdown Spectroscopy

As soil is an important natural resource on the earth's surface, the composition and characterization of soil have a significant impact on agricultural production, the ecological environment, and human health. Traditional soil identification methods need to deal with a large number of samples and complex chemical analysis, which requires a lot of time and effort. In this paper, a method combining laser-induced breakdown spectroscopy (LIBS) and adaptive particle swarm optimization radial basis neural network (APSO-RBF) is proposed to classify and identify soil standard samples from different geographical regions. By selecting the appropriate principal component of LIBS spectral data as input, the computational complexity can be reduced, the redundancy of the original spectral data can be reduced, and the samples can be classified quickly and accurately. For the soil from 10 different regions, the first 6 principal components with the highest contribution rate in principal component analysis were used as the input of APSO-RBF classification model, and the classification accuracy of the test set could reach 98.81%. In comparison with the back propagation (BP) algorithm, back propagation based on adaptive particle swarm optimization (APSO-RBF) algorithm and radial basis function neural network (RBF) algorithm, the powerful classification performance of the model is verified. The results show that LIBS technology greatly improved the accuracy of soil identification in different regions with the help of APSO-RBF model.