Rock classification plays a very important role in geological research. In order to simulate rock classification under unmanned conditions, we selected 13 kinds of rock samples and obtained their classification from spectral information, without any pretreatment, by using laser-induced breakdown spectroscopy (LIBS). Firstly, we collected all the characteristic regions of each rock and used principal component analysis to reduce the dimension of each LIBS spectral signal, to improve the accuracy and speed of the classification algorithm. Secondly, three classification algorithms were used to classify dimension-reduced spectral data, namely linear discriminant analysis, random forest classification and support vector machine (SVM). At the same time, the classification results were evaluated by confusion matrix. The final average classification accuracy was 27%, 91% and 100%, respectively, showing that the SVM algorithm can be applied to the LIBS classification of rocks.

岩石分类在地质研究中起着非常重要的作用。为了模拟无人条件下的岩石分类，我们选择了13种岩石样品，并使用激光诱导击穿光谱法（LIBS）在不进行任何预处理的情况下从光谱信息中获得了它们的分类。首先，我们收集了每个岩石的所有特征区域，并使用主成分分析来减小每个LIBS频谱信号的维数，以提高分类算法的准确性和速度。其次，采用三种分类算法对降维光谱数据进行分类，即线性判别分析，随机森林分类和支持向量机（SVM）。同时，通过混淆矩阵对分类结果进行评估。