Aluminum alloys are irreplaceable in key lightweight components of automobiles, aircraft, aerospace vehicles, and ships. The main raw material of aluminum alloys is bauxite, and so the high-precision sorting and identification of bauxite is very important in guaranteeing the performance of aluminum alloys. This article describes a convolutional neural network (CNN) structure that, combined with principal component analysis-assisted laser-induced breakdown spectroscopy, can identify different types of bauxite samples. First, the data collected by a spectrometer are normalized to eliminate the influence of different dimensions on the excitation intensities of each spectral line. The feature dimensionality of the normalized samples is then reduced through principal component analysis. The features of the input data are extracted multiple times through convolution and pooling operations in the CNN. Experimental results show that the classification accuracy under a single convolution and pooling structure reaches 97.4%, whereas that under multiple convolution and pooling structures can reach 99.6%. To evaluate the performance of the proposed model, models based on k-nearest neighbors, random forest, support vector machine, and full-spectrum feature inputs are constructed. The results show that CNNs have great potential in the field of bauxite identification and classification, and provide a reliable data processing method that enables laser-induced breakdown spectroscopy to classify materials with similar chemical properties.