One-Dimensional Deep Convolutional Neural Network for Mineral Classification from Raman Spectroscopy

Raman spectroscopy is often used for the composition determination and rapid classification of materials because it can reflect the molecular information of materials. Its accuracy mainly depends on the performance of the classification algorithm. In addition to classic machine learning classifiers such as support vector machines and k-nearest neighbor, 1D convolutional neural networks (CNNs) have also been applied to recognize Raman spectra. However, most of the research on 1D CNNs is still in the shallow, simple structure of the network model, and its application scope is only for the classification of a few classes. Therefore, this paper proposes a spectral data classification model based on a 1D deep CNN for classifying and recognizing hundreds of classes. We used the RRUFF Raman spectrum database containing many mineral samples to construct two sub-datasets, which were used to evaluate the classification effect of the model from different perspectives without data enhancement. The experimental results show that the model has a high recognition accuracy for data sets with hundreds of classes and sufficient samples. Compared with the classic machine learning classification algorithm and the 1D CNN models proposed by other scholars, our model has higher recognition accuracy and better performance and has better applicability to datasets with thousands of classes and imbalanced class distribution.