Achieving Accurate Rock Classification Using LIBS

A team of researchers from three Chinese institutions, which include the University of Science and Technology of China, the Chinese Academy of Sciences, and Anhui University, recently collaborated on a study that tested a new laser-induced breakdown (LIBS) spectroscopy device for in situ rock classification. This study, which was published in the journal Chemosensors, highlights the continued advancement of spectroscopic instruments to help solve real-world applications (1). In the case of this study, the researchers demonstrated how machine learning (ML)-driven LIBS analysis can be beneficial in geological exploration and petroleum logging (1).

Scientific expedition team examining a significant fault line in a remote mountain region, with geologists taking measurements and samples from the exposed rock layers. Generated with AI. | Image Credit: © tynza - stock.adobe.com

LIBS is an advanced atomic spectroscopy technique. It uses a high-energy laser pulse to create a microplasma on a sample’s surface, breaking it down into its atomic components (2). As the plasma cools, the emitted light is analyzed to determine the elemental composition of the material (2). It has been in use for more than 30 years in various applications when elemental analysis is required (2). Some of these applications include geology, environmental monitoring, and industrial quality control.

Recently, handheld and portable LIBS devices have emerged on the market, opening more opportunities for the technique to be used. By integrating ML, LIBS can achieve even greater accuracy in classification tasks, as demonstrated in recent geological applications (2).

In this study, researchers demonstrate how ML does improve classification accuracy. When it comes to traditional rock classification techniques, the problem researchers encounter is that they often require laboratory analysis, which is time-consuming and involves extensive sample preparation (1). In contrast, LIBS is a laser-based atomic emission spectroscopy technique that enables real-time elemental analysis with minimal sample preparation, making it ideal for field applications (1).

The new LIBS device tested in this study was designed primarily for outdoor environments and applications, such as geological surveys. The device allows for rapid classification of seven common rock types found in oil logging sites: mudstone, basalt, dolomite, sandstone, conglomerate, gypsolyte, and shale (1). Unlike conventional laboratory LIBS systems that rely on high-energy laser pulses and wide-range spectrometers, this portable device is adapted for field conditions where performance is often constrained by lower energy output and limited spectral resolution (1).

Using the LIBS device, the research team compiled spectral data from random areas of each rock sample and applied pre-processing techniques, including normalization, Savitzky-Golay (SG) filtering, and principal component analysis (PCA) for dimensionality reduction (1). Four ML algorithms were tested for classification accuracy: linear discriminant analysis (LDA), k-nearest neighbor (KNN), support vector machine (SVM), and extreme gradient boosting (XGBoost).

All four machine learning algorithms achieved high accuracy, but XGBoost performed the best. LDA achieved 95.71% accuracy, KNN 93.57%, SVM 92.14%, and XGBoost 98.57% (1). Notably, the XGBoost model reached 100% accuracy for the training set and 98.57% for the test set, making it the most reliable approach for in situ rock classification (1).

One of the main challenges in LIBS-based rock analysis is the variability of spectral data in outdoor conditions. Laboratory-based LIBS systems often analyze powdered rock samples with smooth surfaces, leading to higher signal stability and better spectral consistency (1). However, real-world rock samples encountered in the field lack such uniformity, making spectral data more complex and harder to interpret (1).

The newly developed portable LIBS device addresses these challenges through multi-directional spectral acquisition, ensuring that data collected from rough and irregular surfaces remain reliable (1). The integration of PCA further refines spectral features by reducing noise and improving classification accuracy (1).

By enabling real-time rock classification in the field, this portable LIBS system could streamline exploration workflows, reduce costs, and minimize the need for labor-intensive laboratory analyses. The high accuracy of the XGBoost model also suggests that ML can play a crucial role in refining LIBS-based classification techniques (1).

This study, therefore, shows how artificial intelligence (AI) is being integrated into spectroscopic devices. It is expected that this trend will continue in the future, which will help advance more studies in mineral exploration, environmental monitoring, and planetary science.

References

1. Zhang, M.; Fu, H.; Wang, H.; et al. In Situ Classification of Original Rocks by Portable Multi-Directional Laser-Induced Breakdown Spectroscopy Device. Chemosensors 2025, 13 (1), 18. DOI: 10.3390/chemosensors13010018
2. SciAps, LIBS: Handheld Laser Induced Breakdown Spectroscopy. SciAps.com. Available at: https://www.sciaps.com/products/libs/what-is-libs (accessed 2025-03-10).