Quantitative analysis of major elements in logging cuttings using laser-induced breakdown spectroscopy and a...

Laser-induced breakdown spectroscopy (LIBS) has emerged as a new technique for elemental logging because it allows for the simultaneous detection of multiple elements rapidly and with almost no loss. However, the application of this technology in logging cuttings still faces significant challenges. Complex logging conditions and contaminated core samples can lead to significant baseline drift and noise, obscuring useful information in element signatures and reducing the accuracy of quantitative analysis of key elements in cuttings. To enhance the accuracy and robustness of LIBS, this study proposes a novel method that combines LIBS with the Dual-Scale Attention Residual Network (DSARNet) model. LIBS data from 603 samples were collected and preprocessed through smoothing and denoising, baseline correction, and dimensionality transformation prior to modeling. The prediction model utilises the DSARNet model, which primarily comprises two components: a shallow feature fusion module and a deep feature extraction module. The shallow feature fusion module consists of two parallel convolutional branches operating at different scales and an attention-based feature fusion module. The deep feature extraction module comprises two parallel symmetric residual branches and a prediction section. The experimental results show that, compared with traditional machine learning algorithms and the ResNet deep learning model, the DSARNet model performs the best in terms of both root mean square error (RMSE) and R² values. The R² values for the predicted values of Na, Al, Fe, P, and S are 0.9267, 0.9542, 0.9066, 0.9454 and 0.8372, respectively. This indicates a significant improvement in performance and confirms that this technology has broad application prospects and great potential for widespread adoption in the field of core logging.