Low-cost smartphone-based LIBS combined with deep learning image processing for accurate lithology recognition

Small-sample stacking model for qualitative analysis of aluminum alloys based on femtosecond laser-induced breakdown spectroscopy

Material characterization using laser-induced breakdown spectroscopy (LIBS) often relies on extensive data for effective analysis. However, data acquisition can be challenging, and the high dimensionality of raw spectral data combined with a large-scale sample dataset can strain computational resources. In this study, we propose a small sample size stacking model based on femtosecond LIBS to achieve accurate qualitative analysis of aluminum alloys. The proposed three-layer stacking algorithm performs data reconstruction and feature extraction to enhance the analysis. In the first layer, random forest spectral feature selection and specific spectral line spreading are employed to reconstruct the data. The second layer utilizes three heterogeneous classifiers to extract features from the reconstructed spectra in different feature spaces, generating second-level reconstructed data. Finally, the third layer utilizes the reconstructed dataset for qualitative prediction. Results indicate that the Stacking algorithm outperforms traditional methods such as k-nearest neighbors (KNN), support vector machine (SVM), and random forest (RF), including those combined with principal component analysis (PCA). The Stacking algorithm achieves an impressive 100% recognition rate in classification, with Accuracy, precision, recall, and F1 scores reaching 1.0. Moreover, as the number of samples decreases, the gap between the recognition accuracy of the Stacking algorithm and traditional approaches widens. For instance, using only 15 spectra for training, the Stacking algorithm achieves a recognition accuracy of 96.47%, significantly surpassing the improved RF's accuracy of 71.76%. Notably, the model demonstrates strong robustness compared to traditional modeling approaches, and the qualitative prediction error remains consistently below 5%. These findings underscore the model's enhanced generalization ability and higher prediction accuracy in small sample machine learning. This research contributes significantly to improving the applicability of the LIBS technique for fast detection and analysis of small samples. It provides valuable insights into the development of effective methodologies for material characterization, paving the way for advancements in the field.

Visualization and accuracy improvement of soil classification using laser-induced breakdown spectroscopy with deep learning

Deep learning method is applied to spectral detection due to the advantage of not needing feature engineering. In this work, the deep neural network (DNN) model is designed to perform data mining on the laser-induced breakdown spectroscopy (LIBS) spectra of the ore. The potential of heat diffusion for an affinity-based transition embedding model is first used to perform nonlinear mapping of fully connected layer data in the DNN model. Compared with traditional methods, the DNN model has the highest recognition accuracy rate (75.92%). A training set update method based on DNN output is proposed, and the final model has a recognition accuracy of 85.54%. The method of training set update proposed in this work can not only obtain the sample labels quickly but also improve the accuracy of deep learning models. The results demonstrate that LIBS combined with the DNN model is a valuable tool for ore classification at a high accuracy rate.

Automatic rock classification of LIBS combined with 1DCNN based on an improved Bayesian optimization

To achieve automated rock classification and improve classification accuracy, this work discusses an investigation of the combination of laser-induced breakdown spectroscopy (LIBS) and the use of one-dimensional convolutional neural networks (1DCNNs). As a result, in this paper, an improved Bayesian optimization (BO) algorithm has been proposed where the algorithm has been applied to automatic rock classification, using LIBS and 1DCNN to improve the efficiency of rock structure analysis being carried out. Compared to other algorithms, the improved BO method discussed here allows for a reduction of the modeling time by about 65% and can achieve 99.33% and 99.00% for the validation and test sets of 1DCNN.