Opt Lett. 2023 Jul 1;48(13):3567-3570. doi: 10.1364/OL.494308.
In this Letter, a rapid origin classification device and method for Baishao (Radix Paeoniae Alba) slices based on auto-focus laser-induced breakdown spectroscopy (LIBS) is proposed. The enhancement of spectral signal intensity and stability through auto-focus was investigated, as were different preprocessing methods, with area normalization (AN) achieving the best results-increasing by 7.74%-but unable to replace the improved spectral signal quality provided by auto-focus. A residual neural network (ResNet) was used as both a classifier and feature extractor, achieving higher classification accuracy than traditional machine learning methods. The effectiveness of auto-focus was elucidated by extracting LIBS features from the last pooling layer output using uniform manifold approximation and projection (UMAP). Our approach demonstrated that auto-focus could efficiently optimize the LIBS signal, providing broad prospects for rapid origin classification of traditional Chinese medicines.
本研究提出了一种基于自动对焦激光诱导击穿光谱（LIBS）的白芍切片快速产地分类装置和方法。研究了自动对焦对光谱信号强度和稳定性的增强作用，以及不同的预处理方法，其中面积归一化（AN）效果最好，提高了 7.74%，但不能替代自动对焦提供的改进光谱信号质量。残差神经网络（ResNet）既作为分类器又作为特征提取器，其分类准确性高于传统的机器学习方法。通过使用一致流形逼近和投影（UMAP）从最后一个池化层输出中提取 LIBS 特征，阐明了自动对焦的有效性。我们的方法证明了自动对焦可以有效地优化 LIBS 信号，为中药材的快速产地分类提供了广阔的前景。
