| 引用本文: |
陈赵懿1，冯柯1，陈志斌2，杨小强1，李焕良1.增量学习的拉曼光谱识别算法[J].陆军工程大学,2023,(4):62-69
[点击复制] |
|
| |
|
|
| 本文已被：浏览 81次 下载 96次 |
| 增量学习的拉曼光谱识别算法 |
| 陈赵懿1，冯柯1，陈志斌2，杨小强1，李焕良1 |
| (1.陆军工程大学 野战工程学院，江苏 南京 210007;2.陆军研究院，河北 石家庄 050000) |
| 摘要:针对拉曼光谱仪内置数据库样本数量少和新增物质光谱无法识别的问题，提出了一种增量学习的拉曼光谱识别算法。构建了基于空间注意力机制的一维膨胀卷积神经网络（convolutional neural network model，CNN_SA），用数据库已有光谱数据对CNN_SA预训练，提取拉曼光谱特征，与逻辑回归（logistic regression，LG）和支持向量机（support vector machine，SVM）等算法相比，CNN_SA具有更好的识别性能。并采用类别增量的学习方式，将CNN_SA学习的先验知识迁移到新增物质的拉曼光谱识别中，极大提高了新增物质小样本拉曼光谱的识别准确率。 |
| 关键词: SERS 光谱识别 卷积神经网络 注意力机制 增量学习 |
| DOI：10.12018/j.issn.2097-0730.20220923001 |
| 投稿时间：2022-09-23 |
| 基金项目:军内科研项目(LY20202GK005) |
|
| Raman Spectrum Identification Algorithm Based on Incremental Learning |
| CHEN Zhaoyi1，FENG Ke1，CHEN Zhibin2，YANG Xiaoqiang1，LI Huanliang1 |
| (1.College of Field Engineering，Army Engineering University of PLA，Nanjing 210007，China;2.Army Research Institute，Shijiazhuang 050000，China) |
| Abstract: In view of the small number of samples in the built-in database of Raman spectrometer and the inability to identify the spectra of new substances, a Raman spectrum identification algorithm based on incremental learning is proposed. Firstly, a convolutional neural network model (CNN_SA) based on spatial attention mechanism was constructed, and the existing spectral data in the database was used to pre-train CNN_SA to extract the Raman spectral characteristics. Compared with the traditional algorithms, such as logistic regression (LG) and support vector machine （SVM）, CNN_SA has better recognition performance. In addition, with class increment learning, the prior knowledge learned by CNN_SA can be transferred to the Raman spectrum recognition of the newly added substances, which greatly increases the identification accuracy of the Raman spectra of the newly added substances. |
| Key words: surface enhanced Raman scattering (SERS) spectrum identification convolutional neural network (CNN) attention mechanism incremental learning |
|
