差分拉曼光谱结合RCSC-Transformer对快递面单的检验研究

Differential Raman Spectroscopy Combined with RCSC and Improved Transformer for Courier Face Sheet Inspection Research

关键词:  差分拉曼光谱  快递面单  正交约束主成分分析  元素比例-余弦相似度聚类  Transformer  

Key Words:Differential Raman Spectroscopy  Express Face Sheet  Orthogonally Constrained Principal Component Analysis  Elemental Ratio-Cosine Similarity Clustering  Transformer  

摘要:快递面单作为案件现场常见物证，其材料成分分析对溯源具有重要意义。针对热敏纸类快递面单字迹易消退、填料成分稳定的特点，提出一种结合差分拉曼光谱、改进正交约束PCA与元素比例-余弦相似度聚类（RCSC）的方法，并引入稀疏注意力机制的Transformer模型进行分类预测。采集173个不同品牌及打印时间的快递面单样本，利用正交约束PCA将光谱数据维度压缩95.6%，利用RCSC聚类并结合人工分类为4类。进一步采用稀疏注意力Transformer模型进行分类，准确率提升至90%，显著优于随机森林、支持向量机等传统方法，提供了高效可靠的检验预测方法。

Abstract:As a common type of physical evidence at crime scenes, express delivery labels hold significant value for traceability through material composition analysis. Addressing the challenges of easily fading handwriting and stable filler components in thermal paper-based labels, this study proposes a novel method integrating differential Raman spectroscopy, modified orthogonally constrained principal component analysis (PCA), and element ratio-cosine similarity clustering (RCSC), combined with a Transformer model incorporating a sparse attention mechanism for classification prediction. A total of 173 express delivery label samples from various brands and printing dates were collected. Orthogonally constrained PCA reduced the spectral data dimensionality by 95.6%, while RCSC clustering, supplemented by manual validation, categorized the samples into four classes. Further classification using the sparse attention-based Transformer model achieved an accuracy of 90%, significantly outperforming traditional methods such as random forest and support vector machines. This approach provides an efficient and reliable framework for forensic examination and predictive analysis.