分析测试学报2025，Vol.44Issue(12)：2478-2485,8.DOI:10.12452/j.fxcsxb.250326231
基于高光谱成像技术的喷墨打印文件形成时间鉴别研究
Hyperspectral Imaging-based Investigation for Identification of the Production Timeline of Inkjet-printed Documents
摘要
Abstract
This study proposes a novel nondestructive methodology for determining the formation time of inkjet-printed documents through the integration of hyperspectral imaging technology with ma-chine learning regression algorithms.Experimental data were collected from chronologically printed samples produced by nine inkjet printers of distinct brands and models.The methodology framework comprises two principal phases:dimensionality reduction of hyperspectral data and subsequent re-gression modeling.In the initial phase,principal component analysis(PCA)and partial least squares regression(PLSR)were employed to address the high dimensionality inherent in hyperspectral datas-ets.The interpretation rate of dimensional reduction of principal components derived from these di-mensionality reduction techniques were systematically quantified,providing preliminary interpret-ability for the model architecture.This phase effectively condensed the hyperspectral information while preserving critical temporal features associated with document aging.Four advanced regression methodologies were subsequently implemented on the dimensionality-reduced data:least absolute shrinkage and selection operator regression(LASSO),partial least squares regression(PLSR),ridge regression(RR),and Bayesian regression(BR).The dataset was partitioned into training and testing subsets at a 1∶4 ratio to ensure robust model validation.Quantitative evaluation metrics revealed ex-ceptional model performance across all regression approaches.The coefficient of determination(R2)values for LASSO-PLSR,PLSR,RR,and BR models consistently approached unity(R2>0.99),indicating near-perfect explanatory power.Furthermore,all models demonstrated minimal error met-rics:mean absolute error(MAE),mean squared error(MSE),and root mean squared error(RMSE)values were all proximate to zero,confirming high predictive accuracy.Statistical signifi-cance testing yielded F-statistic values of substantial magnitude(p<0.001)for all PLSR-based regres-sion models,confirming the exceptional statistical validity of the results.Comparative analysis re-vealed superior performance characteristics in PLSR and BR models relative to LASSO and RR imple-mentations.Further explain the relationship between principal components and spectral bands through the PLSR weight coefficients to clarify the role of key spectral bands.External validation using inde-pendent hyperspectral datasets confirmed the models'generalizability,with R2 values exceeding 0.9(p<0.05)across all test scenarios.Error analysis indicated that residual variations fell within accept-able methodological thresholds,confirming measurement reliability.This investigation establishes that hyperspectral imaging technology exhibits significant potential for forensic document analysis,particularly in temporal authentication of inkjet-printed materials.The implemented machine learning framework successfully decouples temporal signatures from complex spectral data while maintaining complete non-invasiveness.The method proposed in this article provides a new solution for the prob-lem of dating documents,offering substantial improvements over conventional destructive chemical analysis methods.关键词
高光谱成像/机器学习/喷墨打印机/打印文件形成时间Key words
hyperspectral imaging/machine learning/inkjet printer/timeline of inkjet-printed documents分类
化学化工引用本文复制引用
梁塬,崔岚,杨雪颖..基于高光谱成像技术的喷墨打印文件形成时间鉴别研究[J].分析测试学报,2025,44(12):2478-2485,8.基金项目
"十三五"国家重点研发计划资助项目(2016YFC08007005) （2016YFC08007005）
公安部科技强警基础工作计划项目(2022JC03) （2022JC03）
中国刑事警察学院研究生创新能力提升项目(2023YCZD07) （2023YCZD07）
