基于红外光谱和AdaBoost算法的印泥紫外老化状态定量分析

Quantitative Analysis of the UV Light Aging Condition of Inkpad Based on Infrared Spectroscopy and AdaBoost Algorithm

针对紫外老化文件,设计一套基于红外光谱分析和机器学习的印泥紫外老化时间分析方法,构建不同老化程度的印泥-红外光谱数据库,使用多元散射校正(MSC)、标准正态变量变换(SNV)和Savitzky-Golay卷积平滑(SG)3种方法对光谱数据进行平滑处理以提高信噪比;利用AdaBoost算法建立印泥紫外老化时间回归模型,并通过网格搜索法对模型参数调优;将优化后的最佳模型与支撑向量机回归、随机森林回归和梯度增强回归等方法进行对比。实验结果显示,经SNV处理后的印泥红外光谱数据建模效果优于MSC和SG预处理的数据;选择决策树模型作为AdaBoost算法的基础模型,决策树深度d=4时,决策树数量n≥50即可获得最佳表现,随着决策树深度增加,最佳模型需要的决策树数量相应降低;AdaBoost模型的最佳均方误差、相对绝对误差、决定系数与可释方差分别为0、0、1、1,与对比算法相比,各项指标均存在显著提升。

This paper presents a method for analyzing the UV light aging time of inkpads using infrared spectroscopy coupled with machine learning techniques.A spectral database reflecting various degrees of aging was established,and the spectral data were refined using multiplicative scatter correction(MSC),standard normal variate transform(SNV),and Savitzky-Golay convolution smoothing(SG)to enhance the signal-to-noise ratio.An AdaBoost regression model for predicting the UV light aging time of inkpads was developed,with its parameters optimized through a grid search approach.This optimized model was benchmarked against support vector regression,random forest regression,and gradient boosting regression models.The study found that the SNV preprocessed infrared spectrum yielded the most accurate modeling results,outperforming those preprocessed with MSC and SG.The AdaBoost algorithm performed optimally with a decision tree depth of 4and when the number of trees was 50or more.As the decision tree depth increased,the optimized model required fewer trees.The AdaBoost model achieved perfect scores in mean square error,relative absolute error,coefficient of determination,and explainable variance,all of which were significantly better than the comparative algorithms.