基于偏最小二乘法的血迹陈旧度高光谱预测研究  

Age imaging by hyperspectral estimation of bloodstains technology based on partial least squares regression

作　　者：孙威 刘怀策 刘金坤 刘玉海 李浩 蔡能斌 陈蕊丽[1] SUN Wei;LIU Huai-ce;LIU Jin-kun;LIU Yu-hai;LI Hao;CAI Neng-bin;CHEN Rui-li(College of Investigation,People’s Public Security University of China,Beijing 100038,China;Shenzhen Wayho Technology Co.,Ltd.,Shenzhen 518108,China;Institute of Criminal Science and Technology,Public Security Bureau of Zhengzhou,Zhengzhou 450016,China;Shanghai Key Laboratory of Crime Scence Evidence,Shanghai 200083,China)

机构地区：[1]中国人民公安大学侦查学院,北京100038 [2]深圳市中达瑞和科技有限公司,广东深圳518108 [3]郑州市公安局刑事科学技术研究所,河南郑州450016 [4]上海市现场物证重点实验室,上海200083

出　　处：《应用化工》2022年第1期273-276,285,共5页Applied Chemical Industry

基　　金：中央高校基本科研业务费(2021JKF103);上海市现场物证重点实验室开放课题基金资助(2021XCWZK02)。

摘　　要：血迹是现场勘查中常见的生物检材,血迹陈旧度与案发时间有着必然的联系。使用高光谱成像仪,以BaSO_(4)白板作为漫反射标准参照板,每间隔1 h采集12个血迹样本在0~10 h共计132份高光谱数据。基于标准正态变换预处理光谱数据,随机选取预处理后的8个血迹样本用作训练集,4个样本用作测试集。分别以全波段和特征波段构建基于偏最小二乘法(PLS)的血迹陈旧度预测模型,同时与以主成分回归(PCR)、支持向量机回归(SVR)、最小二乘支持向量机回归(LS-SVR)算法构建的模型结果进行比较。结果表明,以偏最小二乘法构建的预测模型,训练集决定系数(R^(2)_(c))=0.996,平均绝对误差(MAE_(c))=0.002,测试集决定系数(R^(2)_(p))=0.962,平均绝对误差MAE_(p)=0.010,具有最高的预测精度和稳定性。Blood is a common biological sample in on-the-spot investigation,and there is an inevitable relationship between the obsolescence of blood and the time of the crime.Using hyperspectral imager,using BaSO_(4) whiteboard as the standard reference board of diffuse reflection,12 blood samples were collected every 1 h and a total of 132 hyperspectral data were collected in 0~10 h.Based on the standard normal variate transform to preprocess the spectral data,8 blood samples were randomly selected as the training set and 4 samples were used as the test set.The prediction models of bloodstains age based on partial least square(PLS)were constructed in all bands and feature bands respectively,and the results were compared with those constructed by principal component regression(PCR),support vector machine regression(SVR)and least square support vector machine regression(LS-SVR).The results show that for the prediction model constructed by partial least square method,the determination coefficient of training set is 0.996,the mean absolute error is 0.002,the determination coefficient of test set is 0.962,and the mean absolute error is 0.010,this method has the highest prediction accuracy and stability.

关 键 词：光谱学 高光谱成像技术 血迹陈旧度 偏最小二乘回归 

分 类 号：TQ016[化学工程] O433.4[机械工程—光学工程]