迁移成分分析结合直接校正的模型传递方法  

A Model Transfer Method Based on Transfer Component Analysis and Direct Correction

作　　者：李灵巧 王卓健 陈江海 卢丰 黄殿贵[2] 杨辉华 李泉 LI Ling-qiao;WANG Zhuo-jian;CHEN Jiang-hai;LU Feng;HUANG Dian-gui;YANG Hui-hua;LI Quan(School of Computer Science and Information,Guilin University of Electronic Technology,Guilin 541004,China;Guangxi Zhuang Autonomous Region Center for Analysis and Test Research,Nanning 530022,China;School of Artificial Intelligence,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]桂林电子科技大学计算机与信息安全学院,广西桂林541004 [2]广西壮族自治区分析测试研究中心,广西南宁530022 [3]北京邮电大学人工智能学院,北京100876

出　　处：《光谱学与光谱分析》2024年第12期3399-3405,共7页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62262010);桂林电子科技大学研究生创新项目(2023YCXS052)资助。

摘　　要：近红外光谱因快速、不破坏样品等优点,被广泛应用。近红外光谱仪之间存在一致性现象会导致主机模型预测其从机仪器光谱时准确度不够,如果重新建模则会成本较高。为解决上述问题,提出了一种迁移成分分析直接校正算法(TCADS),首先采用改进后的TCA算法对服从不同分布的主机光谱和从机光谱进行变换,将其映射到高维的再生核希伯尔特空间,再将两者的光谱矩阵进行降维,最后对经过TCA变换后的主机光谱和从机光谱再次采用直接校正算法,进一步提高模型传递性能。该算法将非线性校正与线性校正相结合,相比于传统的线性校正算法有效缓解了过校正问题,具有较好的鲁棒性。为验证算法有效性,在公开数据集进行实验,并与传统的直接校正法(DS)、分段直接校正法(PDS)、斜率偏差校正法(SBC)进行比较。研究表明所提出的TCADS算法有效降低了不同仪器之间的光谱差异,相比于传统的模型传递算法进一步提高了模型传递效果,实现了主机上建立的近红外光谱模型在从机上共享。Near-infrared spectroscopy has been widely used due to its high efficiency and non-destructive properties advantages.However,the consistency phenomenon between near-infrared spectrometers can lead to insufficient accuracy when the master model predicts the spectra of its slave instruments.If the calibration model is rebuilt based on the offset spectrum,it will lead to higher consumption.This paper proposes a transfer component analysis direct standardization(TCADS)algorithm to address the above issues.The algorithm initially employs an enhanced TCA algorithm to convert master and enslaved person spectra,which adhere to distinct distributions,by projecting them into high-dimensional reproducing kernel Hilbert space.Subsequently,it reduces the dimensionality of their spectral matrices.Finally,a direct standardization algorithm is reapplied to the master and slave spectra post-TCA transformation,further enhancing the model's transfer performance.This algorithm combines nonlinear correction with linear correction,effectively alleviating the problem of overcorrection compared to traditional linear correction algorithms,and is robust.To verify the effectiveness of the algorithm,experiments were conducted on public datasets and compared with traditional direct standardization(DS),piecewise direct standardization(PDS),and slope and bias correction(SBC)methods.The experiment demonstrates that the TCADS algorithm proposed in this article efficiently minimizes spectral disparities between the master instrument and the slave instrument.This enhancement notably outperforms traditional model transfer algorithms,facilitating the effective sharing of near-infrared spectral models established on the master instrument to the slave instrument.

关 键 词：近红外光谱 模型传递 迁移学习 迁移成分分析 直接校正 

分 类 号：O657.3[理学—分析化学]