摘要
为探索光谱融合结合深度学习对玉米成分定量检测的可行性,针对80个玉米样本的原始、一阶导数、二阶导数光谱和前3类的串行融合光谱分别构建一维卷积神经网络(one-dimensional convolution neural network,1D-CNN)模型,对样本中水分、油脂、蛋白质和淀粉4种成分含量进行定量建模。结果表明,基于串行融合光谱的1D-CNN的4种成分模型性能指标均优于单独基于一种光谱的模型。与传统偏最小二乘回归和支持向量机回归对比,所建立的定量模型性能均为最优。针对测试集,4种成分模型的决定系数和均方根误差分别为0.956和0.211、0.972和0.118、0.982和0.239、0.949和0.428。实验结果表明,串行光谱融合结合卷积神经网络的方法能够充分挖掘光谱所蕴含的信息,增强模型预测能力,为近红外光谱定量分析提供新思路。
In order to explore the feasibility of spectral fusion combined with deep learning for quantitative detection of maize components,one-dimensional convolution neural network(1D-CNN)models were constructed for the original,first-order derivative,second-order derivative spectra and the first three types of serial fusion spectra of 80 maize samples,then quantitative regression models of four components of moisture,oil,protein and starch in maize samples were built.The results showed that the performance of the four component models of the 1D-CNN based on serial fusion spectra were all superior to the other three models based on a single spectrum.Compared with the traditional partial least squares and support vector machine regression,the performance of the quantitative modes established by this method is optimal.For the test set,the coefficients of determination and root mean square errors of the four component models were 0.956 and 0.211,0.972 and 0.118,0.982 and 0.239,0.949 and 0.428,respectively.The experimental results showed that the method of serial spectrum fusion combined with convolutional neural network can fully mine the information contained in the spectrum,thus to enhance the model prediction ability,which provides a new idea for the quantitative analysis of near infrared spectroscopy.
作者
谈爱玲
王晓斯
楚振原
赵勇
TAN Ailing;WANG Xiaosi;CHU Zhenyuan;ZHAO Yong(School of Information and Science Engineering,Yanshan University,The Key Laboratory for Special Fiber and Fiber Sensor of Hebei Province,Qinhuangdao 066004,China;School of Electrical Engineering,Yanshan University,The Key Laboratory of Measurement Technology and Instrumentation of Hebei Province,Qinhuangdao 066004,China)
出处
《食品与发酵工业》
CAS
CSCD
北大核心
2020年第23期213-219,共7页
Food and Fermentation Industries
基金
国家重点研发计划项目(2019YFC1407904)
河北省自然科学基金项目(C2020203010)
河北省科技计划支撑项目(19975704D)
燕山大学博士基金项目(B779)。
关键词
近红外光谱
玉米成分
深度学习
一维卷积神经网络
光谱融合
near infrared spectroscopy
maize components
deep learning
one-dimensional convolution neural network
spectrum fusion
