Study On Detection Method Of Moisture Content In Livestock And Poultry Feed Based On Terahertz/Near-infrared Spectroscopy

Feed moisture content is the key factor to evaluate feed quality,and is closely related to the calculation of various nutritional components in feed products.In order to ensure the quality and safety of feed,it is necessary to strictly control the feed moisture content within a certain range in the production and storage of feed.In view of the shortcomings of traditional detection methods of feed moisture content,such as long detection time,complex operation and high energy consumption,it is urgent to explore an accurate,fast and energy-saving detection method.Based on terahertz,near infrared spectroscopy and data fusion technology,this paper studies and explores the detection method of water content in livestock and poultry feed.The main contents are as follows.(1)The prediction model of feed moisture content is established based on terahertz(THz)spectroscopy.The research object is 108 feed samples,and the THz spectra of the samples are collected.Firstly,three methods are used to divide the sample set.Secondly,seven preprocessing methods are used to preprocess the spectra.Thirdly,three algorithms are used to extract the characteristic variables of the sample spectrum,and the moisture content prediction model is established combined with partial least squares regression(PLSR).Finally,the correlation coefficients(Rc and Rp),root mean square errors(RMSEC and RMSEP)and residual predictive deviation(RPD)are used to evaluate the prediction effect of the model.The results showed that the optimal models Rc,RMSEC,Rp,RMSEP and RPD for THz prediction of feed moisture content were 0.9807,0.0119,0.9811,0.0093 and 6.1613 respectively.The results show that this method has better prediction effect and can predict feed moisture content quickly and accurately,which provides a reference method for follow-up research.(2)A prediction model of feed moisture content based on near-infrared(NIR)spectroscopy was established.The NIR spectra of 324 feed samples were collected.Firstly,the abnormal samples are removed and three methods are used to divide the sample set.Then,the spectra were pretreated by seven methods.Next,three algorithms are used to extract the characteristic variables of the sample spectrum,and the water content prediction model is established combined with PLSR algorithm.Finally,the evaluation parameters Rc,Rp,RMSEC,RMSEP and the ratio of standard error of correction set to standard error of verification set(SEC/SEP)are used to evaluate the prediction effect of the model.The results showed that the optimal models Rc,RMSEC,Rp,RMSEP and SEC/SEP for NIR prediction of feed moisture content were 0.9879,0.0085,0.9855,0.0095 and 0.8854 respectively.The results show that this method has better prediction effect,can predict feed moisture content quickly and calibrated,and provides a reference method for follow-up research.(3)In order to improve the prediction effect of feed moisture content model,this paper proposes to construct feed moisture content prediction model by THz and NIR spectral data layer fusion(LLDF)and feature layer fusion(MLDF).Firstly,THz and NIR spectral data of 108 feed samples were collected and preprocessed.Then,the spectral data were fused through head to tail splicing,and the data layer feed moisture content prediction model was established combined with PLSR.Next,the spectral characteristic variables are extracted for feature layer fusion,and the feed moisture content prediction model of feature layer is established combined with PLSR.Finally,Rp,RMSEP and RPD are used to evaluate the prediction effect of the model.The results show that the prediction effect of feature layer spectral data fusion model is the best,and Rp,RMSEP and RPD reach 0.9933,0.0069 and 8.7386 respectively.The Rp,RMSEP and RPD of the feature layer fusion model are improved compared with the prediction models established by single THz and single NIR feature variables.The results show that compared with the feed moisture content prediction model established separately by THz or NIR spectral technology,the THz and NIR spectral data fusion prediction model can reflect more feed moisture concentration information and predict feed moisture content more accurately,which provides some theoretical and technical support for new ideas and methods for quantitative analysis of feed moisture content of livestock and poultry.