| Wheat quality is susceptible to variation due to factors such as grain harvest time,moisture content,variety genetics,planting area and impurity content.According to Chinese standards,grain and oil storage units are required to record and classify the different varieties,origins and freshness of wheat for storage,and prohibit wheat containing excessive water and impurities from being put into storage.Wheat of different quality should be stored and sold classically,however,at present,when the grain is stored and traded,there is often a phenomenon that wheat of inferior quality is mixed into storage,which not only harms the interests of consumers,but also increases the difficulty of supervising the quality of grain storage,meanwhile,affects the national food security.Therefore,it is of great practical significance to test the incoming quality of wheat.Taking wheat as the experimental object,this paper used terahertz time-domain spectroscopy and imaging technology to detect the quality of wheat,such as its varieties,origin,freshness,moisture and impurities.The terahertz spectral characteristics of wheat of different varieties,origins,freshness,moisture and impurities were studied separately,and the predictive ability of different pre-processing and feature extraction methods on the model was compared and analyzed.A wheat incoming quality detection model based on terahertz time-domain spectroscopy and imaging technology was established,which laid a foundation for the establishment of a wheat incoming quality detection system.At the same time,it also provides effective technical guidance for trade settlement and grain purchase and storage.The main research contents and results of the paper are as follows:(1)Wheat varieties were identified using a combination of terahertz time domain spectroscopy and improved convolutional neural network(CNN).Taking different varieties of wheat as the research object,Terahertz time-domain spectra were combined with convolutional neural nets for the identification study of 12 wheat varieties.The raw time domain spectral information of 0.1～2.0 THz of wheat was collected,and the spectral information of frequency domain,refractive index and absorption coefficient from 0.2 to 1.4 THz was obtained after removing noise and irrelevant information by Fourier transform and standard normal transformation(SNV)pretreatment.The competitive adaptive reweighted sampling(CARS)and uninformative variable elimination(UVE)algorithms were used to screen the spectral data of the Absorption Coefficient,and the filtered spectra were used as input parameters for the three machine learning models for training;a 7-layer convolutional neural network was also designed to train the full spectrum information matrix.Furthermore,the Momentum optimized gradient descent Algorithm(SGDM)and the Dropout operation were introduced into the convolutional neural network to accelerate the convergence rate of the model and avoid over-fitting.The accuracy of the calibration set of the experimentally designed convolutional neural network model reached 98.7%,the accuracy of the prediction set reached 97.8%,and the error rate was only 2.2%;while the accuracy of the traditional SVM model was 89.6%and the error rate was 10.4%.The accuracy rate of BPNN model is 87.8%with a 12.2%error rate.The accuracy of LS-SVM model was 90.6%and the error rate is 9.4%.The accuracy of all three machine learning traditional models was lower than that of the experimental convolutional neural network.The accuracy of external validation of CNN model reached 99.2%,and the misjudgment rate was 0.8%.The results show that terahertz time-domain spectroscopy can effectively obtain the related information of wheat varieties,and THz spectroscopy combined with CNN can be used as an effective method to identify wheat varieties,which has important practical significance and positive application prospect.(2)A study on the identification of the origin of wheat based on terahertz time-domain spectroscopy was carried out.A total of 240 wheat samples from Shandong,Shaanxi,Henan,Hebei and Anhui provinces were selected for experiment using terahertz time-domain spectrometer.Firstly,the time domain spectrum,frequency domain spectrum and absorption coefficient spectrum data of wheat were obtained.In order to reduce the influence of noise signal,the terahertz spectrum was preprocessed by multiple scattering correction(MSC),convolution smoothing savitzky-golay(S-G),standard normal transformation(SNV)and mean-centered(MC)methods,respectively.Then,terahertz absorption coefficients were screened using the competitive adaptive weighting algorithm(CARS)and the uninformative variable elimination(UVE)algorithm.Finally,the qualitative identification models of PLS-DA,LS-SVM,BPNN and CNN were established.By comparing the classification results of different models,it was found that the SNV-CNN model had a relatively optimal classification effect with a 98.15%correct rate of comprehensive identification of wheat of different origins.The experimental results show that this method has good effect on wheat origin identification and provides a feasible identification method for origin traceability analysis of grain.(3)Wheat freshness identification based on improved sparrow optimization support vector machine(ISSA-SVM)and terahertz spectroscopy was studied.Firstly,the terahertz time-domain spectral information of 325 wheat samples from different years(2016,2017,2018,2019 and 2020)was collected,and the frequency domain information of wheat was obtained by Fourier transform(FFT)calculation.By comparison,it was found that the frequency domain spectral of wheat samples in different years were all different.The frequency domain spectrum data value of wheat in 2020 was significantly lower than that of wheat samples in other years.Then the competitive adaptive reweighted sampling(CARS)algorithm was applied to extract features from the frequency domain spectra,and based on this,back propagation neural network(BPNN),support vector machine(SVM),sparrow algorithm optimized support vector machine(SSA-SVM)and improved sparrow algorithm optimized support vector machine(ISSA-SVM)wheat freshness prediction models were constructed.Finally,the accuracy of the prediction models was compared,and the ISSA-SVM prediction model was found to be superior,with an accuracy of 96.41%in the prediction set identification.The experimental results showed that the improved sparrow optimization support vector machine combined with terahertz spectroscopy technology was feasible for the freshness detection of wheat,which provided a theoretical basis for wheat quality detection,and had practical application value.(4)A study of wheat moisture detection based on TS algorithm and terahertz imaging technology was carried out.Using wheat seeds as samples,240 wheat seeds with different moisture contents were firstly scanned with terahertz,and the terahertz spectral information within the frequency range of 0.1～1.6THz was extracted.The spectra were pre-processed with MC algorithm,S-G algorithm,SNV algorithm and MSC algorithm.Then,the features of pretreated spectra were extracted by applying the tabu search(TS)algorithm and successive projections algorithm(SPA)in combination with the imaging features.Finally,the frequency domain spectra and absorption coefficient spectra after feature extraction were combined with the partial least squares(PLS)method to build a quantitative analysis model for the calibration samples,which shows appropriate correlation between the calibration set and prediction set.For this model,the related coefficient of calibration(R_c),the related coefficient of prediction(R_p),the root mean square error of calibration(RMSEC)and the root mean square error of prediction(RMSEP)were 0.9522,0.4730,0.9531 and 0.5396,respectively;The R_pand RMSEP of external validation were 0.9462 and 0.6338,respectively.The results showed that the S-G+MSC+TS+PLS model constructed using frequency domain spectroscopy was better than the S-G+SNV+TS+PLS model constructed using absorption coefficient spectroscopy,which provided a better analysis method for the detection of water content in wheat.(5)A study of wheat impurity detection based on the combination of terahertz imaging and Wheat-V2 model was carried out.Firstly,the spectral characteristics of wheat,wheat husk,wheat straw,wheat leaf,wheat grain,weed,and insect within the range of 0.2～1.6 THz were studied using the THz spectral imaging.THz pseudo-color imaging was conducted next on wheat and its impurities according to the principle of maximum frequency-domain imaging,and a novel Wheat-V2 convolutional neutral network was designed to identify wheat impurity image information.Finally,the designed Wheat-V2 model was compared with the Res Net-V2＿50 and Res Net-V2＿101models under the same conditions.In addition,the loss function and confusion matrix indicators were used to evaluate the experimental results.By comparison,it was found that the recognition accuracy of the Wheat-V2 model was 97.56%and 98.58%for Top＿1 and Top＿5 on the validation set,respectively;In addition,the designed Wheat-V2 model achieved an average F1-score of 97.83%in terms of image recognition of various impurities,which is higher than that achieved by conventional models,i.e.Res Net-V2＿50 and Res Net-V2＿101.This indicates that the Wheat-V2 model can effectively identify the impurities in wheat images.The results show that the combination of terahertz imaging and convolutional neural network can be applied to wheat impurity detection.In addition,the research results also demonstrate the potential of convolutional neural networks in terahertz imaging detection,which can provide a nondestructive detection method for other grain impurity identification.In this paper,terahertz time-domain spectroscopy and imaging technology were used to detect the incoming quality of wheat stored in storage,which provides a new method for the detection of wheat varieties,origin,freshness,moisture and impurities.As a reliable spectral detection method,the method proposed in this paper is helpful to the quality detection of grain purchasing enterprises in the process of grain storage and transaction.It lays a theoretical foundation for the practical application of terahertz spectroscopy and imaging technology in wheat quality detection,and also provides a certain technical reference for the detection of other agricultural products quality. |
PDF Full Text Request