Research On Rapid Detection Method Of Wheat Freshness Based On Hyperspectral Imaging Technology

Wheat is an important strategic reserve grain in China,and the detection of its freshness is particularly important.In order to solve the problems of complex operation,strong destructiveness and inaccurate results in traditional detection methods,a set of rapid non-destructive detection method of wheat freshness was established by using hyperspectral imaging technology,to provide data reference for on-line detection of wheat freshness.Artificially accelerated aging(40±1?,90%±5%RH)of the newly harvested strong,medium and weak gluten wheat in2020,to study the changes of catalase activity,fatty acid value and gluten water absorption during aging.The quantitative prediction models of three indexes were established by hyperspectral imaging technology combined with chemometrics method,the freshness of wheat with different storage years was further classified,and the visualization of chemical information distribution was realized.The main results are as follows:(1)Changes of wheat catalase activity,fatty acid value and gluten water absorption during aging.In the artificial accelerated aging process,the catalase activity of strong,medium and weak gluten wheat decreased with the extension of aging time;the fatty acid value increased with the extension of aging time;the gluten water absorption fluctuated without obvious trend;the catalase activity and fatty acid value of wheat were highly sensitive,which could be used as indicators to characterize the freshness of wheat;the sensitivity of gluten water absorption was not high,so it is not suitable to be used as an indicator to characterize the freshness of wheat.(2)Study on hyperspectral quantitative prediction method of catalase activity in wheat.BP neural network(BPNN)model and support vector regression(SVR)model were established by combining the full band spectral data of original and pretreated by first derivative(1ST),multiplicative scatter correction(MSC)and standard normal variable transformation(SNV).The predictive effect of 1ST-SVR model was the best,R²=0.9689.The successive projections algorithm(SPA)was used to extract the characteristic bands of wheat catalase activity and modeled.After MSC preprocessing,11 characteristic bands were extracted,which acco unted for 4.30%of the total 256 bands.The corresponding wavelengths were 918.7 nm,956.2 nm,986.8 nm,1061.4 nm,1125.4 nm,1149.0nm,1192.6 nm,1368.9 nm,1382.1 nm,1653.3 nm and 1672.7 nm.The predictive effect of MSC-SPA-SVR model was the best,R²=0.9664.Therefore,MSC-SPA-SVR model can be used to quantitatively predict wheat catalase activity.(3)Study on hyperspectral quantitative prediction method of fatty acid value in wheat.BPNN and SVR model were established by combining t he full band spectral data of original and pretreated by 1ST,MSC and SNV.The predictive effect of 1ST-SVR model was the best,R²=0.9440.The SPA was used to extract the characteristic bands of wheat fatty acid value and modeled.The predictive effect of SNV-SPA-BPNN model was the best,R²=0.9207,21characteristic bands were extracted,accounting for only 8.20%of the total bands.The corresponding wavelengths were 857.2 nm,997.0 nm,1030.9 nm,1118.7 nm,1145.6 nm,1192.6 nm,1325.8 nm,1368.9 nm,1395.3 nm,1415.1 nm,1424.9 nm,1477.5 nm,1617.7 nm,1621.0 nm,1630.7 nm,1653.3 nm,1669.5 nm,1672.7 nm,1676.0 nm,1698.6 nm and 1701.8 nm.Therefore,SNV-SPA-BPNN model can be used to quantitatively predict th e fatty acid value of wheat.(4)Study on hyperspectral quantitative prediction method of wheat gluten water absorption.BPNN and SVR model were established by combining the full band spectral data of original and pretreated by 1ST,MSC and SNV.The prediction results were general.Among them,the predictive effect of 1ST-SVR model was the best,R²=0.6065.The SPA was used to extract the characteristic bands of wheat gluten water absorption and modeled,and the prediction eff ect of the model is general,among which,the predictive effect of SNV-SPA-BPNN model was the best,10 characteristic bands were extracted,which only account for 3.91%of the total band,R²=0.6425,which is better than the model established by the full band data.The quantitative prediction of wheat gluten water absorption by hyperspectral technique is not ideal,so it is not recommended to continue the following experiments.(5)Study on fast classification of wheat freshness by hyperspectral.The wheat stored for 0-4 years were classified based on the average spectral data,and the accuracy of BPNN classification and support vector classification(SVC)can reach up to 100%.7 characteristic wavelengths were extracted by SPA,the corresponding wavelengths are 942.6 nm,1105.2 nm,1192.6 nm,1322.5 nm,1643.6 nm,1672.7 nm and 1692.1 nm.The accuracy of classification is still100%,and the classification effect of BPNN is better.Classified based on CAT,the accuracy of BPNN classification was 100%.CAT?71.42 mg H₂O₂/g,wheat freshness was I grade;71.42 mg H₂O₂/g>CAT?37.91 mg H₂O₂/g,wheat freshness was II grade;CAT<37.91 mg H₂O₂/g,wheat freshness was III grade.Classified based on fatty acid value,the accuracy of BPNN was 97.96%.fatty acid value?15.13 mg KOH/100g,wheat freshness was I grade;15.13 mg KOH/100g<fatty acid value?20.66 mg KOH/100g,wheat freshness was II grade;20.66mg KOH/100g<fatty acid value?35.16 mg KOH/100g,wheat freshness was III grade;fatty acid value>35.16 mg KOH/100g,wheat was IV grade.The freshness of wheat was visualized based on CAT and fatty acid value,and the effect was obvious.In this paper,hyperspectral imaging technology was used to establish a rapid non-destructive detection method of wheat freshness.The classification of wheat freshness was based on CAT and fatty acid value,which breaks the rough estimation barrier of traditional chemical methods.While obtaining the grade of wheat freshness,the specific chemical values of CAT and fatty acid value can be obtained,which makes the prediction of wheat freshness more accurate,fast and non-destructive.