Research On The Quality Of Fresh Smoke Based On Hyperspectral Technology

Tobacco,as one of the most important economic crops in China,the quality of flue-cured tobacco leaves is the lifeblood of the entire tobacco industry,directly affecting the quality of cigarettes products,farmers'income and social benefits.Fresh tobacco leaves are the basis of tobacco leaves after curing,and their quality is directly related to the quality of tobacco leaves after curing.However,the main factors affecting the quality of fresh tobacco include the maturity,moisture content and pigment content of the fresh tobacco leaves.At present,the maturity,moisture content,and pigment content of fresh tobacco leaves mostly rely on traditional manual identification and detection,lacking objective and quantitative evaluation criteria and the experiment takes a long time and damages the samples.Therefore,it is of great significance to propose a fast and non-destructive method for detecting the maturity,moisture content and pigment content of fresh tobacco leaves.This paper took the fresh tobacco leaves in the middle of Yunyan 87 in Tianzhu County,Xifeng County,Anlong County,Daozhen County and Weining County of Guizhou Province as the research object.First,using the self-built hyperspectral image information acquisition system to shoot hyperspectral images of fresh tobacco under natural light conditions and determine its maturity.Then the moisture content and pigment content of fresh tobacco leaves were determined by lab experiments.Finally,the relationship between the hyperspectral characteristics of fresh tobacco leaves and their maturity,moisture content and pigment content was analyzed.The maturity discriminant model and prediction models of moisture content and pigment content based on hyperspectral technology were established respectively.The main research contents and results are as follows:(1)A model for judging the maturity of fresh tobacco leaves based on hyperspectral image technology was established.According to the maturity of fresh tobacco leaves,their physical and chemical properties and spectral characteristics are also different.The 6spectral indices(SI_S)and 4 visible light spectral parameters(VL_S),combined with the partial least square discriminant analysis(PLS-DA)were used to establish fresh tobacco leaf maturity discriminant models(SI_S-PLS-DA,VL_S-PLS-DA).Through comparative analysis,it was concluded that the effect of VL_S-PLS-DA is better,the discriminant accuracy of the modeling set has reached 99.11%,and the prediction set has reached98.96%.(2)A prediction model for the moisture content of fresh tobacco leaves based on hyperspectral image technology was established.First,the hyperspectral data of fresh tobacco leaves were preprocessed by 10 preprocessing methods,including multivariate scattering correction,normalization,Savitzky-Golay smoothing algorithm and first derivative.Then,in order to eliminate abnormal samples,principal component analyses combined with Mahalanobis distance method(PCA-MD)were used.Finally,the partial least square algorithm was introduced to establish a fresh tobacco leaf moisture content model.The results showed that the Savitzky-Golay convolution smoothing method was useful for preprocessing,and the fresh tobacco moisture content prediction model(SG-PCA-DA-PLS)was reliable.The model eliminated 3 abnormal samples,the corrected Set_CR and validation setR_CV were 0.8569 and 0.8527,and the RMSEC and RMSECV were 1.2115 and 1.3766.The accuracy of the model is good,and the moisture content of fresh tobacco leaves can be predicted.(3)A prediction model for the pigment content of fresh tobacco leaves based on hyperspectral image technology was established.The hyperspectral data was pre-processed by the standard normal variable method(SNV)and the sensitive feature was extracted by the successive projections algorithm(SPA).The BP neural network model was used to predict the pigment content based on the above features.The results showed that the SNV-SPA-BP prediction model of chlorophyll a content had the highest accuracy(_CR²=0.882,RMSEC=0.190,R²_CV=0.863,RMSECV=0.203).The fresh tobacco chlorophyll b content prediction model(SG2-SPA-BP)was established with the highest accuracy(_CR²=0.892,RMSEC=0.073,R²_CV=0.871,RMSECV=0.100)after the hyperspectral data was preprocessed by the Savitzky-Golay method and second derivative method and the characteristic sensitive band was selected by the continuous projection transformation method.The prediction model of carotenoid content in fresh tobacco leaf(SG-SPA-BP)was established with the highest accuracy(_CR²=0.845,RMSEC=0.042,R²_CV=0.833,RMSECV=0.207)after the data was preprocessed by Savitzky-Golay smoothing algorithm and the characteristic sensitive band was selected by continuous projection transformation method.