Rapid Non-destructive Detection Of Internal Defects In Carya Cathayensis Sarg By Near Infrared Spectroscopy

Carya cathayensis Sarg,as a special agricultural product with high value,produce different degrees of internal damage in some segments,and internal damage cannot be distinguished by the naked eye.In this paper,we investigated the effects of different lighting conditions on the detection of internal defects,and used different pretreatment,characteristic wavelengths,and modeling methods to classify and identify internal defects(oil seeds,empty pods,and black spots)of pecans.The main research and conclusions of this paper are as follows:(1)A near-infrared spectral detection device was built for pecans to detect internal defects in pecans,and the effect of different lighting conditions on the detection of internal defects was investigated,and the results showed that the differences in the spectral curves at the characteristic peaks were greater in lighting condition 1 than in lighting condition 2.(In addition,the specificity,sensitivity and correctness of the LDA model for oilseed,empty bud and black spot were better than those for light condition 1 than those for light condition 2.(2)Normal,oilseed,hollow bush and black spot pecans were used as the subjects for the study,and different spectral pretreatment,feature wavelength and modeling methods were used for classification model building.The results showed that most of the oilseed and hollow bush samples were clustered in PCA,but the black spot samples were scattered everywhere;the quadratic discriminant function LDA was suitable for oilseed defect identification,the linear discriminant function LDA was suitable for hollow bush defect identification,and the Marxian distance discriminant function LDA was suitable for black spot defect identification.First derivative(FD),standard normal variate transform(SNV)and multiplicative scatter correction(MSC)are the best pre-processing methods for the classification of oilseed,hollow bush and black spot defects,respectively.For the three internal defects,PCA could not distinguish between the four,the quadratic discriminant function LDA was suitable for the identification of the three internal defects,and MSC was the best preprocessing method.Competitive adaptive reweighted sampling(CARS)is the best feature wavelength screening method,which can greatly reduce the amount of data and computation time while ensuring the accuracy of the model.(3)To investigate the effects of primary and secondary data fusion in data fusion techniques on model accuracy.The results showed that the data fusion technique was significantly effective in improving the accuracy of Carya cathayensis Sarg internal defects.the quadratic discriminant function LDA of SNV combined with primary data fusion was the optimal model for oilseed defects with sensitivities,specificities and correctnesses of 0.96,0.99 and 0.99;the linear discriminant function LDA of MSC combined with intermediate data fusion was the optimal model for hollow pod defects with sensitivities,specificities and correctnesses The sensitivity,specificity and correctness were 0.87,0.99 and 0.97;the FD combined with the intermediate data fusion Marxian distance discriminant function LDA was the optimal model for the vacuole defect,with sensitivity,specificity and correctness of 0.84,0.96 and 0.93.For the three internal defects of pecan,the best classification model was MSC combined with the intermediate data fusion quadratic discriminant function LDA,and the correctness of the model calibration set and prediction set were 0.94 and 0.85,respectively.