Detection and recognition of veterinary drug residues in beef using hyperspectral discrete wavelet transform and deep learning

A fast,non-destructive recognition method for veterinary drug residues in beef was proposed to mitigate the laborious sample preparation and long detection times associated with conventional chemical detection techniques.Control beef samples free of veterinary drug residues and four groups of beef sprayed with relevant concentrations of metronidazole,ofloxacin,salbutamol,and dexamethasone under ambient conditions were analyzed by 400-1000 nm hyperspectral imaging followed by multiplicative scatter correction preprocessing.Data dimension reduction was performed using Competitive Adaptive Reweighted Sampling(CARS),Principal Component Analysis(PCA),and Discrete Wavelet Transform(DWT)based on Haar,db3,bior1.5,sym5,and rbio1.3 wavelet basis functions.Treated data were subjected to Convolutional Neural Network(CNN),Multilayer Perceptron(MLP),Random Forest(RF),and Support Vector Machine(SVM)modelling.CNN,MLP,SVM,and RF algorithms achieved overall accuracies of 91.6%,88.6%,87.6%,and 86.2%,respectively,when combined with DWT(wavelet basis functions and numbers of transform layers being Haar-4,db3-2,bior1.5-4,and sym5-3,respectively).The algorithm Kappa coefficients(0.89,0.86,0.85,and 0.83,respectively)and time consumption for prediction(140.60 ms,57.85 ms,70.67 ms,and 87.16 ms,respectively)were also superior to models based on CARS and PCA.DWT combined with deep learning can shorten prediction times,considerably improve the accuracy of classification and recognition,and alleviate the Hughes phenomenon,thus providing a new method for the fast,non-destructive detection and recognition of veterinary drug residues in beef.