Comparative study of distance discriminant analysis and Bp neural network for identification of rapeseed cultivars using visible/near infrared spectra

Abstract

The potential of visible/near infrared spectra as a method of nondestructive discrimination of various rapeseed cultivars was evaluated, discrimination ability of distance discriminant analysis (DDA) and BP neural network (BPNN) for identification of rapeseed cultivars was shown in this article. The spectral curves ranging from 350 to 2500 nm of rapeseed cultivars were obtained by VIS/NIR spectroscopy, and the principal component analysis (PCA) was applied to perform the clustering analysis. The first 6 principle components (PCs) extracted by PCA were employed as the inputs of DDA and BPNN, respectively, and then two different discrimination models for rapeseed cultivars were built. Forty-five samples from each species and a total of 225 samples from 5 categories rapeseed were extracted. One hundred fifty samples were elected randomly as training sets to set up the training model which was validated by the samples of prediction sets formed by the remaining 75 samples. The result error of BPNN model was set to be ±0.15, and the result indicated that no samples exceeded threshold value, therefore the distinguishing rate was 100%. The result of DDA model displayed that the recognition rate of 100% was achieved. Although the methods mentioned in the presented paper were good approaches for nondestructive discrimination of rapeseed cultivars, DDA model with prediction functions was more intuitive than BPNN and convenient to machine recognition. © 2011 IFIP International Federation for Information Processing.