As seen from the principles of multivariate analysis
methods (Teye et al. 2014), they play important roles in the
analysis of electronic signals from different rice types. Good
accuracies in the calibration sets were obtained with SVMand
KNN, indicating that these models were effective and scientific.
However, the difference in accuracies of the prediction
sets by different models was relatively large. The
PCA-based models, with the combination of electronic
tongue and nose, possessed the higher accuracy than that
of electronic tongue alone, but lower than that of electronic
nose alone. This result suggested that linear features of the
combined electronic signals extracted by PCA algorithm
cannot adequately reflect the difference between conventional
and hybrid rice. PCA algorithm was not suit to the
data process of the developed multielectrode electronic
tongue combined with multisensor electronic nose. The
LLE-based models with the combination of electronic
tongue and nose achieved the highest accuracy compared
to that of either electronic tongue or electronic nose, indicating
that the LLE-based models were better than the
PCA-based models. It was shown that LLE was effective
for extracting the features of rice. This is because that the
LLE algorithm can keep nonlinear features of the combined
electronic signals to reflect the difference between conventional
and hybrid rice in the volatile compound and chemical
composition, which can predict unknown samples more
accurately. SVM based on kernel mapping tries to find the
optimal hyperplane using support vectors to reflect the
implicit nonlinear relationship, so the classification ability
is better than that of linear classifier KNN, which obtained
5 % lower prediction accuracy by the combination of
electronic tongue and nose.
As seen from the principles of multivariate analysismethods (Teye et al. 2014), they play important roles in theanalysis of electronic signals from different rice types. Goodaccuracies in the calibration sets were obtained with SVMandKNN, indicating that these models were effective and scientific.However, the difference in accuracies of the predictionsets by different models was relatively large. ThePCA-based models, with the combination of electronictongue and nose, possessed the higher accuracy than thatof electronic tongue alone, but lower than that of electronicnose alone. This result suggested that linear features of thecombined electronic signals extracted by PCA algorithmcannot adequately reflect the difference between conventionaland hybrid rice. PCA algorithm was not suit to thedata process of the developed multielectrode electronictongue combined with multisensor electronic nose. TheLLE-based models with the combination of electronictongue and nose achieved the highest accuracy comparedto that of either electronic tongue or electronic nose, indicatingthat the LLE-based models were better than thePCA-based models. It was shown that LLE was effectivefor extracting the features of rice. This is because that theLLE algorithm can keep nonlinear features of the combinedelectronic signals to reflect the difference between conventionaland hybrid rice in the volatile compound and chemicalcomposition, which can predict unknown samples moreaccurately. SVM based on kernel mapping tries to find theoptimal hyperplane using support vectors to reflect theimplicit nonlinear relationship, so the classification abilityis better than that of linear classifier KNN, which obtained5 % lower prediction accuracy by the combination ofelectronic tongue and nose.
การแปล กรุณารอสักครู่..