Abstract:In order to develop a nondestructive method for identifying expanded kiwifruits, near-infrared diffused spectra of 120 expanded kiwifruits and 120 normal kiwifruits were obtained between 833 and 2500nmusing a Fourier transformation near-infrared diffused spectrograph. Standard normal variate transformation was used to preprocess original spectra. The samples were divided into calibrationset and prediction set based on Kennard-Stone method. Eleven principal components and 6 characteristic wavelengths were selected by principal component analysis (PCA) and successive projections algorithm (SPA). Partial least squares (PLS), support vector machine (SVM), and error back propagation (BP) neural network identification model were established based on full spectrum (FS), PCA, and SPA, respectively. The results showed that correct identification rates of all models were higher than 96.7% and 93.3% for calibration set and predication set, respectively. The established models based on PCA and SPA were much simpler than those based on FS, since the variable numbers of them were only about 0.53% and 0.29% of that of FS, respectively. The identification performance of PLS and SVM were better than that of BP. The best model was PCA-PLS, whose accuracy rate reached 100% for calibration set and predication set. The results clearly indicate that near-infrared diffused spectra technique has the potential as an efficient, accuracy and non-invasive method for distinguishing expanded kiwifruits from normal kiwifruits.
