Abstract: In this research, hyperspectral technique combined with chemometrics was used to discriminate between Xinjiang-grown Fuji apples with and without watercore. Visible and near infrared hyperspectral images were acquired within the wavelength range of 380 to 1 004 nm. The region of interest was selected from the images to calculated average spectra. The original spectra were preprocessed by 9 different methods such as direct difference first-order derivative, and then principal component analysis (PCA), independent component analysis and correlation coefficient method were used to reduce the dimensionality of the spectral data. Finally, Bayes discriminant, K nearest neighbor method, Mahalanobis distance discriminant, least squares support vector machine, quadratic linear discriminant method were combined to perform pattern recognition. Results indicated that 15 principal components were extracted by PCA. The model developed using standard normal variate (SNV) or multiple scatter calibration (MSC) combined with PCA and least squares support vector machine (LSSVM) exhibited the best recognition performance with identification rates for the calibration and prediction sets of 100% and 91.2%, respectively.

Key words: apple watercore, hyperspectral imaging, chemometrics, principal component analysis, pattern recognition