Improving of prediction model for apple sugar content using independent component analysis

To improve the prediction model of sugar content, independent component analysis (ICA) was used to preprocess the near infrared (NIR) spectra of apples. Compared with those original spectra, apple spectra after ICA pretreatment were smoother, but their shape showed not much difference. This indicated that the major information in apple spectra could be reserved while noise was removed by ICA method. The partial least square (PLS) was used to establish the calibration models of sugar content against apple spectra after ICA and averaging pretreatment. Compared with just being averaged, the results show that the number of factors used in PLS model against the spectra pretreated by ICA decreased, and the precision was also improved. The optimum PLS calibration model was obtained with 6 factors, the correlation coefficient (r(c)) of 0.9549 with the standard error of calibration ( SEC) of 0.3361 and the standard error of prediction (SEP) of 0.4355. This result shows that the ICA preprocessing NIR spectra can not only improve precision, but also simplify the model.