Development of a common calibration model for determining the Brix value of intact apple, pear and persimmon fruits by near infrared spectroscopy

Calibrations developed using near infrared (NIR) spectroscopy to determine the quality of fruits or vegetables are usually applicable to a single species. The ability to determine the quality of several species using a common calibration would have advantages in some production and processing situations. If consumer NIR devices for fruit inspection in a supermarket become a reality in the near future, the need for common calibrations will become critical to widespread adoption because consumers will expect one calibration to correctly determine the quality of many fruit species. A method to develop a common NIR calibration model that could be applied to many fruit species was examined. Before the spectral acquisition, the samples temperature was controlled at 25C for at least 30 min. NIR spectra of apples (Malus domestica Borkh.), pears (Pyrus pyrifolia Nakai.) and persimmons (Diospyros kaki Thunb.) were measured in the short-wavelength region using an interactance method by a NIR instrument Model 6500. After the NIR measurement, each fruit was immediately cut about 7 - 10 mm deep from the peel at the portion which was illuminated by NIR radiation for the Brix value measurement. This study compared the performance of a calibration model for each fruit species and a common calibration model which were using spectral data of each single species and three fruit species combination for calibration and validation sets, respectively. The calibration models developed were applied to the three fruit species. PLS regressions based on second derivative spectra with Savitsky-Golay (10 points of each left and right side) with wavelength region of 850 - 1000 nm by trial and error, were performed for Brix value determination using calibration samples comprising each fruit species independently and the three species combined. The regression coefficient plots revealed the weight at each wavelength of the calibration models. Strong negative peaks for apple, pear, persimmon and common calibration models were observed around 906, 910, 912 and 910 nm, respectively. These were attributed to absorption related to sugar which has been reported at 918 nm. Each single species calibration model predicted the Brix value in validation samples of the same species with a low SEP 0.34 - 0.40 Brix and with low bias (0.01 - 0.08 Brix) but with much higher SEP and bias errors when used to predict the Brix value in other species (SEP = 0.46 - 0.54 Brix and Bias = -0.72 - 1.29 Brix). Whereas, the common calibration model developed from the combined sample set predicted Brix values in the apples, pears and persimmons with an SEP = 0.43 Brix and a bias of -0.03 Brix, wherewith 5 factors, R2 = 0.88 and SEC = -0.46Brix. This study indicates that it is possible to develop a calibration model which can reliably determine Brix value in several fruit species with similar flesh structure including thin peel such as apples, pears and persimmons. To make a universal or robust common calibration model, further study is needed using many samples having different seasons, locations, varieties, maturity and so on.