Author(s)

Kun Zhang

Corresponding Author

Kun Zhang

ABSTRACT

To measure Brix in Shatang mandarin, FieldSpec 3 porTable spectroradiometer made Inc. (ASD) was used to measure 200 Shatang mandarin’s absorbance. The PLS model of Shatang mandarin was built. 8 different popular spectroscopy pretreatment approaches were used to find the suiTable optimal pretreatment methods and the optimal modeling band of the model of Shatang mandarin. Then the BP neural network model of Shatang mandarin was built. The results show the l correlation coefficient R of BP neural network model of Shatang mandarin based on the PLS mode calibration model of Shatang mandarin was 0.967. On validation the correlation coefficient R was 0.868. The mean absolute error was 0.56 Brix. The standard error of the prediction was 0.74.

KEYWORDS

Shatang mandarin, Brix value, Near infrared spectrum, Partial least squares, BP artificial neural network