ABSTRACT
Hetao muskmelon is a unique fruit in the hetao area of northwest china, which has been loved by consumers. Sugar content is the important indicator of measuring the quality and mature of muskmelons. This research uses Maya 2000 pro portable spectrometer and PR-101α portable digital refractometer to get spectrum and sugar content values of "jinhongbao" muskmelon, researches the effect of different extraction methods of characteristic wavelength (stepwise multiple linear regression (SMLR), interval partial least squares(iPLS), backward interval partial least squares (biPLS) and synergy interval partial least squares (siPLS)) on model accuracy and prediction results. The results show: using biPLS method on extraction of characteristic wavelength will the full spectrum evenly divided into 20 subintervals, the PLS factors of 14, when removing 8 subintervals, and choosing the wavelength variable numbers of 218, getting the biPLS model is best, RMSE of corresponding calibration and prediction models is 0.9961 and 1.18. So using the biPLS method of extraction on spectrum wavelength could extract effectively the characteristic wavelengths of melon sugar content, increase the ability of model prediction, and achieve rapid detecting of sugar content about muskmelons.
