Analysis of hyperspectral scattering characteristics for predicting the bruising susceptibility of apple

Abstract.

Hyperspectral scattering is a promise technique for assessing the apple quality, and extraction of the most useful information from the spectral scattering profiles is critical for accurate assessment of apple bruising susceptibility. This paper analyzed three methods including a diffusion theory model, a four-parameter Lorentzian distribution function and a generalized Gaussian distribution for characterization of the spectral scattering profiles acquired from300 ‘Golden Delicious’ apples by a hyperspectral imaging system over the wavelength region of 500-1000 nm. Partial least squares (PLS) algorithm was used to develop the calibration models for predicting the bruising susceptibility. The correlation coefficient of prediction () was 0.796, 0.794 and 0.784, and the standard error of prediction () was75.6, 75.1 and 76.7 (mm³J^-1) for diffusion theory, Lorentzian distribution and generalized Gaussian distribution models, respectively. Results demonstrated that each scattering characteristic could be used to predict bruise susceptibilities of sound ‘Golden Delicious’ apples, and provide auxiliary information for fruit quality sorting and packaging.