Location: Sugarbeet and Bean ResearchTitle: Analysis of hyperspectral scattering profiles using a generalized Gaussian distribution function for prediction of apple firmness and soluble solids content Author
Submitted to: ASABE Annual International Meeting
Publication Type: Proceedings
Publication Acceptance Date: 8/8/2011
Publication Date: 8/12/2011
Citation: Zhu, Q., Huang, M., Zhao, X., Lu, R. 2011. Analysis of hyperspectral scattering profiles using a generalized Gaussian distribution function for prediction of apple firmness and soluble solids content. ASABE Annual International Meeting. Paper No. 11388. Interpretive Summary: Hyperspectral scattering has been used as a nondestructive means for assessing the firmness and soluble solids content (SSC) of apples, two important attributes for quality control and grading. The technique requires accurate description of light scattering profiles using an appropriate mathematical method. This research proposed a general Gaussian distribution (GGD) function, coupled with mean reflectance, to model the hyperspectral scattering profiles for predicting the firmness and SSC of 600 'Golden Delicious' apples for the wavelengths of 500-1000 nm. The performance of the proposed GGD-mean reflectance method was compared with two other mathematical methods, i.e., mean reflectance and the modified Lorentzian function, which were used in previous studies. Compared with the two previously developed methods, the GGD-mean reflectance method had better prediction results for the firmness and SSC of the apples, with the improvements in the standard error of up to 3% for firmness and between 4% and 21% for SSC. The generalized Gaussian distribution function is thus useful for modeling hyperspectral scattering profiles to predict the firmness and SSC of apple fruit.
Technical Abstract: Hyperspectral scattering provides an effective means for characterizing light scattering in the fruit and is thus promising for noninvasive assessment of apple firmness and soluble solids content (SSC). A critical problem encountered in application of hyperspectral scattering technology is analyzing and modeling hyperspectral scattering profiles. A generalized Gaussian distribution (GGD) function, coupled with mean reflectance (GGD-mean), was proposed to describe the spectral scattering profiles of 600 ‘Golden Delicious’ apples for the spectral region of 500–1,000 nm. The three-parameter GGD-mean model included mean, variance parameter and shape parameter from the GGD probability density function. A fast estimation algorithm was utilized for the variance and shape parameters. Calibration models for fruit firmness and SSC were developed for 400 apples, using multi-linear regression (MLR) and partial least squares (PLS), and the models were validated using the remaining 200 fruit. The GGD-mean model yielded better prediction results for fruit firmness and SSC with the average values of r obtained with PLS being equal to 0.854 and 0.864, respectively, for 10 validation runs, compared with those obtained using mean reflectance (r = 0.843 and 0.849 for firmness and SSC, respectively) and Lorentzian function (r = 0.847 and 0.802 for firmness and SSC, respectively). The GGD-mean method is thus recommended for firmness and SSC prediction because it can accurately characterize spectral scattering profiles for apples.