Skip to main content
ARS Home » Midwest Area » East Lansing, Michigan » Sugarbeet and Bean Research » Research » Publications at this Location » Publication #291782

Title: Analysis of hyperspectral scattering images using a moment method for apple firmness prediction

Author
item ZHU, Q - Jiangnan University
item HUANG, MIN - Jiangnan University
item Lu, Renfu
item Mendoza, Fernando

Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/1/2013
Publication Date: 7/13/2014
Citation: Qibing, Z., Huang, M., Lu, R., Mendoza, F. 2014. Analysis of hyperspectral scattering images using a moment method for apple firmness prediction. Transactions of the ASABE. 57(1):75-83.

Interpretive Summary: Firmness is an important internal quality attribute in determining fruit maturity and harvest time and for assessing postharvest quality of apples. Hyperspectral scattering is a promising technique that was recently developed for nondestructive firmness assessment of apple and other fruits. It measures light scattering in the fruit over the spectral region of 500-1,000 nm and then extracts scattering features for prediction of fruit firmness. Since hyperspectral scattering images provide a large amount of spatial and spectral information for each fruit, an effective method for fast extraction of scattering features from the images is needed. This study proposed a new method, called moment method, to characterize scattering features for apples over the wavelengths of 500-1,000 nm. Hyperspectral scattering images were acquired, using an inhouse developed online hyperspectral imaging system, from 6,894 apples of three varieties (i.e., ‘Delicious’, ‘Golden Delicious’, ‘Jonagold’) harvested in 2009 and 2010. The firmness of apples was measured using a standard destructive method. The zeroth and first order moments were calculated from the hyperspectral scattering images. Two mathematical modeling methods, partial least squares (PLS) and least squares support vector machine (LSSVM), were used to develop firmness prediction models for the zeroth order moment, the first order moment and their combined data. Overall the zeroth moment was similar in firmness prediction to the first order moment. The models for the combined data were significantly better (ranging between 1.5%-15.2%) than the models using the zeroth order moment alone. The LSSVM models for the combined data achieved good firmness predictions with the correlation coefficients of 0.85-0.95 for the three apple varieties, which were 7.2% to 17.7% better than the PLS models in the prediction error. The proposed moment method is much faster and simpler in computation, compared to other features extraction methods developed previously, and can thus be implemented for rapid assessment of fruit firmness.

Technical Abstract: This article reports on using a moment method to extract features from the hyperspectral scattering profiles for apple fruit firmness prediction. Hyperspectral scattering images between 500 nm and 1000 nm were acquired online, using a hyperspectral scattering system, for ‘Golden Delicious’, ’Jonagold’, and ‘Delicious’ apples harvested in 2009 and 2010. The zeroth order moment (ZOM), which is equivalent to the mean reflectance, and the first order moment (FOM) were calculated from the hyperspectral scattering profiles for each wavelength. Firmness prediction models were developed for the ZOM data, FOM data and their combined data (or Z-FOM) respectively, using partial least squares (PLS) and least squares support vector machine (LSSVM). The PLS models based on the Z-FOM data improved prediction results by 1.5%-12.5% for the prediction set, compared with the PLS models using the ZOM data alone. The LSSVM models for the prediction set of Z-FOM data yielded better prediction results, with the improvements of 8.6%-21.2% over the PLS models for the ZOM data, 7.2%-17.7% over the PLS models for the Z-FOM data and 2.9%-15.2% over the LSSVM models for the ZOM data. The Z-FOM method provided a simpler, faster and effective means to extract features from the hyperspectral scattering profiles, and it has led to significant improvements in firmness prediction accuracy, when used with either PLS or LSSVM.