Neuro - Fuzzy and Multivariate Statistical Classi cation of Fruit Popula - tions Based on Visible - Near Infrared Spectrophotometry

Variations in fruit development can aaect fruit composition, maturity, storage attributes and sensory properties. While these have major importance to the horticultural industry, there is a lack of suitable tools for discriminating between fruit. Visible-near infrared (VNIR) reeectance spectroscopy collects a large volume of data rapidly and non-destructively at any stage of development. The postprocessing of this data to yield such qualitative information has not been studied extensively in horticultural systems. Spectroscopic data was processed using both multivariate statistical (principal component and canonical discriminant analysis) and by Neuro-Fuzzy hybrid approaches, using kiwifruit treated during the period of growth to resemble natural extremes in fruit populations. The classiication performance of the Neuro-Fuzzy hybrid approach was in general superior to that obtained by multivariate statistical methods.