Fourier-transform near-infrared spectroscopy as a tool for olive fruit classification and quantitative analysis

The potential of diffuse reflectance near-infrared spectroscopy combined with pattern recognition to discriminate between olives (Olea europaea L.) of different qualities has been tested. The sample set was formed of sound olive fruits and those showing the most common alterations of olives, which lead to decreased oil quality, namely freeze damages, harvest after falling on the ground, fermentation due to prolonged storage time, and olive tree diseases. Near-infrared (NIR) spectra were recorded between 9900 and 4100 cm(-1). Preliminary studies of the data set structure were performed using hierarchical cluster analysis and principal component analysis. Discriminant analysis provided prediction abilities of 100% for sound, 79% for frostbite, 96% for ground, and 92% for fermented olives using a leave-a-fourth-out cross-validation procedure. Quantification of oil and water content in the olives was also approached by using partial least squares (PLS) regression. Results, in terms of predictive ability using a leave-one-out cross-validation, were compared for calibration using the whole sample set and calibrations for the sound and damaged samples separately. Relative errors of prediction using all samples were 7.2% and 3.4% for oil content and humidity, respectively. Using only sound samples, relative errors of prediction of 3.8% and 2.8% for oil and water content, respectively, were obtained. Thus, better accuracy could be achieved when classification of the olive samples prior to quantitative analysis was performed. These results demonstrate the utility of NIR spectroscopy to differentiate olives of different qualities. Using NIR, a fast selection of sound olives in a quality-orientated production facility becomes feasible.