Kiwifruit flesh firmness determination by a NIR sensitive device and image multivariate data analyses

A prototype based on a NIR sensitive camera and a Xenon lamp was set up and used to capture 8 bit gray scale (from 0 = black to 255 = white) image of the radiation that passes through the fruit. The count of the pixels with different gray tone was used to build statistical-mathematical models to correlate and predict the kiwifruit flesh firmness. One hundred sixteen fruits conveniently stored to obtain firmness within a range of penetrometric force from 0.8 N to 87 N, were submitted to the optical measurements. Simple regression between the gray tone having the maximum number of pixels and the firmness showed an exponential correlation with R-2 values of 0.717. On the contrary, the tone uniformity (maximum number of pixels with the same gray tone) resulted linearly correlated with hardness (R-2 = 0.687). PLS algorithm allowed prediction of the flesh firmness with R-2 of 0.777 (RMSE = 13 N). Artificial neural networks produced similar results. Although the current technique does not fully satisfies the need of an accurate selection, it could be considered for on-line applications by improving performances (e.g. acting on lamp spectral emissions and camera detection) and with easy mechanical modifications of the sorting lines. (C) 2019 Elsevier B.V. All rights reserved.