Using Eq. (5) to assess the fruit mass from optically detected
length did not render satisfactory results. The fruit mass was
overestimated by 29% and the calculated values showed a high variability
(R2 = 0.56). It is assumed that this is the combined effect of L
being the factor that shows the least correlation toM, and the somewhat
higher variability of the photographically determined values
for L, as compared to the manually measured values (Fig. 4B).
As previously shown, image processing enables the reliable
reproduction of a 2D raster model of the measured fruit (Yimyam
et al., 2005), based on RGB color analyses, which can be used to
create a 3D model if several pictures from different angles are
combined (Chalidabhongse et al., 2006). Even though data on computational
time has not been published, the amount of data which
needs to be processed to create and evaluate a 3D model may be
rather high, and thus somewhat unwieldy if used for an automated
process such as sorting. In this study it has been shown that, alternatively,
mass can be estimated based on the data obtained from
two orthogonally taken photographs.