Abstract:
A non-destructive image analysis method for classification of tomato (Lycopersicon
esculentum) varieties (Sri Lankan hybrids) has been developed. Maheshi, Rajitha , T245
and Thilina varieties grown in Atipola, Sri Lanka during Yala season, 2009 were used in
the study. Multispectral imaging system with white LEDs as light source was used for data
acquisition. Two plane polarizers have been employed in order to acquire multiple
scattered light coming from the interior of the sample. The acquired data were corrected
for intensity variations occurred due to the 3-dimensional nature of objects by transferring
them to 2-dimensional images. The 2- dimensional data were subsequently transformed
into a 1-dimentional array in order for efficient computer processing. The acquired data
were compressed using singular value decomposition and trained them with a linear model
with an offset. After successful training of data, the system was found to be capable of
discriminating Maheshi, Rajitha and T245 / Thilina varieties. However, the system was
unable to discriminate T245 and Thilina varieties individually. Discrimination capability
of the system was tend to deteriorate with the shelf life
