p

g

m Fourier transform of the spectra of this data set.

he estimated baseline for one of 45 replicates of the milk data using BWH. The

and for the raw spectra and the thick line stands for the estimated baseline.

the transformation, a matrix was generated. The data was further

m transformed. A visualisation map was produced using the self-

g map algorithm [Kohonen, 1982] based on the logarithm-

med short-term Fourier transform data. To do so is because the

nising map algorithm can normally produce a map to well-reserve

alise the structure of a multi-dimensional data. The visualisation

ed out through a process in which the 45 replicates were mapped

nt neurons, which were arranged as a two-dimensional array of

e cell is one neuron on this array. If two spectra profile were

o the same neuron (cell), it is believed that these two spectra profile

great degree of similarity compared with others. In addition, similar

ere supposed to be mapped to neighbouring neurons (cells). The R

kohonen was used for this data visualisation.

elf-organising map for this data set was composed of 36 neurons

h each of 45 spectra data was mapped with a cow milk

ation percentage printed on the cells of the map. Some neurons

no spectrum, some attracted one spectrum and some attracted

n one spectrum. If two spectra were mapped to the same neuron,

pected that their spectra profiles were very similar. It is expected

cow milk concentration percentage values were similar. Figure

w such a map. The self-organising map shows a good separation