| Summary: | The pre-processing using seven normalization models was carried out to
improve the degree of pattern discrimination of electronic nose (e-nose) outputs
which are initially similar each other. All models are adopted from the well
known models that are usually used in data processing. They are: range scale1 ,
range scale2, relative scale1, relative scale2, baseline subtraction, global method,
and local method. In this case, high discrimination degree is important in
improving the performance of pattern recognition system of e-nose. To provide a
set of trained samples, six variants of mix ratio between turmuric and wild ginger
were exposed to e-nose. Before being trained to the pattern recognition system,
all the patterns were normalized. The pattern recognition system used in this
research is back propagation artificial neural network (BP-ANN).
Effect of the seven models on the performance of pattern recognition system
is marked by the percentage of recognition repeatibility in recognizing the patterns
of test samples. In this case, the test samples were treated using the same
procedures similar to that of the trained samples. As the result, the normalization
effect may improves the pattern discrimination visually. However, from the seven
normalization models, range scale1 and range scale2 are the best models shown by
the high (90%) pattern recognition degree and repeatibility.
|
|---|