Full text
| |
| Author(s): |
Giovani Baratto
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | São Paulo. |
| Institution: | Universidade de São Paulo (USP). Escola Politécnica (EP/BC) |
| Defense date: | 1998-03-11 |
| Examining board members: |
Francisco Javier Ramírez Fernandez;
Euvaldo Ferreira Cabral Junior;
André Carlos Ponce de Leon Ferreira de Carvalho;
Jose Carlos Teixeira de Barros Moraes;
Homero Enrique Banados Perez
|
| Advisor: | Francisco Javier Ramírez Fernandez |
| Abstract | |
A classifier was developed using artificial neural nets for the recognition of patterns of aromas in an electronic nose. The classifier was developed starting from the biological model KIII of Freeman. The classifier is composed of 4 neural nets, calls of R), R1, R2, R3 and R4, interlinked for delays elements denominated of L1, L2, L3 and L4. The net R0 reconstructs the sign of entrance of the sensor ones, the net R1 classifies the state of the sensor ones and the nets R2 and R3 generate focalization signals for the net R1. An alectronic nose was built and measures were accomplished in two series. In the first series the ethyl alcohol, the isopropyl alcohol and the acetone were used. In the second series of measures four commercial Brazilian brandy were used, I blink and alcohol of rice. Simulations were accomplished with experimental and synthetic patterns with a multilayer perceptron. The smallest mistake rate was reached with the use of the tension on the sensor ones in permanent regime before and after the injection of the species as attribute. A training algorithm was developed for the artificial neural nets of the classifier. (AU) | |