ConstruÃÃo de um nariz artificial usando redes neurais

AUTOR(ES)
DATA DE PUBLICAÇÃO

2000

RESUMO

The animals interact with environment in several ways. The smell around them has an extremely importance in the information transport, essential for survival. Such as for the animals, the smell plays an important rule for human beings. Due to this importance, the study in the human olfactory system is very helpful and started very long time ago. Although it is important, up to now there is many questions without response.Researchers are working in this area trying to get responses for these questions, and artificial noses are being developed to help to understand and to mimic the human olfactory system. The commercial artificial nose has the frequently cited applications: the food quality control, beverage industry, environmental monitoring, medical diagnoses, perfumes, robotics and safety. Hence, there are high demand for electronics instruments for mimic the human olfactory sense, with low cost and sensorial information fast and precise. An artificial nose is a device composed of a chemical sensitivity system and of a pattern recognition system. The chemical sensor system has a set of different sensors for vapours detection. There are different technologies to prepare the sensors, resulting in different sensitivity and work condition. In this work we applied neural network for solving the pattern recognition system of an artificial nose completely designed and constructed by us (team of computer science, chemistry and physics). The use of neural network is essential as a technique for pattern recognition due to flexibility. Especially in our case some particular neural network architectures has shown good results for recognising smell signals. Our work involved four prototypes of artificial nose during four years, with a gradual improvement in chemical sensor technology, data acquisition instrumentation and pattern recognition. The sensor technology changes from: manually prepared electrodes with a gap of polypyrrole electrochemically deposited as an active material; to micrometrical four leads electrodes and the same active material of the first prototype; and finally the technology of the thin organic films prepared by the free growth technique. The data acquisition system followed this changing improving the signal/noise ratio and the operational performance. Marizete Silva Santos ConstruÃÃo de um nariz artificial usando redes neurais The neural network system also had several versions to follow this development and to adapt to new situations involving others approaches. In the beginning we accomplished tests with several neural network models to treat smell data from the first prototype and analysed the data to find a good model for recognition of this data. The data from the second prototype were more precise and noiseless then the first one. We use many methods to evaluate the performance of the neural network models with this data and we found the good results with MultiLayerPerceptron model with an optimised architecture. The last stage of this thesis was a development of new approaches for neural network especially designed for treating smell data. We suggested artificial noses based on biological models of human olfactory system. In these models the neural network architecture is a fundamental element of the new concept of artificial noses. Despite others artificial nose models that focalize the dynamic and physiology of signals in internal organs that control and recognised the signal already pre-processed from the more external elements near the sensor, we suggest an approach based on anatomy of organs nearest the sensorial neurons (the primary human sensors). This aspect of modelling is not treated in others works, but it is essential to understand the coding of the message sent to brain from the molecules that interact with the primary sensors. The fundamental point of our model is the coding of signal through the connections between mitrals cells, glomerulis and sensorial neurons. This stage corresponds to a pre-processing of signal, selecting and concentrating the essence of the information of a human chemical sensor, the olfactory system

ASSUNTO(S)

odor sensors ciencia da computacao rede neurais neural networks artificial nose sensores de aroma e nariz artificial

ACESSO AO ARTIGO

Documentos Relacionados