Due to olfactory pathway is relatively simple and well knows functionally and morphologically, it is an interesting system for classification tasks. Many different aspects of olfaction, such as the nature of stimuli or the mechanisms of reception and central processing have been widely studied and modeled on last decades. This dissertation research a bionic neural network based on main features of the olfactory system and its applications to different pattern recognition tasks.
Firstly, this thesis introduces the main issues on olfactory neural system and odour researches during the last years.
Secondly, after describing the anatomic structure of olfactory neural system some olfactory model like K-set of Prof. Freeman, the bulb model of Prof. Li and cortical model of Prof. Liljenstrom are exposed.
Thirdly, a bionic model mimics the olfactory system and its application on pattern recognition processes is researched. Based on bulb and cortical areas of the olfactory system the model mimics the main features of the olfactory system. One of the main characteristic of our model is that patterns come into the network by the bulb model and using afferent connection the patterns are learned and stored for future recall on the cortical model. In order to improve the classification task a modified Hebbian learning rule is applied to bulb and cortical model. Furthermore optimization processes are applied to improve the model performance.
Four different standard datasets are used to test the pattern recognition capacity of our bionic model. The performance of our bionic model is also compared with some classical ANNs and with former results obtained by other researchers using same datasets. Finally the classification results are summarized for each dataset, showing the unquestioned capacity of our olfactory bionic model to learn and classify patterns using a very small training set and a few learning trials.