Research On Morphological Neural Network Endowed With Dendrites

Morphological neural network endowed with dendrites ( MNNED) is a new paradigm for neural computing. Its morphological neuron model incorporates the dendritic processes and yields a more realistic neuron model mimicking the biological model. Due to the advantages, MNNED has been applied widely to the fields of intelligent information processing, such as image processing and pattern classification. The thesis does the further research into MNNED. in which the following contents are involved and the corresponding results are obtained:(1) Image algebra is analyzed, and the relationship between image algebra and neural network model is established.(2) The morphological neural network that incorporates the dendritic processes is investigated, and the geometric properties of the dendritic processes are identified through the analysis of the learning algorithms for single layer morphological perceptron endowed with dendrites.(3) In accordance with the position characteristics of patterns, an approach to selecting the parameters, which can make morphological associative memory endowed with dendrites (MAMED) more robust, is presented.(4) By replacing the sphere neighborhoods with hyper-boxes generated by dendrites in the covering algorithm, the covering algorithm is modified. Based on this modification, a learning algorithm for multi-layer morphological perceptron endowed with dendrites (MMPED) is presented, which can not only adjust the parameters of the network, but determine its structure.(5) The programs for implementing MAMED and MMPED are developed. Thus these neural networks are applied to solve image recovering and pattern...