直線振動運動する模擬動物を用いた遺伝的アルゴリズムによる図形の特徴抽出

Feature extraction from simple patterns by utilizing a genetic algorithm of artificial animal (AAA-2) having one-dimensional oscillatory motion

Creators Kawauchi Yoshihiro

Creators Miike Hidetoshi

Creators Koga Kazutoshi

Creators Yokoyama Etsuro

This paper pesents a new approach for featuer extraction of a simple pattern utilizing a genetic algorithm. In the algorithm, we assume artificial animals (AA-2) having one-dimensional oscillatory motion. The animals are moving aroundin an artificial world where the foods of the animals are distribured in a shape of a spatial pattern. Each animal has motion parameters of direction of motion (DM) and amplitude of oscillation (AO) as their own genes. When they meet together in a certain condition new generations are produced by mutation and intersection of genes. Darwin's rules of selection and adaptation bring an evolution of the artificial animals. After adaptation in, the genes of survived animals represent features of the spatial pattern. Features such as directions and characteristic size of the pattern can be extracted by analyzing the type of survived genes.

[ISSN]0372-7661

[NCID]AN00244228