- 著者
-
本庄 鉄弥
土屋 範芳
中塚 勝人
- 出版者
- 一般社団法人 資源・素材学会
- 雑誌
- 資源と素材 (ISSN:09161740)
- 巻号頁・発行日
- vol.111, no.4, pp.205-211, 1995-04-25 (Released:2011-01-27)
- 参考文献数
- 37
Self-Organizing Neural Network (SONN) was constructed for the purpose of mineral identification. This system consists of two different kinds of networks, Kohonen's Self-Organizing Map and three layer feedforward neural network based on the back-propagation learning algorithm. The former step, Self-Organizing Map, could divide minerals into some categories by the similarities on the selected characteristics of minerals. This rough division of whole input patterns on feature maps was closely analogous to the first step of classification by human brains. The later step, each category had the three layer feedforward neural network independently, and then the minerals belonging to the same category could be identified.In this study, 82 minerals were identified by 5 characteristics of cleavage, metallic luster, Mohs hardness, streak, and color. Some minerals have plural input patterns on the 5 characteristics mentioned above. Therefore, total number of input patterns was 119 for 82 minerals.After constructing the feature maps and the back-propagation learning, this system could suggest the suitable mineral name for unlearning input patterns. The advantage of the proposed method is that scaling up of the system is possible with relatively small increase in learning times. Further, it should be stressed that this technique can be used in other problems where recognition and identification are necessary.