1998 Fiscal Year Final Research Report Summary
Reconstruction of Object Image From Millimeter Wave Image By Intelligent Process
| Project/Area Number |
08650514
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
計測・制御工学
|
|---|
| Research Institution | TOYO UNIVERSITY |
|---|
Principal Investigator |
YONEYAMA Masahide TOYO UNIVERSITY,INFORMATION & COMPUTER SCIENCES,PROFESSOR, 工学部・情報工学科, 教授 (60277358)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
MIZUNO Kouji TOUHOKU UNIVERSITY,RESEARCHINSTITUTE OF ELECTRICAL COMMUNICATION,PROFESSOR, 電気通信研究所, 教授 (30005326)
|
|---|
| Project Period (FY) |
1996 – 1998
|
| Keywords | millimeter wave / imaging system / image reconstruction / feedforward neural network / dynamic type neural network / Hopfield neural network / Boltzmann machine |
|---|
| Research Abstract |
As the research for the millimeter electro-magnetic wave imaging system, the optimal values of several physical parameters such as the object size and the antenna array size were determined.
26 alphabet capital letters (A-Z) having unevenness of the surface were made as the objects for test. The great number of the millimeter-wave images for each object were measured under several experimental conditions.
On the other hand, as the post processing using neural networks, the image reconstruction were experimented by the computer simulation. As the neural networks, both type of the feedforward type and the dynamic type were employed. Two kind of dynamic type neural networks such as Hopfield neural network and Boltzmann machine are used.
Generally, the image obtained from the millimeter electro-magnetic wave imaging system is so degraded that it is impossible for human eyes to recognize the original object's shape from the millimeter wave image.
The aim of the the post processing using neural networks is to reconstruct the original object's images being recognizable for human eyes from the degraded millimeter-wave images.
As the experimental results, it became clear that Boltzmann machine was superior to Hopfield neural network in the capability of recalling the correct images.
Although the dynamic type neural networks have a merit being easy to prepare the learning data set, the feedforward type is generally better than the dynamic type at the point of the image reconstruction capability
|
