Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 76
PROCEEDINGS OF THE THIRD INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY
Edited by: B.H.V. Topping and Z. Bittnar
Paper 84

Application of Self-Organizing Maps to Classification of Bridge Images

H. Shiro, M. Hirokane and H. Furuta

Faculty of Informatics, Kansai University, Osaka, Japan

Full Bibliographic Reference for this paper
H. Shiro, M. Hirokane, H. Furuta, "Application of Self-Organizing Maps to Classification of Bridge Images", in B.H.V. Topping, Z. Bittnar, (Editors), "Proceedings of the Third International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 84, 2002. doi:10.4203/ccp.76.84
Keywords: self-organizing map, image classification, image characteristics,.

Summary
Recently, the importance of aesthetics design for the public structures has been recognized. In such a situation, the design that reflected the sensitivity of citizens is required. However, when the citizens participate in the design for the public structures, their opinion is influenced with the advice of some engineers with enough experience and it is difficult to suggest voluntarily their own opinions for the design. Although almost all the public structures are still designed by the engineers with enough experience, in recent years, some structures show a tendency to be designed with the requests of citizens. However, there are some differences between the requests of citizens and the constructed structures and some citizens recognize the requests of them are not reflected to the design.

In this study, for decreasing such differences of design between the requests of citizens and the constructed structures, we attempted to apply the SOM [1,2,3] to support the design that reflected the requests of citizens or engineers. As the first step for this purpose, the photo images of bridges were classified into some similar groups by using the SOM. In this implementation, 90 photo images of bridges in the bridge yearbook [4] that was published in 1993 were actually used for classifying into some similar groups. In the recent report for the aesthetics design [5], it was reported that the colour of bridge had a considerable influence for the sensitivity of citizens. Therefore, the RGB values that were included in the photo images were used in this research as the characteristics for the photo images. Moreover, four types of templates [6] were used for taking the spatial distribution of colors into consideration. All of 90 photo images were divided into 4 parts by using each template such as the vertical line, the horizontal line, the diagonal line and the concentric circle. Each of the RGB values was divided into three ranges. The classification of photo images was implemented by using these values as the characteristics for each photo image.

The self-organizing maps (SOM) that was proposed by T. Kohonen is a kind of neural network technique and this method should be considered as the algorithm based on unsupervised competitive learning. This method provides the visualized map that is mapped from a high dimensional space to a lower dimensional one that is called as the map units. Map units, or neurons, are usually considered as a two- dimensional grid and such a SOM algorithm is actually mapping a high-dimensional space onto a two-dimensional plane. The similar input data vectors on the neurons that are nearby each other among the input vectors are placed nearby on the unit of the output map. As such a result, the similarity among the input data that is located in the high-dimensional space can be realized visually on the two-dimensional map.

The main findings of the present study are as follows:

As compared with the classification results based on four types of templates, the photo images that the background, the position of bridge girders and so on were very similar, were rearranged at the near position and classified into the same group in all classification results. For example, No5 and No29 as shown in Figures 19 and 20 of the full paper were classified into the same group in all classification results. These two images are photographed from the far spots and the colours of bridges are clear blue and go straight from the right end to the left end, the background include the sky and the river, and the colours of these background is also blue. However, the photo images that the background, the position of the bridge girders and so on are different, were classified into the different group by each template. For example, No17 and No49 as shown in Figures 14 and 15 of the full paper that were classified into the same group by the diagonal line based template were classified into the different group by the other templates. From the viewpoint of the spatial distribution that is included the colours, the photographing spots and so on, the classification results by the concentric circle based template were better than the others. It can be thought that in case of photographing an object, its object is placed around in the center area.

References
1
H. Tokutaka, S. Kishida, K. Fujimura, "Self-Organizing Map", Springer Verlag Tokyo, 1996.
2
H. Tokutaka, S. Kishida, K. Fujimura, "Application of Self-Organizing Map", Kaibundo Publications, Japan, 1999.
3
T. Kamisaka, E. Izumi, I. Yamakawa, "Self-Organizing Method with Conversion of Input Vectors and Its Application to Figure Detection", 15th Fuzzy System Symposium, Japan, 1999.
4
"Bridge Yearbook", Association of Japan Bridge Construction, Japan, 1993.
5
W. Shiaki, N. Ito, K. Yasuda, M. Adachi, "Reseach on Evaluation for Aesthetics Design of Bridges Using Method in Kansei Engineering", Symposium for Optimization of System, 79-84, Japan, 1997.
6
"Advanced Use Technology of Information", The Institute of Systems, Control and Information Engineers, Japan, 2000.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £85 +P&P)