Comparison of Pattern Classification Methods in System for Crossarm Reuse Judgement on the Basis of Rust Images

M. Yamana, H. Murata, T. Onoda, T. Ohashi, and S. Kato (Japan)


Machine Learning, Pattern Classification, SVM, Crossarm,Rust Images


Japanese electric power companies aim to utilize existing equipment fully and maintain facilities effectively. Hu man experts presently judge various hardware regarding whether they are reusable or not, in order to utilize equip ment fully. In particular, this paper considers crossarm reuse judgement. This judgement by human experts is performed on the basis of the rust which adheres to the crossarms. However, this judgement depends on human expertise, so it is difficult to maintain a constant degree of accuracy of judgement. Electric power companies want to ensure constant and accurate judgement. Therefore, we try to develop a crossarm reuse judgement system based on rust images that uses machine learning techniques. The system consists of a commercial microscope and a standard note PC to keep the cost for the reuse judgement. We es timate the degree of accuracy of the judgement of various pattern classification methods without special image pro cessing techniques such as the extraction of features. The results show that a support vector machine is the most suit able instrument for this judgement system.

Important Links:

Go Back