列車前方画像を用いた木まくらぎ検査手法の精度向上

抄録

In order to save labor work in the inspection of track facility by a low-cost and simple inspection method, we have developed a wooden sleeper inspection method analysing forward view image from train cab. The developed method uses deep learning to diagnose the degree of deterioration of wooden sleepers in five condition stages.However, with regard to the worst degree of deterioration, there is a problem that the diagnosis accuracy is lowered because the learning data obtained is extremely small. Therefore, the diagnosis accuracy was improved by expanding the training data. This paper reports the verification results of the accuracy improvement.