列車前方画像を用いた木まくらぎ劣化度判定AIシステムの開発

抄録

In railway track maintenance, the inspection of wooden sleepers is crucial for the safety of train operations. However, the inspection of wooden sleepers is basically performed through manual visual inspection by workers, which is time-consuming and labor-intensive. Therefore, we developed a wooden sleeper inspection system that uses images of the front of a train obtained by a camcorder. The proposed system uses deep learning to judge the deterioration of the images and ranks the degree of deterioration. This paper outlines the developed system, evaluates its judgment accuracy, and compares the results with judgments made by multiple train maintenance engineers. Accordingly, the accuracy of the proposed method in accurately judging the degree of deterioration reached 86.3% to 94.1% when compared with the experts' results, indicating that the developed deterioration judgment model has comparable performance to that of the train maintenance engineers in terms of judgment accuracy and distribution of answers.