打音検査における変状検出のための時間周波数パターン抽出

抄録

Hammering test is adequate for the auto-inspection of social infrastructures because of its high accuracy and easiness of operation. Recently, a lot of machine learning approaches to construct defect detectors of hammering test are studied. However, difficulty in obtaining training dataset of hammering sound decreases the performance of the detectors due to overfitting to the training dataset. In this paper, against the problem, a hammering sound feature that is liftered in the quefrency domain is validated.