4.7 Article

Weakly Supervised Acoustic Defect Detection in Concrete Structures Using Clustering-Based Augmentation

Journal

IEEE-ASME TRANSACTIONS ON MECHATRONICS
Volume 26, Issue 6, Pages 2826-2834

Publisher

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TMECH.2021.3077496

Keywords

Inspection; Mel frequency cepstral coefficient; Automation; Training data; Acoustics; Mechatronics; Task analysis; Augmentation; clustering; defect detection; infrastructure inspection; weak supervision

Funding

  1. JSPS KAKENHI [JP21K17829]
  2. Satomi scholarship foundation
  3. Suzuki foundation

Ask authors/readers for more resources

The article introduces a new approach for weakly supervised acoustic defect detection in concrete structures, which allows for significant performance gains with low amounts of weak supervision.
The automation of inspection methods for concrete structures is a pressing issue worldwide. Weakly supervised approaches, i.e., approaches based on supervision in other forms than traditional class labels, offer a unique mix of automation and human involvement that is highly effective for critical tasks such as inspection work. Generating weak supervision is less tedious than generating training data for supervised learning approaches. However, since it is less informative, high amounts of weak supervision are often needed. In practice, it is often the case that only scarce amounts of weak supervision are available. In this article, we propose a novel approach for weakly supervised acoustic defect detection in concrete structures that augment human-provided weak supervision. Experiments in both laboratory and field conditions showed that the proposed method allows for considerable performance gains for low amounts of weak supervision.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.7
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available