Review of Scientific Instruments, Год журнала: 2025, Номер 96(4)

Опубликована: Апрель 1, 2025

Surface cracks are crucial for structural health monitoring of various types buildings. Despite substantial advancements in crack detection through deep neural networks, their reliance on pixel-level annotation escalates labeling costs and renders the procedure time-intensive. Consequently, academics have suggested multiple Explainable Artificial Intelligence (XAI) methodologies to enhance efficacy pseudo-labeling. However, fractures’ slender, continuous, inconspicuous characteristics render current XAI approaches ineffective adequately gathering feature information. This work examines many strategies extensive experimentation. It synthesizes advantages each strategy mitigate uncertainty error associated with a singular model fracture region. Moreover, we formulate implement integration discrepancies across distinct algorithms two separate datasets. The experimental results indicate that proposed method provides more accurate basic annotations weakly supervised segmentation.

Язык: Английский