Measurement Science and Technology, Journal Year: 2024, Volume and Issue: 36(2), P. 026007 - 026007

Published: Dec. 23, 2024

Abstract Tunnel cracks pose a significant threat to structural integrity, potentially leading localized collapse of the infrastructure. Traditional manual crack detection methods are prohibitively expensive, highlighting need for an efficient and accurate automatic segmentation model. To address this challenge, we propose novel model subway tunnel lining surface based on DeepLabV3+ architecture. In model, design improved Swin transformer V2 Base (SwinV2*) as backbone enhance performance. Considering tubular morphology cracks, introduce snake convolution module better capture their unique features. prevent performance degradation when fusing shallow deep features, incorporate spatial feature calibration that facilitates alignment grouping along channel dimension. We assess our model’s effectiveness using thousands images captured by image acquisition system designed surfaces. Experimental results show achieves strong metrics: 68.96% IoU, 84.33% mIoU, 87.57% PA. Compared original DeepLabV3+, approach demonstrates superior performance, with 2.89% improvement in 1.45% increase mIoU notably, 10.39%

Language: Английский