V²net: A Nested Framework for Crack Segmentation Based on Multiscale Feature Learning

Crack detection is crucial for infrastructure safety assessment and maintenance. However, the diversity of crack patterns, complex topologies, and variable imaging conditions pose significant challenges to automatic segmentation systems. In addition, issues with multiscale feature extraction and fusion, along with insufficient consideration of external factors like weather and human interference in existing datasets, leave room for improvement in model accuracy and generalization. In response to these challenges, we propose a V2net, a multiscale nested network designed to enhance generalization by progressively connecting multiple VNet submodules. We introduce a simple yet efficient multiscale feature extraction and fusion module [multichannel fusion attention (MCFA)] to improve feature extraction. To address the limitations of existing datasets, we developed and released the SUES-CRACK dataset, which incorporates complex scenarios, including variations in lighting conditions, blurred crack boundaries caused by water infiltration, and human interference. Both quantitative and qualitative results demonstrate the effectiveness of our proposed model for crack segmentation. Extensive experiments show that V2net achieves MIoU of 66.51% and 44.23% on two public crack segmentation datasets. To further validate its generalization, the model achieved a MIoU of 50.91% on the SUES-CRACK dataset. The dataset and source code will be made available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/nanxiang11/NxV2net</uri> following the publication of the paper.