混凝土构件表面裂缝分割的改进SegFormer方法

打开文本图片集

中图分类号：TU375 文献标志码：A DOI： 10.15935/j.cnki.jggcs.202504.0005

AbstractConcrete structures commonly exhibit visible surface cracks and detecting these surface cracks is of great significance for structural condition assessment.This paper presents an improved attention mechanism mode for surface crack segmentation.The method leverages the network framework and innovative module （CA）of the improved attention mechanism mode，adds a coordinate attention module to the decoder，and by introducing pooling along the diagonal direction significantly enhances the model's understanding of elongated spatial relationships in pixel-level predictions.Comparative experiments on the public dataset Concrete3k demonstrate that the improved attention mechanism mode achieves the highest IoU value，indicating superior performance in crack segmentation tasks.Additionally，the improved atention mechanism mode maintains high performance with a relatively low parameter count.

Keywordsconcrete components， surface cracks，image segmentation，improved atention mechanism mode，deep learning，robustness

0 引言

在过去二十年里，我国建筑行业飞速发展，许多基础设施和建筑物相继拔地而起。（剩余12352字）