Related Experiment Video
Updated: Jul 6, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Boundary-Sensitive Hybrid Attention Network for Multi-Scale Crack Fine Segmentation.
Yaotong Jiang1, Tianmiao Wang1, Congyu Shao1
1School of Mechanical Engineering and Automation, Beihang University, Beijing 100191, China.
Sensors (Basel, Switzerland)
|May 27, 2026
Summary
This study introduces BSA-Net, a novel network for concrete crack segmentation in bridge health monitoring. It improves accuracy in challenging conditions, offering a reliable solution for automated infrastructure inspection.
Area of Science:
- Civil Engineering
- Computer Vision
- Artificial Intelligence
Background:
- Concrete crack segmentation is vital for infrastructure safety and longevity.
- Traditional methods struggle with weak contrast, background noise, and multi-scale cracks.
Purpose of the Study:
- To develop an advanced deep learning model for accurate concrete crack segmentation.
- To address limitations of existing methods in complex and noisy environments.
Main Methods:
- Introduced the Boundary-Sensitive Hybrid Attention Network (BSA-Net).
- Employed a hierarchical Transformer encoder (Hiera-A) for multi-scale feature extraction.
- Utilized a multi-scale context module (Light-ASPP) for efficient context aggregation.
- Implemented a dual-branch boundary-aware decoder (BAD) for precise boundary detection.
Main Results:
- BSA-Net demonstrated superior performance over existing crack detection models.
- Achieved high accuracy in segmentation, boundary clarity, and recall rates, especially for subtle cracks.
- Showcased effectiveness in complex, noisy environments and on benchmark datasets.
Conclusions:
- BSA-Net offers a scalable and reliable solution for automated infrastructure monitoring.
- The model enhances crack segmentation performance in real-world conditions.
- Provides improved defect detection capabilities for civil infrastructure.
