Related Experiment Video
Updated: Jun 23, 2025
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.0K
Detection of Road Crack Images Based on Multistage Feature Fusion and a Texture Awareness Method
Maozu Guo1,2, Wenbo Tian1,2, Yang Li1,2
1School of Electrical and Information Engineering, Beijing University of Civil Engineering and Architecture, Beijing 102616, China.
Sensors (Basel, Switzerland)
|June 19, 2024
Summary
This study introduces FetNet, a deep learning method for road crack detection using Swin transformers. FetNet achieves high accuracy in road crack segmentation, improving transportation infrastructure inspection.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Civil Engineering
Background:
- Road structural health monitoring is crucial for transportation infrastructure.
- Accurate road crack detection and segmentation are essential for effective inspection.
- Existing methods may struggle with complex road conditions and feature extraction.
Purpose of the Study:
- To propose an automatic pixel-level semantic road crack image segmentation method.
- To enhance the accuracy and generalizability of road crack detection using deep learning.
- To introduce the FetNet model, leveraging Swin transformers for improved feature extraction.
Main Methods:
- Employed Swin transformer (Swin-T) as a backbone for multi-level feature extraction from road crack images.
- Utilized a texture unit to capture crack texture and edge characteristics.
- Integrated Refinement Attention Module (RAM) and Panoramic Feature Module (PFM) for feature merging and result refinement.
Main Results:
- FetNet achieved state-of-the-art performance on the Crack500 dataset.
- Key metrics include 90.4% precision, 85.3% recall, 87.9% F1 score, and 78.6% mean intersection over union.
- Demonstrated superior crack segmentation accuracy and generalizability compared to other advanced models.
Conclusions:
- The FetNet method significantly advances automatic road crack segmentation.
- The model shows excellent performance in complex road scenes, validating its practical applicability.
- FetNet offers a robust solution for structural health monitoring of road infrastructure.
