Related Experiment Videos
ECS-Net: An adaptive feature learning semantic segmentation network for accurate concrete porosity prediction
Tiantian Liang1, Dunxin Gu1, Zerun Guo1
1School of Electrical Engineering, Dalian Jiaotong University, Dalian 116028, China.
Iscience
|August 2, 2026
Summary
This study introduces an efficient concrete segmentation network (ECS-Net) for improved porosity prediction in concrete engineering. ECS-Net enhances the detection of internal concrete damage and structural integrity.
Area of Science:
- Materials Science
- Civil Engineering
- Computer Vision
Background:
- Accurate concrete porosity prediction is crucial for structural safety and quality assessment.
- Computed tomography (CT) image segmentation for concrete pores faces challenges like complex structures, small pore sizes, low contrast, and intricate morphologies.
- Existing methods struggle with reliable segmentation of internal concrete features.
Purpose of the Study:
- To develop an efficient concrete segmentation network (ECS-Net) for enhanced pore feature extraction from complex CT images.
- To improve the accuracy and reliability of porosity prediction in concrete structures.
- To support non-destructive detection of internal concrete damage and structural integrity diagnosis.
Main Methods:
- Development of an encoder-decoder semantic segmentation network (ECS-Net) with specialized modules for adaptive pore feature extraction.
- Implementation of a multiscale encoder to integrate global structural information and local pore details.
- Design of a complexity-aware decoder to enhance segmentation at fuzzy boundaries and low-contrast regions.
Main Results:
- ECS-Net demonstrated superior segmentation performance and generalization ability on concrete CT, Crack500, and aggregate datasets compared to mainstream models.
- Experiments confirmed stable performance in porosity prediction through 3D reconstruction using limited specimens.
- The network effectively handles challenges posed by evolving pore structures, tiny pores, low contrast, and complex morphologies.
Conclusions:
- ECS-Net offers a robust solution for accurate concrete CT image segmentation and porosity prediction.
- The developed network can significantly aid in the non-destructive detection of internal concrete damage.
- Findings support the application of ECS-Net for structural integrity diagnosis and long-term performance evaluation of concrete structures.