DEEP LEARNING-BASED CRACK DETECTION AND STRUCTURAL DEFECT MAPPING IN REINFORCED CONCRETE SYSTEMS
Fuente:
Zenodo
Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Recurso digital |
| Veröffentlicht: |
Zenodo
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866902233436651520 |
|---|---|
| author | Dr. SriSudha Garugu, Dr. Tadi Chandrasekhar, Vikas Gulati, Dr. Aasheesh Raizada, Anup Saxena, Dr. Vimal Bibhu, Ranjan Banerjee |
| author_facet | Dr. SriSudha Garugu, Dr. Tadi Chandrasekhar, Vikas Gulati, Dr. Aasheesh Raizada, Anup Saxena, Dr. Vimal Bibhu, Ranjan Banerjee |
| contents | <p class="MsoBodyText">The structural support of the built environment is made of reinforced concrete systems, but their overall performance is jeopardized by the progressive surface deterioration that is challenging to evaluate at any given time with the help of manual inspection. This paper will suggest a deep learning-based architecture of crack detection and mapping structural defects in reinforced concrete infrastructure with an emphasis on pixel-level semantic<span> </span>segmentation<span> </span>and<span> </span>spatial<span> </span>explanations.<span> </span>Multi-class<span> </span>annotation<span> </span>of<span> </span>high-resolution<span> </span>concrete<span> </span>images was done by a U-Net based convolutional neural network trained under class-imbalanced conditions. This method blends the processing of the data, optimization of weighted losses, quantitative results of Dice and Intersection-over-Union metrics, and mapping defect areas that can be interpreted as representations of the surface conditions. Experimental findings prove that the framework is effective to segment visually dominant defects like corrosion, spalling, and exposed reinforcement whereas fine cracks can be difficult to segment because<span> </span>of<span> </span>their<span> </span>sparse<span> </span>geometry.<span> </span>The<span> </span>systematic<span> </span>patterns of<span> </span>over-<span> </span>and<span> </span>under-estimation are<span> </span>seen in the analysis of defects areas, which demonstrates that predictions should be interpreted as relative values, but not<span> </span>as absolute ones. The research places semantic segmentation as a powerful decision support method. It aids architectural inspection, documentation of conditions, prioritization of maintenance, and development of<span> </span>trusted digital processes of reinforced concrete infrastructure system in the world.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20379125 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | DEEP LEARNING-BASED CRACK DETECTION AND STRUCTURAL DEFECT MAPPING IN REINFORCED CONCRETE SYSTEMS Dr. SriSudha Garugu, Dr. Tadi Chandrasekhar, Vikas Gulati, Dr. Aasheesh Raizada, Anup Saxena, Dr. Vimal Bibhu, Ranjan Banerjee Reinforced concrete inspection, deep learning, semantic segmentation, structural defect mapping, crack detection. <p class="MsoBodyText">The structural support of the built environment is made of reinforced concrete systems, but their overall performance is jeopardized by the progressive surface deterioration that is challenging to evaluate at any given time with the help of manual inspection. This paper will suggest a deep learning-based architecture of crack detection and mapping structural defects in reinforced concrete infrastructure with an emphasis on pixel-level semantic<span> </span>segmentation<span> </span>and<span> </span>spatial<span> </span>explanations.<span> </span>Multi-class<span> </span>annotation<span> </span>of<span> </span>high-resolution<span> </span>concrete<span> </span>images was done by a U-Net based convolutional neural network trained under class-imbalanced conditions. This method blends the processing of the data, optimization of weighted losses, quantitative results of Dice and Intersection-over-Union metrics, and mapping defect areas that can be interpreted as representations of the surface conditions. Experimental findings prove that the framework is effective to segment visually dominant defects like corrosion, spalling, and exposed reinforcement whereas fine cracks can be difficult to segment because<span> </span>of<span> </span>their<span> </span>sparse<span> </span>geometry.<span> </span>The<span> </span>systematic<span> </span>patterns of<span> </span>over-<span> </span>and<span> </span>under-estimation are<span> </span>seen in the analysis of defects areas, which demonstrates that predictions should be interpreted as relative values, but not<span> </span>as absolute ones. The research places semantic segmentation as a powerful decision support method. It aids architectural inspection, documentation of conditions, prioritization of maintenance, and development of<span> </span>trusted digital processes of reinforced concrete infrastructure system in the world.</p> |
| title | DEEP LEARNING-BASED CRACK DETECTION AND STRUCTURAL DEFECT MAPPING IN REINFORCED CONCRETE SYSTEMS |
| topic | Reinforced concrete inspection, deep learning, semantic segmentation, structural defect mapping, crack detection. |
| url | https://doi.org/10.5281/zenodo.20379125 |