An Improved HRNetV2-Based Semantic Segmentation Algorithm for Pipe Corrosion Detection in Smart City Drainage...

Area of Science:

• Civil Engineering
• Computer Science
• Artificial Intelligence

Background:

• Urban drainage pipelines are critical for smart city infrastructure.
• Internal pipeline corrosion presents significant structural and safety risks.
• Accurate detection of corrosion is vital for maintaining underground systems.

Purpose of the Study:

• To develop an enhanced semantic segmentation framework for accurate identification of corroded pipeline regions.
• To improve the precision and robustness of corrosion detection in closed-circuit television (CCTV) images.
• To support intelligent infrastructure inspection and automated maintenance systems.

Main Methods:

• Utilized High-Resolution Network Version 2 (HRNetV2) for semantic segmentation.
• Integrated Convolutional Block Attention Module (CBAM) to enhance feature representation of corrosion patterns.
• Incorporated Lightweight Pyramid Pooling Module (LitePPM) for improved multi-scale context modeling.

Main Results:

• Achieved a mean Intersection over Union (mIoU) of 95.92 ± 0.03%.
• Reached a Recall of 97.01 ± 0.02%.
• Obtained an overall Accuracy of 98.54% on a real-world corrosion dataset.

Conclusions:

• The proposed HRNetV2-based framework effectively detects pipeline corrosion.
• The method demonstrates high precision and robustness in identifying corroded areas.
• Results provide technical insights for advancing automated maintenance in smart cities.