Multimodal Pipe Defect Diagnosis with Transformer-Based Cross-Modal Learning

View at publisher Summarise with VIRAT
Authors: Ping Zhong, Gang Li, Weijiao Kong, Yue Yu
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The integrity of underground pipelines is critical for urban infrastructure and environmental safety. Traditional inspection methods, such as Magnetic Flux Leakage (MFL) and optical imaging, often suffer from inherent limitations, failing to provide a comprehensive diagnosis. MFL excels at detecting volumetric defects like corrosion and pitting but is insensitive to non-metallic or planar defects such as cracks and deformations. Conversely, optical imaging identifies surface cracks and structural anomalies but is highly susceptible to turbid water and limited lighting conditions. This paper proposes a novel approach for multimodal defect diagnosis by synergistically fusing MFL and optical imaging data using a Transformer-based cross-modal learning framework. Our method leverages the self-attention mechanism of the Transformer to effectively model the intricate dependencies within each modality and, more importantly, the cross-attention mechanism to learn a shared, robust feature representation that captures complementary information from both data streams.

Fields of Study & Subjects

Non-Destructive Testing Techniques Infrastructure Maintenance and Monitoring Industrial Vision Systems and Defect Detection

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
152–157
Publication date
28 January 2026
Language
EN