Multimodal Information Fusion-Based Methods for Pipeline Inspection and Defect Quantification

Abstract

This paper focuses on the detection and quantification of pipeline defects in magnetic flux leakage (MFL) inspection under practical in-line inspection conditions. In real pipeline environments, MFL signals are affected by structural components, operating-condition fluctuations, and environmental noise. These factors cause overlapping target distributions and increase the difficulty of weak-defect identification and reliable defect quantification. To address defect detection, this paper proposes a dual-modal information fusion framework in which multichannel MFL signals are transformed into time–intensity maps and grayscale maps to jointly characterize temporal evolution and spatial channel distributions. A spatial–frequency collaborative enhancement mechanism is introduced to suppress interference from strong structural responses, and a distance-aware cross-modal fusion strategy is designed to improve information interaction and detection robustness. For defect quantification, this paper proposes a physics-informed multi-axis probabilistic decision framework to address geometric coupling and uncertainty accumulation in the inversion process. Physical and engineering priors derived from MFL mechanisms are incorporated to guide depth estimation, and complementary triaxial information is used to improve prediction stability. Probabilistic uncertainty modeling is further introduced to characterize prediction confidence and reduce decision risk in ambiguous cases. Experiments conducted in laboratory, simulation, and field pipeline environments demonstrate that the proposed framework improves detection stability, regression accuracy, and uncertainty representation, validating its effectiveness for practical pipeline inspection and quantitative defect assessment.