Title: Edge-Aware and Content-Adaptive Infrared Gas Leak Detection for Industrial Safety Monitoring

Abstract:

While infrared gas leak detection is critical for ensuring industrial safety and protecting the environment, automated identification remains difficult due to the elusive nature of gas plumes, which are frequently faint, diminutive, semi-transparent, and poorly defined. To address these challenges in cluttered thermal environments, this study introduces the Edge-Aware and Content-Adaptive Feature Fusion Detector (ECAF-Det), a specialized solution for detecting weak plumes.

ECAF-Det incorporates three purpose-built architectural components. First, a plume-oriented local-global feature enhancement block is employed to retain subtle boundary details while capturing long-range contextual continuity. Second, a multi-scale edge perception module converts directional gradient and phase-consistency signals into hierarchical edge priors, facilitating a boundary-sensitive representation of the plumes. Third, a content-adaptive sparse routing path aggregation network dynamically manages multi-scale feature propagation, effectively highlighting informative plume characteristics while minimizing redundant background noise.

Performance evaluations on the IIG dataset demonstrate that ECAF-Det attains an Average Precision (AP) of 29.8%, an AP at 50% IoU of 84.3%, and a small-object AP of 25.3%. These metrics represent improvements of 3.0, 6.5, and 5.4 percentage points, respectively, over the RT-DETR-R18 baseline, all while maintaining a computational cost of 43.7 GFLOPs and 14.9 million parameters. Furthermore, tests on the LangGas dataset yielded an AP of 36.3% and an AP50 of 68.5%, confirming the model’s ability to generalize across varying infrared gas plume appearances.

The primary AI innovation lies in the integration of edge-aware representation learning with content-adaptive sparse feature routing, designed specifically for perceiving weak infrared plumes. Ultimately, this detector offers a robust visual perception tool for early warning systems and remote inspection tasks within industrial gas leak monitoring frameworks.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC