Visual detection of workers’ personal protective equipment for underground construction safety management

Underground construction plays an important role in the development of geological energy resources and underground spaces, including tunnels, mines, caverns, utility corridors, and underground energy storage facilities. In these confined and visually degraded environments, the correct use of personal protective equipment is essential for worker safety. However, existing visual inspection methods usually detect workers and protective equipment as independent objects, which may lead to incorrect worker-equipment association under low illumination, dust, occlusion, motion blur, and crowded working conditions. This paper proposes TCSS-Net, a Topology-Causal Safety State Network for worker-level personal protective equipment detection in underground construction safety management. The proposed method constructs a person-centric human topology and anchors equipment evidence to anatomically corresponding body regions, including the head, face, torso, hands, waist, and feet. A topology-aware safety evidence construction module extracts body-part-level visual, geometric, and visibility features. A causality-invariant topological reasoning module then propagates safety information across anatomical regions and introduces counterfactual constraints to improve robustness against underground environmental bias while preserving sensitivity to true protective-equipment evidence. Finally, an uncertainty-aware risk decision module integrates unsafe probability, visibility reliability, visual degradation, and model uncertainty to generate worker-level safety decisions with interpretable body-part risk explanations. The proposed framework addresses key challenges in underground safety monitoring, including person-equipment mismatch, anatomical inconsistency, environmental shortcut learning, and uncertain visual evidence. It provides a practical and extensible vision-based solution for intelligent safety supervision in underground construction and geological energy resource engineering.