Abstract: The intelligent identification of mining accident hazards represents a critical technical problem in the intelligent
transformation of mine safety. Traditional hazard identification methods face dual challenges of limited applicability and insufficient
semantic reasoning capabilities,which fail to meet the requirements of intelligent management in this field. This research
analyzes the limitations of independently applying object detection and knowledge graphs in mine safety,proposing an
integrated framework that synergizes their technical advantages. An information fusion mechanism is designed to structurally
map visual perception results to semantic nodes in the knowledge graph,establishing a closed-loop cognitive chain of "mine imagery-
knowledge graph-hazard reasoning" for intelligent hazard identification. The methodology begins by establishing a threetier
classification system for mining entities based on industry standards,constructing a mining entity knowledge graph,and formalizing
safety rules into Cypher-based inference rules to support hazard reasoning. Subsequently,the YOLOv8n model is improved
and the SE attention mechanism is introduced to improve the recognition accuracy of mining entities,and combined with
the mining accident inference rule base to realize the recognition inference of mining accident hazards. A B/ S architecture prototype
system is implemented to visualize identification and inference results. Experimental results demonstrate that the enhanced
object detection achieves 71% mean average precision (mAP) across 14 critical entity categories in complex mining
environments. The hazard identification and reasoning module within the prototype system can automatically infer whether the
current scenario contains safety hazards,thereby validating the framework′s feasibility in identifying and reasoning about mining
hazards in complex environments. This research provides novel insights for intelligent hazard discrimination in mine safety management.

Key words: mine safety,object detection,knowledge graph,intelligent mine,accident hazard inference