Risk identification method for power operation and inspection based on multi-element interaction relationship analysis

Introduction The interaction between power operation personnel and tools is a critical factor in ensuring maintenance safety and operational standardization. However, due to the diversity and complexity of human-object interactions in both spatial structure and action semantics, existing methods still face challenges in local features extraction and fine-grained identification of semantically similar behaviors. Methods To address these challenges, we propose an interactive risk recognition method for power operations based on semantic prompts and locally-aware enhanced Transformer, aiming to improve the analysis capabilities of key interaction behaviors such as climbing ladders or electricity checking in operational scenarios. The method first introduces a multimodal semantic prompt module, which synergistically guides visual and linguistic prompts to effectively enhance the model's semantic understanding of complex interaction behaviors. It then integrates a locally perceptual-enhanced Transformer module to strengthen the feature expression capabilities of visual and linguistic networks, thereby improving the performance of multimodal features. Results Experiments conducted on a self-constructed distribution network operation interaction dataset demonstrate that the proposed method achieves high recognition accuracy in key interaction detection tasks and can effectively identify potential safety risks, providing timely alerts. Discussion These results highlight its broad applicability in intelligent operation supervision and risk management.