Enhanced Faster R-CNN with Temporal and Context Modules for Power Plant Safety Monitoring

Abstract

In order to improve the intelligence and automation level of power plant safety monitoring systems, this study proposes an improved Faster R-CNN algorithm by integrating multi-scale feature extraction, context-aware RPN, temporal information fusion, and RoI Align optimization. The model is trained and tested on a power plant safety monitoring dataset covering diverse and complex scenarios. Comparative experiments against baseline methods including original Faster R-CNN, YOLOv5, and SSD demonstrate that the improved algorithm achieves a mean Average Precision (mAP) of 0.85 and a Recall of 0.82, outperforming the baselines by margins of up to 13% in mAP and 12% in Recall. The enhanced algorithm also shows superior adaptability to small targets, occlusions, low light, and complex backgrounds. These results indicate that the proposed method significantly enhances the performance of target detection in challenging power plant environments.