Identifying Indoor Natural Gas Work Targets with an Enhanced YOLOv8 Model Incorporating RMT Backbone

This study aims at the real-time recognition of indoor natural gas work targets utilizing an enhanced YOLOv8 model integrated with an Retentive Networks Meet Vision Transformers (RMT) backbone. The model is specifically designed to detect and recognize key work instruments and objects, including gas stoves, water heaters, flue pipes, valves, and more, from video footage captured by wearable devices equipped with cameras. To further improve the detection accuracy, we introduce the MPD-Inner-IoU loss function to better handle the positioning of bounding boxes. The enhanced model is trained and validated using a comprehensive dataset gathered from real-world indoor gas operation scenarios. Experimental results demonstrate the model’s capability for accurate and real-time identification of indoor natural gas work targets, highlighting its potential for improving safety and efficiency in gas operation environments.