FEDNet: A real-time deep-learning framework for object detection
Pan Jian, Zhe Li, Wei Yi, Zhibin Chen, Yingxiong Nong, Bin Zhou, Wuqing Huang, Jun Zou, Zilang Pan, Weiwei Liu · 2023
Regular security facilities check plays a key role in preventing potential dangers in production safety. Those facilities generally include helmets, protective suits, and fire extinguishers. Amongst them, fire extinguishers are essential facilities for fire hazard prevention and fire extinguishing. However, manual checks of fire extinguishers’ presence may suffer from low efficiency and are error-prone, so we developed a real-time deep-learning framework, called FEDNet (Fire Extinguisher Detection Network), for the automatic detection of fire extinguishers on the scenes via cameras. Based on YOLOv5, which is a well-known one-stage object detection framework, we developed our FEDNet by introducing the state-of-the-art techniques including our proposed attention module, transformer-like modules and label assignment strategy. Our developed FEDNet showed much more advantageous performance in terms of detection accuracy. On a private dataset, our proposed framework achieved a [email protected] of 94.0% with an FPS (Frames Per Second) of 74, which surpassed the original YOLOv5 by a big margin. Compared to other existing methods, our method also showed overwhelming performance in terms of speed and accuracy. More importantly, the developed framework can be extended for other object detection scenarios.