Coal mine external fire detection method based on edge intelligence

Abstract

The detection of external fire in coal mines and the reliable identification of initial fire are of great significance for improving the level of coal mine fire detection. It is also an important direction of intelligent mine construction in the future. In order to improve the speed, precision and real-time of coal mine external fire detection, a coal mine external fire detection method based on edge intelligence is proposed. The feature scale of the backbone network of the YOLOv5s model is improved. The model can fully learn the shallow features and improve the small target detection performance. At the same time, an adaptive attention module is added on the basis of the original feature pyramid network (FPN) to improve the detection precision of the model. There are problems of image detection error and missed detection caused by poor light conditions, more dust and camera shooting angle in the underground mine. In order to solve the above problems, the YOLOv5s-as model is constructed by using multi-sensor auxiliary detection and weighting fusion identification of video detection information and multi-sensor detection information through dynamic weighting algorithm. The YOLOv5s-as model is transplanted to the intelligent edge processor, and lightweight processing is carried out to realize the deployment of edge intelligent devices. The experimental results show that the reasoning time of the YOLOv5s-as model is slightly longer than that of the YOLOv5s-a model without sensor information fusion reasoning, but mean value of average precision when the intersection over union is 0.5 ([email protected]) is increased by 7.24%. Compared with the YOLOv5s model before transplantation, the [email protected] of the YOLOv5s-as model transplanted to the intelligent edge processor and subjected to lightweight processing increased by 15.04%. For small target fire sources, SSD 300, SSD 512 and YOLOv5s models cannot identify them. The YOLOv5s-a and YOLOv5s-as models can detect small target fire sources with good adaptability. When using the edge processing method, the response period of YOLOv5s as model is 238 ms, which is 38.66% shorter than the centralized processing method.

Keywords

• external fire detection
• yolov5s
• deep learning
• adaptive attention mechanism
• multi source information fusion
• edge computing