A Real-Time Smoke and Fire Warning Detection Method Based on an Improved YOLOv5 Model

Abstract

Smoke and fire detection technology has been very mature with the development of deep learning. Still, the early warning detection methods for fires in some factories and gas stations are insufficient. Especially, no clear advancement has been seen in terms of small targets. So, we propose a smoke and fire detection method based on YOLOv5 network as a solution to address this problem. We combine the original YOLOv5 backbone network with an attention mechanism to improve the ability of small object detection. Existing smoke and fire detection methods encounter some challenges in practical industrial applications. Deep neural network learning models require powerful computing processors and servers; secondly, the models trained by general neural networks are too large to be deployed. The improved network can reduce the model size while maintaining high detection accuracy, thereby reducing the storage space of edge devices. In addition, this study uses Deepstream SDK to push streaming fire video, and combines with Jetson NX Xavier edge device for model deployment to realize real-time fireworks warning.