An efficient fire detection algorithm based on multi‐scale convolutional neural network

Abstract

Video fire detection (VFD) technology has shown a broad application prospect with the popularization of camera monitoring systems. Since the initial stage is the best time for firefighting, it's crucial to develop a robust algorithm for early warning. In this paper, an efficient VFD fusion algorithm is presented. First, the fire candidate areas (FCA) are located quickly based on low-level visual features to guarantee well timeliness. Furthermore, a multi-scale convolutional neural network with spatial pyramid pooling is built and trained on the dedicated flame data set without the requirement of sample labeling. In this case, FCA with different aspect ratios and scales can be accurately identified. The method is fully tested on various databases of arbitrarysized images and videos. Experimental results show that the proposed fusion algorithm, in addition to improving the detection efficiency, also ensures the accurate identification of flames on different scales.

Intelligent Methods for Forest Fire Detection Using Unmanned Aerial Vehicles

“…An efficient algorithm for quickly localizing fires based on low-level visual features is presented in [12]. It uses a multi-scale convolutional neural network architecture, the convolutional neural network, based on SPP (spatial pyramid pooling), which is trained on a dedicated dataset without the need for sample labeling.…”

Smoke and Flame Identification Method for the Entire Process of Grassland Fire Based on YOLOv5m-D and Static and Dynamic Characteristics

“…Scholars both domestically and internationally have conducted research on it. Hu Y et al conducted in-depth research on the internal gas characteristics of regional peat smoke events and their expansion into grassland fires by utilizing emission factors, thereby improving the differentiation between early ignition and later diffusion of grassland fires, and effectively enhancing the detection and recognition efficiency of grassland fires [7].Cheng Y and others proposed a fusion algorithm for efficient grassland fire detection, aiming at the relevant application problems of video fire detection technology under the development of camera technology, so as to enhance the actual detection efficiency as well as the accuracy of flame recognition at different scales [8]. In view of the low accuracy and high uncertainty of grassland fire detection, Chen J and others produced three-dimensional products for grassland fire detection by using spatial and practical resolution, thus effectively improving the accuracy of grassland fire recognition and providing data basis for the development of Moderate Resolution Imaging Spectroradiometer [9].…”