A real-time fire detection method from video for electric vehicle-charging stations based on improved YOLOX-tiny

For the current mainstream detection methods are difficult to achieve fire detection in outdoor electric vehicle-charging station, this paper proposes a real-time fire detection method from video for electric vehicle charging stations based on improved YOLOX-tiny. CBAM attention mechanism is introduced to concatenate the spatial and channel attention information, to preserve the salient features of different shapes of flames. Depthwise Separable Convolution is used to replace Conventional Convolution to reduce the number of Parameters and FLOPs of the network, and improves the speed of detection and the deployment of the model on the embedded side. CIoU loss function is used to replace bounding box regression loss function of YOLOX-tiny, and the aspect ratio limit mechanism is added to improve the convergence speed of the loss function and make the prediction results more consistent with the actual target. Experiment shows that mAP value of improved YOLOX-tiny is 94.05%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%$$\end{document}, Precision is 91.76%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%$$\end{document}, and Recall is 83.27%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%$$\end{document} on the embedded side, with the video detection speed is 20 fps, which meets the demand for real-time detection for electric vehicle-charging stations.