【摘要】 针对目前对于充电站内车辆起火现象的检测精度较低、检测速度慢等问题,从实用化角度出发,提出了一种基于YOLOv5s改进的车辆起火检测方法YOLOv5s-Fast。首先在Backbone网络中采用了全局上下文注意力机制与C3模块进行融合,成为一种新的特征提取的模块C3GC,增强模型提取特征的能力,减少了计算量;其次在Neck网络中采用了轻量级上采样算子,能够根据输入图像进行自适应的上采样,提升了检测精度;最后引入解耦头,提高了目标检测的准确率与效率。实验结果表明,所提出的方法YOLOv5s-Fast与原YOLOv5s相比,平均精度提升了4.9%、检测帧率由原先的46 fps提高到59 fps,方法更加实用化。更多还原

【Abstract】 In response to the current problems of low detection accuracy and slow detection speed of vehicle fires in charging stations, this paper proposes a vehicle fire detection method YOLOv5s-Fast based on YOLOv5s improvement from a practical perspective. This article first uses the global context attention mechanism and C3 module to fuse in the Backbone network, becoming a new feature extraction module C3GC, enhancing the model’s ability to extract features and reducing computational complexity. Secondly, in the Neck network, this paper adopts a lightweight upsampling operator that can adaptively upsample based on the input image, improving detection accuracy. Finally, this article introduces a decoupling head to improve the accuracy and efficiency of object detection. The experimental results show that the proposed method YOLOv5-Fast has an average accuracy improvement of 4.9% and frame rate increase from 46 fps to 59 fps compared to the original YOLOv5s, making the method more practical.更多还原