摘要
提出一种基于改进的Yolo v3模型的设备状态检测方法,以实现精准的信号灯检测.工作流程为:(1)通过图像采集设备获取信号灯图像并预处理:(2)通过改进的Yolo v3模型进行信号灯检测.其中改进的Yolo v3模型增加了特征图的加权融合步骤,提高了对小目标检测识别的性能.实验结果表明,相比于原始Yolo v3模型.采用改进的Yolo v3模型进行信号灯检测识别时能够取得更高的交并比,同时,模型的召回率和精度也有相应提升.
A device state detection method was proposed based on improved Yolo v3 model to achieve accurate signal light detection.The entire workflow included two steps:(1)To obtain the image of the signal light through the image acquisition device and normalize it;(2)To perform signal light detection with the improved Yolo v3 model.The improved Yolo v3 model employed weighted fusion of feature maps on three scales to improve its performance in the field of small target detection and recognition.The experiment results showed that comparing with the original Yolo v3 model,the improved model could achieve a higher intersection over union for signal light detection and recognition,as well as recall and accuracy.
作者
王鑫
曾愚
魏怀灏
李嘉周
吴斗
范玉强
Wang Xin;Zeng Yu;Wei Huaihao;Li Jiazhou;Wu Dou;Fan Yuqiang(Information and Communication Company,State Grid Sichuan,Chengdu 610041,China)
出处
《南开大学学报（自然科学版）》
CAS
CSCD
北大核心
2021年第6期7-11,共5页
Acta Scientiarum Naturalium Universitatis Nankaiensis
