摘要
随着人工智能的兴起,人工智能技术已经被广泛的使用。在智慧交通系统中,对路灯对象的检测需要依靠人工智能技术。然而,现有技术在复杂场景图像中检测路灯对象普遍存在检测精度较低的问题。为此,本文提出一种复杂场景图像中路灯对象的人工智能检测算法。该算法首先引入坐标注意力机制,将位置信息编码到通道注意力中,从而获得复杂场景图像中大范围的位置信息;其次,通过双向融合方式将不同尺度的特征图进行融合,形成多尺度特征图,从而增强对复杂场景图像中不同大小和尺度路灯对象的检测能力;最后,通过预设锚框,在不同尺度的特征图上进行路灯对象检测。实验结果表明,本文提出的算法在自建数据集中的精确率为91.44%,较YOLOV5算法增加了2.52%。该算法在检测精确度方面取得了进步,为智慧交通系统提供了有力的支持。
With the rise of artificial intelligence,AI technology has been widely used.In smart transportation systems,the detection of streetlight objects relies on AI technology.However,existing techniques often suffer from low detection accuracy when dealing with streetlight objects in complex scene images.To this end,this article proposes an artificial intelligence detection algorithm for streetlight objects in complex scene images.The algorithm first introduces a coordinate attention mechanism to encode the position information into the channel attention,thereby obtaining a wide range of position information in complex scene images.Secondly,it performs bidirectional fusion to merge feature maps of different scales,creating multi-scale feature maps to enhance the detection capability of streetlight objects of various sizes and scales in complex scene images.Finally,using preset anchor frames,the algorithm performs streetlight object detection on feature maps of different scales.Experimental results demonstrate that the proposed algorithm achieves an accuracy of 91.44%on a self-built dataset,which is a 2.52%improvement over the YOLOv5 algorithm.The algorithm shows advancements in detection accuracy and provides strong support for smart transportation systems.
作者
陈春根
姜玉稀
韩大专
CHEN Chungen;JIANG Yuxi;HAN Dazhuan(Shanghai Sansi Electronic Engineering Co.Ltd.,Shanghai 201100,China;School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China)
出处
《照明工程学报》
2024年第4期168-172,共5页
China Illuminating Engineering Journal
基金
国家重点研发计划(2017YFB0403500)
上海市闵行区重大产业技术攻关计划(2022MH-ZD19)
上海市院士(专家)工作站建站项目。
关键词
智慧交通系统
坐标注意力机制
双向融合
路灯对象检测
smart transportation systems
coordinate attention mechanism
bidirectional fusion
streetlight object detection
