摘要
针对YOLOv5算法对目标物体及局部特征定位不准确的问题,提出了一个基于YOLOv5的YOLOv5 ABN目标检测网络。首先为了得到准确的定位信息,本文将卷积注意力机制中的空间注意力机制与坐标注意力机制相结合,提出了一个融合空间特征信息与坐标信息的空间坐标注意力机制,在YOLOv5的Backbone网络中加入空间-坐标注意力模块,在Neck检测层引入加权双向特征金字塔网络,并对错框参数进行优化,提高了网络对样本模型及局部特征的检测精度及定位能力,分别用改进前后的算法对自定义数据集和康奈尔数据集进行训练。实验结果表明,改进后的算法性能要优于YOLOv5算法,在识别精度没有降低的前提下,对目标物体的定位更加准确,鲁棒性更好。
In order to solve the problem that YOLOv5 algorithm is not accurate in locating the target object and local features,a YOLOv5 ABN target detection network based on YOLOv5 is proposed.Firstly,in order to get accurate positioning information,this paper combines spatial attention mechanism and coordinate attention module in convolution attention mechanism.A spatial and coordinated attention module is proposed,which integrates spatial feature information and coordinate information.The spatial and coordinated attention module is added to the Backbone network of YOLOv5.The weighted bidirectional feature pyramid network is introduced into the Neck detection layer,and the staggered box parameters are optimized,which improves the detection accuracy and positioning ability of the network for sample models and local features.The customized data set and Cornell data set are trained with the improved algorithms respectively.The experiment results show that the improved algorithm is better than YOLOv5 algorithm,and the target object is more accurate and more robust without reducing the recognition accuracy.
作者
常春雷
运凯
王辉
张辉
徐赢
马崇瑞
Chang Chunlei;Yun Kai;Wang Hui;Zhang Hui;Xu Ying;Ma Chongrui(State Grid Xinjiang Information and Telecommunication Company,Wulumuqi 830002,China;Beijing China-power Information Technology Co.,Ltd.;State Grid Dalian Electric Power Supply Company)
出处
《单片机与嵌入式系统应用》
2023年第9期45-48,60,共5页
Microcontrollers & Embedded Systems
