摘要
针对目标检测中小目标及目标尺寸相差较大难以检测的问题,提出了改进YOLOv3的AR-Yolov3(Add Receptive-fieldYolov3)网络。利用改进的感受野模块E-RBF增大网络感受野特性,使网络提取到深层次的全局性语义特征,提高尺寸相差较大目标的检测精度;使用CSP双向特征金字塔(D_CSP-FPN)网络,实现不同层级特征信息充分利用,提高网络浅层预测分类能力和深层定位能力。实验结果显示,AR-Yolov3网络模型检测性能在小目标和多尺寸目标中较当下主流模型有更好的检测效果。
Aimed at the problem that it is difficult to detect small and medium-sized targets in object detection,the AR-Yolov3(Add Receptive Field YOLOv3)network improved by YOLOv3 is proposed.Firstly,the improved receptive field module E-RBF was used to increase the characteristics of the receptive field,so that the deep global semantic features can be extracted from the network,and the detection accuracy of targets with large size difference can be improved.Then,the D_CSP-FPN network was used to make full use of the feature information at different levels,and improve the prediction and classification ability of the network at the shallow level and the location ability at the deep level.Finally,the experimental results show that the detection performance of AR-Yolov3 network model is better than the current mainstream model in the detection of small targets and multi-size targets.
作者
白鹏飞
伊力哈木·亚尔买买提
BAI Peng-fei;Yilihamu Yaermaimaiti(School of Electrical Engineering,Xinjaing University,Urumqi Xinjaing 830047,China)
出处
《计算机仿真》
北大核心
2023年第3期252-257,共6页
Computer Simulation
基金
国家自然科学基金(61866037,61462082)。
关键词
空间感受野
双向金字塔
梯度推理能力
Space receptive field
Two-way pyramid
Gradient reasoning ability
