摘要
本文首先构建了一个无人机检测数据集,然后利用Res2net提取目标多感受野特征并提出一种新的混合特征金字塔结构,从细粒度的多尺度特征提取和层级多尺度特征融合两个方面来提升网络性能,实现了一种基于多尺度特征融合的无人机目标检测网络.在本文构建的无人机检测数据集上进行实验,提出的网络对无人机、鸟类和普通飞机的识别率均能达到93%以上.
In this paper,a drone detection dataset was constructed firstly.Multi-sensory field features of the target were extracted by Res2net and a new hybrid feature pyramid structure was proposed,which improved network performance from the two respects of fine-grained multi-scale feature extraction and hierarchical multi-scale feature fusion,and then a drone target detection network based on multi-scale feature fusion was realized.The discriminations of drones,birds,and airplanes detected by the proposed network are over 93%in the experiments based on the drone detection dataset constructed in this paper.
作者
秦浪
赵德明
苏昕
曾浩
王正宁
QIN Lang;ZHAO Deming;SU Xin;ZENG Hao;WANG Zhengning(School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)
出处
《中国科技论文在线精品论文》
2020年第4期382-391,共10页
Highlights of Sciencepaper Online
基金
四川省科技计划(2018GZ0071)
关键词
人工智能
计算机视觉
无人机检测
多尺度特征融合
artificial intelligence
computer vision
drone detection
multi-scale feature fusion
