摘要
近年来目标检测技术发展十分迅速,出现了很多优秀的目标检测算法,诸如Faster RCNN、YOLO和SSD等,其中尤以SSD目标检测算法表现突出,其运行速度可以和YOLO媲美,检测精度可以和Faster RCNN媲美,但SSD目标检测算法已生成六个特征图,接着单独送进网络里面检测,并没有考虑到多尺度特征融合问题。基于此,文章对SSD目标检测算法的多尺度特征融合技术进行了研究,使用特征图融合技术和三支路多尺度特征融合技术改进SSD目标检测算法,并获得更优的效果。
In recent years,the development of target detection technology has been very rapid,and there have been many excellent target detection algorithms,such as Faster RCNN,YOLO,and SSD.Among them,the SSD target detection algorithm has outstanding performance,its running speed can be comparable to YOLO,and the detection accuracy can be comparable to Faster RCNN.The SSD target detection algorithm generated six feature maps separately and sent them to the network for detection,and does not take into account the problem of multi-scale feature fusion.Based on this,the article studies the multi-scale feature fusion technology of the SSD target detection algorithm,and uses feature map fusion and three-branch multi-scale feature fusion technology to improve the SSD target detection algorithm and obtain better results.
作者
黄和锟
HUANG Hekun(Guangdong University of Technology,Guangzhou 510006,China)
出处
《现代信息科技》
2020年第18期122-124,共3页
Modern Information Technology