第 卷 第 期 电讯技术
62 5
A Real-time Multi-scale Object Detection and Identification Algorithm
ZH U Pei pe i,WU Y uan,LAI Zuome i
(Southwest China Institute of Electronic Technology,Chengdu 610036,China)
Abstract
:I n unm anne d ae rial ve hic le( U AV) object detecti on and identifica tion task,targets’ sizes vary
enor mous ly alon e with t he hei ght of the UAV. Some norma l o bjec t detection and identification algori thms
can only extra ct lim ited sm all objec t fea tures, which results in lo w detec tion per forma nce. R esearchers also
deve lop some deep neural ne tworks for s mall objects. But these mode ls always have small recept ive field
and lo se some large targets’ features . In this paper,a real-tim e o bject dete ctio n and id enti fication netwo rk
based on im proved Feature Pyram id Networ k ( FP N) mo dule is p ropo sed to promote the p erforman ce of
multi-s cale objects detection. Thi s imp roved FPN mo dule in
一种实时多尺度目标检测识别算法 朱佩佩 来自淘豆网www.taodocs.com转载请标明出处.