深度学习小目标检测算法综述

打开文本图片集

关键词：小目标检测；多尺度特征融合；注意力机制；样本均衡；轻量级网络；鲁棒性中图分类号：TP301.6 文献标志码：A 文章编号：1001-3695（2025）10-002-2893-12doi：10.19734/j. issn.1001-3695.2025.03.0067

Survey on deep learning-based small object detection algorithms

Zhang Qin 1，2 ，Guo Weian2+ （1.Collegefoc，n;dU versity，Shanghai201804，China）

Abstract：Smallobjectdetectionisanimportantbranchofobject detection，playingacriticalroleinapplications suchas intellgent surveillance，autonomous driving，medicalimage analysis，andremotesensing.However，duetothesmallpielpro portionof targets，weak featurerepresentationcomplexbackgrounds，andthetrade-offbetweendetectionaccracyandpeed， significanttechnicalchalengesremain.Basedonanextensiveliteraturereview，thispaperoutlinedthetechnicalchalenges andsolutionsforsmallobjectdetection，analyzedtecoreisuessuchasinsuicientfeaturerepresentation，inadequateuilizationofcontextualinformation，andsampleimbalance.Itsummarizedkeyadvances，including multi-scalefeaturefusion，attention mechanisms，andknowledgedistllation.Using MS COCOand TinyPerson datasets，this papercompared thedetection eficiencyandaccuracyofmainstreamalgorithms，highlightingthestrengthsandweaknessesofdiferentmethods.Furthermore，it exploredthfutureresearchdirections，suchas generativefeaturelearning，self-supervised learning，anddyamicarchitecture design，to provide insights for the further development of small object detection technologies.

Key words：smallobjectdetection；multi-scale feature fusion；attentionmechanism；samplebalance；lightweightnetwork; robustness

0 引言

小目标检测是目标检测领域的重要分支，近年来随着深度学习技术的迅速发展，已成为计算机视觉研究中的热点问题。（剩余40167字）