计算机应用研究2025，Vol.42Issue(10)：2893-2904,12.DOI:10.19734/j.issn.1001-3695.2025.03.0067
深度学习小目标检测算法综述
Survey on deep learning-based small object detection algorithms
摘要
Abstract
Small object detection is an important branch of object detection,playing a critical role in applications such as in-telligent surveillance,autonomous driving,medical image analysis,and remote sensing.However,due to the small pixel pro-portion of targets,weak feature representation,complex backgrounds,and the trade-off between detection accuracy and speed,significant technical challenges remain.Based on an extensive literature review,this paper outlined the technical challenges and solutions for small object detection,analyzed the core issues such as insufficient feature representation,inadequate utiliza-tion of contextual information,and sample imbalance.It summarized key advances,including multi-scale feature fusion,at-tention mechanisms,and knowledge distillation.Using MS COCO and TinyPerson datasets,this paper compared the detection efficiency and accuracy of mainstream algorithms,highlighting the strengths and weaknesses of different methods.Further-more,it explored the future research directions,such as generative feature learning,self-supervised learning,and dynamic ar-chitecture design,to provide insights for the further development of small object detection technologies.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Key words
small object detection/multi-scale feature fusion/attention mechanism/sample balance/lightweight network/robustness分类
信息技术与安全科学引用本文复制引用
张琴,郭为安..深度学习小目标检测算法综述[J].计算机应用研究,2025,42(10):2893-2904,12.基金项目
国家自然科学基金资助项目(62273263,72171172,71771176,92367101) （62273263,72171172,71771176,92367101）
上海市自然科学基金资助项目(23ZR1465400) （23ZR1465400）
福建省中青年教师教育科研项目(JAT220652) （JAT220652）
福州职业技术学院校级科研计划资助项目(FZYKJJHZD202401) （FZYKJJHZD202401）
福州职业技术学院引导计划资助项目(FZYKJZXYD202201) （FZYKJZXYD202201）
