计算机与数字工程2025，Vol.53Issue(1)：115-118,175,5.DOI:10.3969/j.issn.1672-9722.2025.01.022
基于双向融合和特征增强的SSD小目标检测算法
SSD Small Target Detection Algorithm Based on Bidirectional Fusion and Feature Ehancement
摘要
Abstract
An improved algorithm based on bidirectional feature fusion and lightweight feature enhancement is proposed to solve the problem of low accuracy of SSD algorithm for small target detection.The feature fusion module of the algorithm constructs a two-way network channel for feature fusion by deep separable convolution and up-sampling operation,which makes the detail ap-parent information and high-dimensional semantic information from different scale feature layers of the backbone network more fully fused.The feature enhancement module of the algorithm adopts the attention mechanism and feature graph stitching method.The at-tention mechanism makes the model focus on the target region rather than the background through weight distribution.Feature graph splicing convolution of only part channels can increase semantic information of feature layer on the premise of ensuring detection speed.By combining the outputs of these two parts,double enhancement of input features can be achieved.Experiments demon-strate its effectiveness,BilFE-SSD improves 11.1%compared with SSD algorithm index mAP on HRRSD small target data sets.关键词
小目标检测/特征融合/特征增强/注意力机制Key words
small target detection/feature fusion/features enhancement/attention mechanism分类
信息技术与安全科学引用本文复制引用
吴钟仁,周莲英,丁腊春..基于双向融合和特征增强的SSD小目标检测算法[J].计算机与数字工程,2025,53(1):115-118,175,5.基金项目
慢病本体知识库开发(编号:8421170004) （编号:8421170004）
市智慧妇幼信息平台本体知识库系统(编号:20180477)资助. （编号:20180477）
