现代电子技术2025，Vol.48Issue(13)：50-56,7.DOI:10.16652/j.issn.1004-373x.2025.13.007
基于改进RT-DETR的小目标检测算法
Small object detection algorithm based on improved RT-DETR
摘要
Abstract
Small object detection often faces challenges such as missed detections and false positives due to the small proportion of the object to the image and the limited semantic information.In view of this,an improved RT-DETR based small object detection model is proposed to enhance detection performance while ensuring real-time performance.The backbone network of the RT-DETR model is modified by designing a partially re-parameterized convolution module to improve feature extraction efficiency.An efficient multi-scale attention(EMA)mechanism is introduced to aggregate spatial and cross-spatial information.The HiLo attention mechanism is employed in the AIFI encoder to reduce the computational costs and enhance the robustness of the detection algorithm.Experiments were conducted on the FloW-Img dataset of the small objects on water.The results show that both the missed detection rate and the false positive rate of the model based on the improved RT-DETR are reduced in comparison with the baseline model RT-DETR.On the test set,the mAP@0.5 of the proposed algorithm achieves 0.841 and its mAP@0.5:0.95 is 0.394,representing improvements of 5.5%and 3.7%,respectively,over the baseline model RT-DETR.The detection performance surpasses those of both the baseline model and the object detection models of YOLO series.关键词
小目标检测/RT-DETR模型/PCov/Transformer/EMA/HiLo注意力机制Key words
small object detection/RT-DETR model/PCov/Transformer/EMA/HiLo attention mechanism分类
电子信息工程引用本文复制引用
王康,王小林,刘心智,邓健志..基于改进RT-DETR的小目标检测算法[J].现代电子技术,2025,48(13):50-56,7.基金项目
广西科技重大专项绿色高效平陆运河建设"尖峰"专项(桂科AA23062035-2) （桂科AA23062035-2）
