| 注册
首页|期刊导航|无线电工程|融合全局与局部注意力的小目标检测

融合全局与局部注意力的小目标检测

周叶 陈慈发 张上

无线电工程2026,Vol.56Issue(7):1185-1195,11.
✕
无线电工程2026,Vol.56Issue(7):1185-1195,11.DOI:10.3969/j.issn.1003-3106.2026.07.006

融合全局与局部注意力的小目标检测

Small Object Detection Based on Global-Local Attention Fusion

周叶 1陈慈发 2张上1

作者信息

  • 1. 三峡大学 水电工程智能视觉检测湖北省重点实验室,湖北 宜昌 443002||三峡大学 计算机与信息学院,湖北 宜昌 443002||三峡大学 湖北省建筑质量检测装备工程技术研究中心,湖北 宜昌 443002
  • 2. 三峡大学 计算机与信息学院,湖北 宜昌 443002||三峡大学 湖北省建筑质量检测装备工程技术研究中心,湖北 宜昌 443002||荆楚理工学院 大数据研究中心,湖北 荆门 448004
  • 折叠

摘要

Abstract

To address the challenges of small object scale,blurred boundaries,complex backgrounds,and frequent missed detections in unmanned aerial vehicle aerial images,a global-local attention-based detection network incorporating mask generation based distillation and slice-aided inference is proposed.Built upon the YOLOv11 framework,a Gated Tri-branch Spatial-Channel Collaborative Attention module(GT-SCCA)is designed to enhance the representation of local fine-grained features and suppress background interferences through the collaborative interaction of Channel Attention(CA),direction-aware spatial attention and Object-Aware(OA)branch.In addition,a Low-Resolution Self-Attention(LRSA)mechanism is introduced to establish global contextual dependencies with low computational overhead,thereby improving the model's holistic perception of small objects.A mask generation based distillation strategy is further employed to guide the student network toward learning more discriminative feature representations and enhancing feature expressiveness.During the inference stage,slice-aided hyper inference is adopted to strengthen feature perception in small-object regions and reduce missed detections.Extensive experiments are conducted on the VisDrone-DET2019 and ITCVD unmanned aerial vehicle aerial image datasets,and comparisons are performed with several mainstream object detection algorithms.Experimental results demonstrate that the proposed method improves the mean Average Precision(mAP)by 2.94%and 3.80%over the baseline model on the VisDrone-DET2019 and the ITCVD dataset respectively,achieving significantly enhanced small-object detection performance while maintaining high detection efficiency.

关键词

小目标检测/YOLOv11/全局-局部注意力/特征提取能力增强

Key words

small object detection/YOLOv11/global-local attention/enhanced feature extraction capability

分类

信息技术与安全科学

引用本文复制引用

周叶,陈慈发,张上..融合全局与局部注意力的小目标检测[J].无线电工程,2026,56(7):1185-1195,11.

基金项目

湖北省大学生创新创业训练计划(202311075046) (202311075046)

国家大学生创新创业训练计划(2020111075012,202011075013) (2020111075012,202011075013)

湖北省中央引导地方专项(2024BSB002)Hubei Provincial College Students'Innovation and Entrepreneurship Training Program(202311075046) (2024BSB002)

National College Students'Inno-vation and Entrepreneurship Training Program(2020111075012,202011075013) (2020111075012,202011075013)

Central Government Special Fund for Local Science and Technology De-velopment of Hubei Province(2024BSB002) (2024BSB002)

无线电工程

1003-3106

访问量0
|
下载量0
段落导航相关论文