《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (9): 2769-2775.DOI: 10.11772/j.issn.1001-9081.2025081018

• 人工智能 • 上一篇    

基于协同查询优化的长尾目标检测框架

陈江彦, 王彦丹, 刘翼虎, 马应龙()   

  1. 华北电力大学 控制与计算机工程学院,北京 102206
  • 收稿日期:2025-09-04 修回日期:2025-11-25 接受日期:2025-11-27 发布日期:2025-12-01 出版日期:2026-09-10
  • 通讯作者: 马应龙
  • 作者简介:陈江彦(2001—),男,湖北麻城人,硕士研究生,主要研究方向:深度学习、长尾目标检测
    王彦丹(1998—),女,山东青岛人,博士研究生,主要研究方向:深度学习、类别不平衡学习
    刘翼虎(1998—),男,山东潍坊人,博士研究生,主要研究方向:大数据
    马应龙(1976—),男,陕西咸阳人,教授,博士,主要研究方向:深度学习、分布式计算。
  • 基金资助:
    国家自然科学基金面上项目(62072450)

Collaborative query optimization-based framework for long-tailed object detection

Jiangyan CHEN, Yandan WANG, Yihu LIU, Yinglong MA()   

  1. School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China
  • Received:2025-09-04 Revised:2025-11-25 Accepted:2025-11-27 Online:2025-12-01 Published:2026-09-10
  • Contact: Yinglong MA
  • About author:CHEN Jiangyan, born in 2001, M. S. candidate. His research interests include deep learning, long-tailed object detection.
    WANG Yandan, born in 1998, Ph. D. candidate. Her research interests include deep learning, class imbalance learning.
    LIU Yihu, born in 1998, Ph. D. candidate. His research interests include big data.
    MA Yinglong, born in 1976, Ph. D., professor. His research interests include deep learning, distributed computing.
  • Supported by:
    National Natural Science Foundation of China(62072450)

摘要:

针对基于Transformer的目标检测方法在面对长尾分布时通常会性能显著下降的问题,提出一种基于协同查询优化的单阶段端到端长尾目标检测框架CQ-DETR(Collaborative Query optimization-based DEtection TRansformer)。设计层自适应编码器融合(LAEF)模块动态整合多尺度编码器特征,兼顾高层语义与底层细节;设计特征感知查询生成(FQG)模块从图像特征中动态生成内容感知的查询,以增强初始内容查询对潜在目标的表征能力;提出类别-定位联合感知查询选择(CLJS)机制以协同优化类别覆盖范围与定位精度。实验结果表明,CQ-DETR在长尾目标检测基准数据集LVIS v1.0上优于RichSem(Rich Semantics)方法,平均精度(AP)和稀有类别的AP(APr)分别提升1.4和1.5个百分点,验证了所提框架在类别不平衡场景下的有效性;同时,在相对平衡的COCO 2017数据集上,与DINO(DETR with Improved deNoising anchOr box)相比,CQ-DETR的AP提升1.1个百分点,验证了该框架在通用目标检测场景下具有良好的泛化能力。

关键词: 目标检测, Transformer, 长尾分布, 多尺度特征融合, 查询选择机制

Abstract:

To address the problem that significant performance degradation of Transformer-based object detection methods when facing long-tailed distributions, a collaborative query optimization-based single-stage end-to-end long-tailed object detection framework, named CQ-DETR (Collaborative Query optimization-based DEtection TRansformer), was proposed. In the framework, a Layer-Adaptive Encoder Fusion (LAEF) module was designed to integrate multi-scale encoder features dynamically, so as to consider both high-level semantics and low-level details; a Feature-aware Query Generation (FQG) module was designed to generate content-aware queries dynamically from image features, thereby enhancing the representation capability of initial content queries for potential objects; a Category-Localization Joint-aware query Selection (CLJS) mechanism was proposed to achieve collaborative optimization of category coverage and localization accuracy. Experimental results indicate that CQ-DETR is superior to RichSem (Rich Semantics) method on the long-tailed object detection benchmark dataset LVIS v1.0, with the Average Precision (AP) and the AP of rare categories (APr) improved by 1.4 and 1.5 percentage points, respectively, verifying the effectiveness of the proposed framework in category-imbalanced scenarios; meanwhile, on the relatively balanced COCO 2017 dataset, compared with DINO (DETR with Improved deNoising anchOr box), CQ-DETR has the AP improved by 1.1 percentage points, verifying the good generalization ability of this framework in general object detection scenarios.

Key words: object detection, Transformer, long-tailed distribution, multi-scale feature fusion, query selection mechanism

中图分类号: