采用置信度分层协同决策的目标检测方法

靳苗苗; 余序宜; 赵英海; 张会鹏; 尚颖慧; 梅魁志

doi:10.7652/xjtuxb202610002

您当前的位置：

首页 >

文章列表页 >

采用置信度分层协同决策的目标检测方法

专题智能驾驶的融合感知与决策规划 | 更新时间：2026-09-29

- 采用置信度分层协同决策的目标检测方法
- An Object Detection Method Based on Confidence-Stratified Collaborative Decision-Making
- 西安交通大学学报 2026年60卷第10期页码：22-33
- 作者机构：
  
  西安交通大学人机混合增强智能全国重点实验室，710049，西安
  西安交通大学人工智能与机器人研究所，710049，西安
  北京华航无线电测量研究所，100013，北京
- 作者简介：
  
  靳苗苗（1999-），女，博士生；
  梅魁志（通信作者），男，教授，博士生导师。
- 基金信息：
  
  国家自然科学基金资助项目(62576269)
- DOI：10.7652/xjtuxb202610002
  中图分类号： TP399
- 收稿：2026-02-26，
  
  网络首发：2026-05-11，
  
  纸质出版：2026-10-10
- 稿件说明：
移动端阅览
靳苗苗, 余序宜, 赵英海, 等. 采用置信度分层协同决策的目标检测方法[J/OL]. 西安交通大学学报,2026,60 (10):22-33. https://doi.org/10.7652/xjtuxb202610002.

JIN Miaomiao, YU Xuyi, ZHAO Yinghai, et al. An Object Detection Method Based on Confidence-Stratified Collaborative Decision-Making[J/OL]. Journal of Xi'an Jiaotong University,2026,60 (10):22-33. https://doi.org/10.7652/xjtuxb202610002.
靳苗苗, 余序宜, 赵英海, 等. 采用置信度分层协同决策的目标检测方法[J/OL]. 西安交通大学学报,2026,60 (10):22-33. https://doi.org/10.7652/xjtuxb202610002. DOI：

JIN Miaomiao, YU Xuyi, ZHAO Yinghai, et al. An Object Detection Method Based on Confidence-Stratified Collaborative Decision-Making[J/OL]. Journal of Xi'an Jiaotong University,2026,60 (10):22-33. https://doi.org/10.7652/xjtuxb202610002. DOI：

摘要

针对实时目标检测模型置信度校准不足以及后处理决策机制单一所引发的误检与漏检问题，提出了一种置信度分层协同决策架构。该架构以YOLOv5模型为基础检测网络，在推理阶段构建置信度分层机制，根据预测置信度分布，将候选框划分为高置信度直通路径和低置信度精细化判别路径。与此同时，引入极端梯度提升（XGBoost）模型，对低置信度样本进行二次分类与置信度修正，实现检测结果的差异化处理与计算资源的分配。在融合阶段设计统一的决策整合模块，对不同路径输出进行结构化融合与筛选，在不改变主干网络结构和基本不增加计算复杂度的前提下，提升了整体检测性能与预测可靠性。研究结果表明：在2080Ti测试环境下、COCO数据集上，改进后YOLOv5s和YOLOv5m模型的平均精度均值

AP，50-95

分别提升了1.98%、1.99%；泛化至YOLOv8m时，模型的

AP，50-95

为55.07%，推理速度为165.4帧/s。该方法构建了深度神经网络与传统机器学习模型协同决策的新型融合范式，为高可靠领域目标识别系统提供了一种兼顾精度、效率与可信度的技术路径。

Abstract

To address the issues of false positives and false negatives caused by insufficient confidence calibration and single post-processing decision mechanisms in real-time object detection models

a confidence-stratified collaborative decision-making framework is p

roposed. Using the YOLOv5model as the base detection network

the framework constructs a confidence stratification mechanism during the inference stage. Based on the predicted confidence distribution

candidate bounding boxes are divided into a high-confidence passthrough path and a low-confidence refined discrimination path. Meanwhile

an extreme gradient boosting (XGBoost) model is introduced to perform secondary classification and confidence recalibration on low-confidence samples

achieving differentiated processing of detection results and optimized allocation of computational resources. In the fusion stage

a unified decision integration module is designed to perform structured fusion and filtering on the outputs from different paths. This improves overall detection performance and predictive reliability without altering the backbone network structure or significantly increasing computational complexity. Experimental results show that

validated on the COCO dataset in a 2080Ti testing environment

the mean average precision

50-95

of the improved YOLOv5s and YOLOv5m models increased by 1.98% and 1.99%

respectively. When generalized to YOLOv8m

the model achieved an

50-95

of 55.07% with an inference speed of 165.4frame/s. This method establishes a novel fusion paradigm for collaborative decision-making between deep neural networks and traditional machine learning models

providing a technical pathway that balances accuracy

efficiency

and credibility for object recognition systems in high-reliability domains.

关键词

Keywords

references

Tang Shiyi, Zhang Shu, Fang Yini.HIC-YOLOv5: improved YOLOv5for small object detection [C]//2024IEEE International Conference on Robotics and Automation (ICRA).Piscataway, NJ, USA: IEEE, 2024: 6614-6619.

邱天衡, 王玲, 王鹏, 等.基于改进YOLOv5的目标检测算法研究[J].计算机工程与应用, 2022, 58(13): 63-73.

Qiu Tianheng, Wang Ling, Wang Peng, et al. Research on object detection algorithm based on improved YOLOv5[J]. Computer Engineering and Applications, 2022, 58(13): 63-73.

邵延华, 张铎, 楚红雨, 等.基于深度学习的YOLO目标检测综述[J].电子与信息学报, 2022, 44(10): 3697-3708.

Shao Yanhua, Zhang Duo, Chu Hongyu, et al. A review of YOLO object detection based on deep learning[J]. Journal of Electronics & Information Technology, 2022, 44(10): 3697-3708.

Ge Zheng, Liu Songtao, Wang Feng, et al. YOLOX: exceeding YOLO series in 2021[PP/OL]. arXiv (2021-07-18)[2025-07-18]. https://arxiv.org/abs/2107.08430.

Jocher G, Chaurasia A, Stoken A, et al. ultralytics/yolov5: v7.0-YOLOv5SOTA realtime instance segmentation [EB/OL]. (2022-11-22)[2025-12-10]. https://github.com/ultralytics/yolov5/releases/tag/v7.0.

Song Baoye, Chen Jianyu, Liu Weibo, et al. YOLO-ELWNet: a lightweight object detection network [J]. Neurocomputing, 2025, 636: 129904.

张凡, 张鹏超, 王磊, 等.基于YOLOv5s的轻量化朱鹮检测算法研究[J].西安交通大学学报, 2023, 57(1): 110-121.

Zhang Fan, Zhang Pengchao, Wang Lei, et al. Research on lightweight crested ibis detection algorithm based on YOLOv5s[J].Journal of Xi'an Jiaotong University, 2023, 57(1): 110-121.

Feng Yifan, Huang Jiangang, Du Shaoyi, et al. Hyper-YOLO: when visual object detection meets hypergraph computation [J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(4): 2388-2401.

Ji Chunlin, Yu Tao, Gao Peng, et al. Yolo-tla: an efficient and lightweight small object detection model based on YOLOv5[J].Journal of Real-Time Image Processing, 2024, 21(4): 141.

Zhao Yian, LüWenyu, Xu Shangliang, et al. DETRs beat YOLOs on real-time object detection [C]//2024IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2024: 16965-16974.

Huang Zixin, Tao Xuesong, Liu Xinyuan.NAN-DETR: noising multi-anchor makes DETR better for object detection [J].Frontiers in Neurorobotics, 2024, 18: 1484088.

Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection [C]//2017IEEE International Conference on Computer Vision (ICCV).Piscataway, NJ, USA: IEEE, 2017: 2999-3007.

Zhu Chenchen, Chen Fangyi, Shen Zhiqiang, et al. Soft anchor-point object detection [C]//Computer Vision-ECCV 2020.Cham, Germany: Springer International Publishing, 2020: 91-107.

Betti A, Tucci M.YOLO-S: a lightweight and accurate YOLO-like network for small target detection in aerial imagery [J].Sensors, 2023, 23(4): 1865.

Yoon I, Kwon H, Kim J, et al. Enhancing source-free domain adaptive object detection with low-confidence pseudo label distillation [C]//Computer Vision-ECCV 2024.Cham, Switzerland: Springer Nature Switzerland, 2025: 337-353.

Bodla N, Singh B, Chellappa R, et al. Soft-NMS: improving object detection with one line of code [C]//2017IEEE International Conference on Computer Vision (ICCV).Piscataway, NJ, USA: IEEE, 2017: 5562-5570.

Guo Chuan, Pleiss G, Sun Yu, et al. On calibration of modern neural networks [C]//Proceedings of the 34th International Conference on Machine Learning.Chia Laguna Resort, Sardinia, Italy: PMLR, 2017: 1321-1330.

Dussert G, Chamaillé-Jammes S, Dray S, et al. Being confident in confidence scores: calibration in deep learning models for camera trap image sequences [J]. Remote Sensing in Ecology and Conservation, 2025, 11(1): 88-99.

Rai A, Athya D.AdaptiVision: dynamic confidence thresholding for enhanced object detection [C]//2025International Conference on Engineering Innovations and Technologies (ICoEIT).Piscataway, NJ, USA: IEEE, 2025: 592-598.

Ding Zhicheng, Lai Zhixin, Li Siyang, et al. Confidence trigger detection: accelerating real-time trackingby-detection systems [C]//20245th International Conference on Electronic Communication and Artificial Intelligence (ICECAI).Piscataway, NJ, USA: IEEE, 2024: 587-592.

Thatikonda M, Pk M K, Amsaad F.A novel dynamic confidence threshold estimation AI algorithm for enhanced object detection [C]//NAECON 2024-IEEE National Aerospace and Electronics Conference.Piscataway, NJ, USA: IEEE, 2024: 359-363.

Dave P, Chandarana A, Goel P, et al. An amalgamation of YOLOv4and XGBoost for next-gen smart traffic management system [J].PeerJ Computer Science, 2021, 7: e586.

Chen Tianqi, Guestrin C.XGBoost: a scalable tree boosting system [C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York, NY, USA: ACM, 2016: 785-794.

Babayomi M, Olagbaju O A, Kadiri A A.Convolutional XGBoost (C-XGBOOST)model for brain tumor detection [EB/OL].arXiv (2023-01-05)[2025-11-05]. https://arxiv.org/abs/2301.02317.

Chen Yuming, Yuan Xinbin, Wang Jiabao, et al. YO-LO-MS: rethinking multi-scale representation learning for real-time object detection [J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(6): 4240-4252.

Reddy K G, Basha S S.Real time object identification: a study on COCO dataset [C]//Proceedings of the International Conference on Advanced Materials, Manufacturing and Sustainable Development (ICAMMSD 2024).Dordrecht, the Netherlands: Atlantis Press, 2025: 860-870.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

复杂天气下铁轨异物入侵目标检测研究

机器学习在沸腾传热过程中的应用进展

跨声速压气机转子通用k-ω湍流模型的机器学习辅助优化

面向CO2捕集的吸附剂研究：从传统设计到机器学习

可解释性机器学习驱动的煤和污泥超临界水共气化产物预测研究