您的位置：山东大学 -> 科技期刊社 -> 《山东大学学报（理学版）》

《山东大学学报(理学版)》 ›› 2026, Vol. 61 ›› Issue (1): 15-25.doi: 10.6040/j.issn.1671-9352.8.2024.013

• • 上一篇    

融合多尺度注意力机制和改进特征融合的轻量化水面小目标检测模型

仲尚1,马丽1,2*,刘文哲1,李雨豪1   

  1. 1.河北地质大学信息工程学院, 河北 石家庄 052161;2.河北地质大学智能传感物联网技术河北省工程研究中心, 河北 石家庄 052161
  • 发布日期:2026-01-15
  • 通讯作者: 马丽(1977— ),女,副教授,硕士生导师,博士,研究方向为知识发现与粒计算、机器学习. E-mail:mali_new@163.com
  • 作者简介:仲尚(1999— ),男,硕士研究生,研究方向为机器学习、目标检测. E-mail:2371019647@qq.com*通信作者:马丽(1977— ),女,副教授,硕士生导师,博士,研究方向为知识发现与粒计算、机器学习. E-mail:mali_new@163.com
  • 基金资助:
    河北省高等学校科学技术研究重点资助项目(ZD2018043);河北省教育科学规划课题一般资助课题项目(2303121);河北地质大学博士基金资助项目(BQ2017045)

Lightweight water surface small object detection model with multi-scale attention mechanism and improved feature fusion

ZHONG Shang1, MA Li1,2*, LIU Wenzhe1, LI Yuhao1   

  1. 1. College of Information Engineering, Hebei GEO University, Shijiazhuang 052161, Hebei, China;
    2. Intelligent Sensor Network Engineering Research Center of Hebei Province, Hebei GEO University, Shijiazhuang 052161, Hebei, China
  • Published:2026-01-15

PDF (PC)

12

摘要： 在复杂水面场景下,针对小目标检测精度较低、漏检率高和计算资源有限的问题,提出一种融合多尺度注意力机制和改进特征融合的轻量化水面小目标检测模型。根据中心度理论,设计新的主干网络,利用多尺度注意力机制增强模型的特征提取能力,利用部分卷积从减少特征图冗余的角度改进颈部网络,有效降低模型计算量。使用大型可分离核注意力模块改进快速空间金字塔池化模块,提升模型的特征融合能力。实验结果表明,与其他模型相比,本文模型的检测精度高、漏检率低、参数量少。

关键词: 小目标检测, 特征融合, 多尺度注意力机制, 特征图冗余

Abstract: In complex water surface scenarios, addressing the issues of low detection accuracy, high missed detection rates, and limited computational resources for small target detection, this paper proposes a lightweight water surface small object detection model with multi-scale attention mechanism and improved feature fusion. Based on centerness theory, a new backbone network is designed, leveraging the multi-scale attention mechanism to enhance the models feature extraction capabilities. Partial convolution is used to improve the neck network by reducing feature map redundancy, effectively lowering the models computational load. A large separable kernel attention module is employed to improve the spatial pyramid pooling module, enhancing the models feature fusion ability. Experimental results demonstrate that, compared to other models, the proposed model achieves higher detection accuracy, lower missed detection rates, and fewer parameters.

Key words: small target detection, feature fusion, multi-scale attention mechanism, feature map redundancy

中图分类号: 

  • TP391
[1] GIRSHICK R. Fast R-CNN[C] //Proceedings of the IEEE International Conference on Computer Vision(ICCV). Boston: IEEE, 2015:1440-1448.
[2] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas: IEEE, 2016:779-788.
[3] LIU W, ANGUELOW D, ERHAN D, et al. SSD: single shot multibox detector[C] //Proceedings of the 14th European Conference on Computer Vision(ECCV). Berlin: Springer, 2016:21-37.
[4] REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6):1137-1149.
[5] TAO Kong, YAO Anbang, CHEN Yurong, et al. HyperNet: Towards accurate region proposal generation and joint object detection[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas: IEEE, 2016:845-853.
[6] LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C] //Proceedings of the 30th IEEE/CVF Confenence on Computer Vision and Pattern Recognition(CVPR). Honolulu: IEEE, 2017:936-944.
[7] CHEN C Y, LIU M Y, TUZEL O, et al. R-CNN for small object detection[C] //Proceedings of the 13th Asian Conference on Computer Vision. Berlin: Springer, 2016:214-230.
[8] 戚玲珑,高建瓴. 基于改进YOLOv7的小目标检测[J]. 计算机工程,2023,49(1):41-48. QI Linglong, GAO Jianling. Small target detection based on improved YOLOv7[J]. Computer Engineering, 2023, 49(1):41-48.
[9] LIN Feng, HOU Tian, JIN Qiannan, et al. Improved YOLO based detection on algoruthm for floating debris in waterway[J]. Entropy, 2021, 23(9):1111.
[10] 王林,汪钰婷. 基于加强特征融合的轻量化船舶目标检测[J]. 计算机系统应用,2023,32(2):288-294. WANG Lin, WANG Yuting. Lightweight ship target detection based on enhanced feature fusion[J]. Computer System Applications, 2023, 32(2):288-294.
[11] OUYANG Daliang, HE Su, ZHAN Jian, et al. Efficient multi-scale attention moudle with cross-spatial learning[C] //Proceedings of the IEEE Interrnational Conference on Acoustics Speech and Signal Processing(ICASSP). Rhodes Island: IEEE, 2023:1-5.
[12] CHEN J R, KAO S, HE H, et al. Run, dont walk: chasing higher flops for faster neural networks[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Vancouver: IEEE, 2023:12021-12031.
[13] LAU K W, PO L M, REHMAN Y A U. Large separable kernel attention: rethinking the large kernel attention design in CNN[J]. Expert Systems with Applications, 2024, 236:121352.
[14] ZHOU Zhiguo, SUN Jiaen, YU Jiabao, et al. An image-based benchmark dataset and a novel object detector for water surface object detection[J]. Forntiers in Neurorobotics, 2021, 15:723336.
[15] HU Jie, SHEN Li, SUN Gang. Squeeze-and-excitation networks[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Piscataway: IEEE, 2018:7132-7141.
[16] WOO S H, PARK J, LEE J Y, et al. CBAM: convolutional block attenetion module[C] //Proceedings of the 15th European Conference on Computer Vision(ECCV). Munich: Springer, 2018:3-19.
[1] 孙迪,郭义童,任超,范海峰,张传雷. 基于多尺度特征融合与改进注意力的锈蚀螺栓螺帽检测[J]. 《山东大学学报(理学版)》, 2026, 61(1): 1-14.
[2] 陈俊芬,李娜娜,谢博鋆,张杰. 双注意力引导特征融合的半弱监督目标检测[J]. 《山东大学学报(理学版)》, 2025, 60(1): 1-13.
[3] 罗奇,苟刚. 基于聚类和群组归一化的多模态对话情绪识别[J]. 《山东大学学报(理学版)》, 2024, 59(7): 105-112.
[4] 廖祥文,徐阳,魏晶晶,杨定达,陈国龙. 基于双层堆叠分类模型的水军评论检测[J]. 《山东大学学报(理学版)》, 2019, 54(7): 57-67.
[5] 李润川,昝红英,申圣亚,毕银龙,张中军. 基于多特征融合的垃圾短信识别[J]. 山东大学学报（理学版）, 2017, 52(7): 73-79.
[6] 于然1,2,刘春阳3*,靳小龙1,王元卓1,程学旗1. 基于多视角特征融合的中文垃圾微博过滤[J]. J4, 2013, 48(11): 53-58.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
版权所有 © 2018 《山东大学学报(理学版)》编辑部　
地址: 山东省济南市山大南路27号山东大学中心校区（250100）　电话: +86-0531-88366917　E-mail: xblxb@sdu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发