时频域协同增强的轻量化低光照交通标志检测
DOI:
CSTR:
作者:
作者单位:

1.太原师范学院计算机科学与技术学院晋中030619;2.智能优化计算与区块链技术山西省重点实验室晋中030619; 3.西北师范大学物理与电子工程学院兰州730070;4.荆楚理工学院人工智能学院荆门448000

作者简介:

通讯作者:

中图分类号:

TP391.4;TN0

基金项目:

国家自然科学基金(62066041)、山西省重点研发计划(202102010101008)、湖北省自然科学基金(2024AFB458)、山西省科技战略研究专项重点项目(202304031401011)、山西省基础研究计划(自由探索类)项目(202403021222276,202503021212249)、山西省高等学校科技创新项目(2024L295)资助


Lightweight low-light traffic sign detection algorithm based on time-frequency domain collaborative enhancement
Author:
Affiliation:

1.School of Computer Science and Technology, Taiyuan Normal University, Jinzhong 030619, China; 2.Shanxi Provincial Key Laboratory of Intelligent Optimization Computing and Blockchain Technology, Jinzhong 030619, China; 3.School of Physics and Electronic Engineering, Northwest Normal University, Lanzhou 730070, China; 4.School of Artificial Intelligence, Jingchu University of Technology, Jingmen 448000, China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对低光照场景中因图像亮度不足、噪声干扰以及目标纹理特征模糊而导致的交通标志检测精度下降等问题,提出一种时频域协同增强的轻量化低光照交通标志检测算法。首先,构建图像增强模块(image enhancement module,IEM),结合时域边缘增强与频域细节补偿,有效提升图像质量与结构清晰度;其次,设计动态特征融合模块(C3K2 dynamic fusion block,C3K2-DFB),通过级联多尺度卷积与动态特征融合机制,增强模型对不同尺度目标的感知能力;此外,优化Neck网络结构,构建轻量化特征融合路径,提升语义信息的传递效率;最后,使用小波池化(wavelet pooling,WaveletPool)替代传统下采样操作,保留高频纹理特征并增强对小目标和边缘细节的敏感性。实验结果表明,改进后的网络模型的平均精度均值为77%,在同一数据集下相较基准模型提升了1.6%,浮点运算数(floating-point operations per second,FLOPs)比原始算法减少了25%,参数量减少了47%。上述结果验证了本文模型在实际应用中的有效性。

    Abstract:

    To address the reduced accuracy of traffic sign detection in low-light scenes due to insufficient image brightness, noise interference, and blurred target texture features, a lightweight low-light traffic sign detection algorithm with temporal and frequency domain co-enhancement is proposed. First, an image enhancement module (IEM) is constructed, combining temporal edge enhancement and frequency domain detail compensation to effectively improve image quality and structural clarity. Second, a dynamic fusion block (C3K2-DFB) is designed, enhancing the model’s ability to perceive targets at different scales through cascaded multi-scale convolutions and a dynamic feature fusion mechanism. Furthermore, the Neck network structure is optimized to construct a lightweight feature fusion path, improving the efficiency of semantic information transmission. Finally, wavelet pooling (WaveletPool) is used to replace traditional downsampling operations, preserving high-frequency texture features and enhancing sensitivity to small targets and edge details. Experimental results show that the improved network model has an average accuracy of 77%, which is 1.6% higher than the baseline model on the same dataset. The number of floating-point operations per second (FLOPs) is reduced by 25% and the number of parameters is reduced by 47% compared to the original algorithm. These results verify the effectiveness of the model in practical applications.

    参考文献
    相似文献
    引证文献
引用本文

张红瑞,崔硕,杨潞霞,马永杰,陈雪.时频域协同增强的轻量化低光照交通标志检测[J].电子测量与仪器学报,2026,40(9):151-162

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-11-13
  • 出版日期:
文章二维码
×
《电子测量与仪器学报》
关于防范虚假编辑部邮件的郑重公告