留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

融合注意力机制的多尺度红外目标检测

李向荣 孙立辉

李向荣, 孙立辉. 融合注意力机制的多尺度红外目标检测[J]. 红外技术, 2023, 45(7): 746-754.
引用本文: 李向荣, 孙立辉. 融合注意力机制的多尺度红外目标检测[J]. 红外技术, 2023, 45(7): 746-754.
LI Xiangrong, SUN Lihui. Multiscale Infrared Target Detection Based on Attention Mechanism[J]. Infrared Technology , 2023, 45(7): 746-754.
Citation: LI Xiangrong, SUN Lihui. Multiscale Infrared Target Detection Based on Attention Mechanism[J]. Infrared Technology , 2023, 45(7): 746-754.

融合注意力机制的多尺度红外目标检测

基金项目: 

河北省重点研发计划项目 20350801D

详细信息
    作者简介:

    李向荣(1998-),女,硕士研究生,研究方向:图像处理、目标检测。E-mail: 243404315@qq.com

    通讯作者:

    孙立辉(1970-),男,博士,教授,研究领域:图像处理、数据分析。E-mail: Sun_lh@163.com

  • 中图分类号: TN215

Multiscale Infrared Target Detection Based on Attention Mechanism

  • 摘要: 针对红外图像存在细节纹理特征差、对比度低、目标检测效果差等问题,基于YOLOv4(You Only Look Once version 4)架构提出了一种融合通道注意力机制的多尺度红外目标检测模型。该模型首先通过降低主干特征提取网络深度,减少了模型参数。其次,为补充浅层高分辨率特征信息,重新构建多尺度特征融合模块,提高了特征信息利用率。最后在多尺度加强特征图输出前,融入通道注意力机制,进一步提高红外特征提取能力,降低噪声干扰。实验结果表明,本文算法模型大小仅为YOLOv4的28.87%,对红外目标的检测精度得到了明显提升。
  • 图  1  SE-YOLOv4网络结构

    Figure  1.  SE-YOlOv4 network structure

    图  2  主干特征提取网络结构

    Figure  2.  Backbone feature extraction network structure

    图  3  优化后的多尺度融合策略

    Figure  3.  Optimized multi-scale fusion strategy

    图  4  SENet网络结构

    Figure  4.  SENet network structure

    图  5  红外数据集样本及人工标注示例

    Figure  5.  Infrared dataset samples and manual labeling examples

    图  6  YOLOv4和SE-YOLOv4检测精度对比

    Figure  6.  Comparison of detection accuracy between Yolov4 and SE-Yolov4

    图  7  SE-YOLOv4与YOLOv4检测效果对比

    Figure  7.  Comparison of detection results between SE-YOLOv4 and YOLOv4

    图  8  YOLOv4和SE-YOLOv4在FLIR数据集上的检测效果对比

    Figure  8.  Comparison of detection effects of Yolov4 and SE-Yolov4 on FLIR dataset

    表  1  本文红外数据集先验框尺寸表

    Table  1.   Anchor box sizes of infrared dataset

    Output feature layer size Anchor1 Anchor2 Anchor3
    104×104 (5, 19) (8, 28) (12, 39)
    52×52 (12, 15) (17, 61) (30, 25)
    26×26 (33, 113) (59, 45) (95, 111)
    下载: 导出CSV

    表  2  相关目标检测模型实验结果

    Table  2.   Experimental results of relevant target detection model

    Model AP(%) mAP/
    (%)
    car person bicycle
    EfficientDet 77 75 62 71.08
    Faster-RCNN 79 65 59 67.53
    SSD 79 67 74 73.54
    YOLOv3 90 84 67 80.43
    YOLOv5 92 90 73 85.00
    SE-YOLOv4 93 90 81 87.85
    下载: 导出CSV

    表  3  消融实验

    Table  3.   Ablation experiment

    Experiments AP/(%) mAP/
    (%)
    Total params Model size/
    MB
    Car Person Bicycle
    YOLOv4 89 85 70 81.28 64, 040, 001 244.29
    Improving effective feature layers 91 87 80 85.95 36, 067, 401 137.59
    CSPDarknet44 91 87 80 85.95 18, 488, 393 70.52
    SE 90 87 72 83.32 64, 041, 801 244.30
    Ours 93 90 81 87.85 18, 490, 193 70.53
    下载: 导出CSV

    表  4  相关目标检测模型实验结果

    Table  4.   Experimental results of relevant target detection models

    Model AP/(%) mAP/(%)
    Car Person Bicycle
    YOLOv3 78 62 42 60.67
    YOLOv4 85 81 59 75.00
    YOLOv5 87 81 63 77.00
    SE-YOLOv4 85 82 79 82.08
    下载: 导出CSV
  • [1] 史泽林, 冯斌, 冯萍. 基于波前编码的无热化红外成像技术综述(特邀)[J]. 红外与激光工程, 2022, 51(1): 32-42. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ202201003.htm

    SHI Zelin, FENG Bin, FENG Ping. An overview of non thermal infrared imaging technology based on wavefront coding (invited) [J]. Infrared and Laser Engineering, 2022, 51(1): 32-42. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ202201003.htm
    [2] CHEN C, LI H, WEI Y, et al. A local contrast method for small infrared target detection[J]. IEEE Transactions on Geoscience & Remote Sensing, 2013, 52(1): 574-581.
    [3] LIU R, LU Y, GONG C, et al. Infrared point target detection with improved template matching[J]. Infrared Physics & Technology, 2012, 55(4): 380-387.
    [4] Teutsch M, Muller T, Huber M, et al. Low resolution person detection with a moving thermal infrared camera by hot spot classification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, Columbus, 2014: 209­216.
    [5] HAO Q, ZHANG L, WU X, et al. Multiscale object detection in infrared streetscape images based on deep learning and instance level data augmentation[J]. Applied Sciences, 2019, 9(3): 565. doi:  10.3390/app9030565
    [6] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.
    [7] GU Jiaojiao, LI Bingzhen, LIU Ke, et al Infrared ship target detection algorithm based on improved Faster R-CNN[J]. Infrared Technology, 2021, 43(2): 170-178. https://www.cnki.com.cn/Article/CJFDTOTAL-BJLG202307012.htm
    [8] REN S, HE K, 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, 2016, 39(6): 1137-1149.
    [9] 刘智嘉, 汪璇, 赵金博, 等. 基于YOLO算法的红外图像目标检测的改进方法[J]. 激光与红外, 2020, 50(12): 1512-1520. https://www.cnki.com.cn/Article/CJFDTOTAL-JGHW202012015.htm

    LIU Zhijia, WANG Xuan, ZHAO Jinbo, et al. An improved method of infrared image target detection based on YOLO algorithm[J]. Laser and Infrared, 2020, 50(12): 1512-1520. https://www.cnki.com.cn/Article/CJFDTOTAL-JGHW202012015.htm
    [10] Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv: 1804.02767, 2018.
    [11] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Pro-ceedings of the IEEE Conference On Computer Vision and Pattern Recognition, 2018: 7132-7141.
    [12] Bochkovskiy A, Wang C Y, LIAO H Y M. Yolov4: Optimal speed and accuracy of object detection[J/OL]. arXiv preprint arXiv: 2004.10934, 2020.
    [13] LIN T Y, Dollar P, Girshick R, et al. Feature pyramid networks for object detection[C]//Computer Vision and Pattern Recognition(CVPR), 2017: 2117-2125.
    [14] LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Computer Vision and Pattern Recognition(CVPR), 2018: 8759-8768.
    [15] LUO Y, CAO X, ZHANG J, et al. CE-FPN: enhancing channel information for object detection[J/OL]. arXiv preprint arXiv: 2103. 10643, 2021.
    [16] 谢俊章, 彭辉, 唐健峰, 等. 改进YOLOv4的密集遥感目标检测[J]. 计算机工程与应用, 2021, 57(22): 247-256. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG202122029.htm

    XIE Junzhang, PENG Hui, TANG Jianfeng, et al. Improved dense remote sensing target detection of YOLOv4[J]. Computer Engineering and Application, 2021, 57(22): 247-256. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG202122029.htm
    [17] 鞠默然, 罗江宁, 王仲博, 等. 融合注意力机制的多尺度目标检测算法[J]. 光学学报, 2020, 40(13): 132-140. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB202013016.htm

    JU Muran, LUO Jiangning, WANG Zhongbo, et al. Multi scale target detection algorithm integrating attention mechanism[J]. Journal of Optics, 2020, 40(13): 132-140. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB202013016.htm
    [18] TAN M, PANG R, LE Q V. Efficient det: Scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 10781-10790.
    [19] LIU W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision, 2016: 21-37.
    [20] 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, 2016: 779-788.
    [21] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 7263-727.
  • 加载中
图(8) / 表(4)
计量
  • 文章访问数:  121
  • HTML全文浏览量:  30
  • PDF下载量:  35
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-04-10
  • 修回日期:  2022-07-20
  • 刊出日期:  2023-07-20

目录

    /

    返回文章
    返回