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

针对多尺度目标的轻量级红外目标检测算法

郑璐 彭月平 周彤彤

郑璐, 彭月平, 周彤彤. 针对多尺度目标的轻量级红外目标检测算法[J]. 红外技术, 2023, 45(5): 474-481.
引用本文: 郑璐, 彭月平, 周彤彤. 针对多尺度目标的轻量级红外目标检测算法[J]. 红外技术, 2023, 45(5): 474-481.
ZHENG Lu, PENG Yueping, ZHOU Tongtong. A Lightweight Infrared Target Detection Algorithm for Multi-scale Targets[J]. Infrared Technology , 2023, 45(5): 474-481.
Citation: ZHENG Lu, PENG Yueping, ZHOU Tongtong. A Lightweight Infrared Target Detection Algorithm for Multi-scale Targets[J]. Infrared Technology , 2023, 45(5): 474-481.

针对多尺度目标的轻量级红外目标检测算法

基金项目: 

装备综合研究项目 WJ20211A030131

科研单位自主选题研究项目 ZZKY20223105

详细信息
    作者简介:

    郑璐(1998-),女,汉族,浙江金华人,硕士研究生。研究方向:深度学习与目标检测。E-mail:1095496345@163.com

    通讯作者:

    彭月平(1974-),男,汉族,湖北人,工学博士后,教授。研究方向:战场环境建模与仿真。E-mail:1095496345@qq.com

  • 中图分类号: TP391.41

A Lightweight Infrared Target Detection Algorithm for Multi-scale Targets

  • 摘要: 针对现有基于深度学习的红外目标检测算法参数量大、复杂度较高、对多尺度目标检测性能较差等问题,提出了一种针对多尺度目标的轻量级红外目标检测算法。算法以YOLOv3为基础,采用MobileNet V2轻量级骨干网络、设计改进的简化空间金字塔结构(simSPP)、Anchor Free机制、解耦头和简化正负样本分配策略(SimOTA)分别对Backbone、Neck和Head进行优化,最终得到模型大小为6.25 M,浮点运算量2.14 GFLOPs的LMD-YOLOv3轻量级检测算法。在构建的MTS-UAV数据集上mAP达到90.5%,在RTX2080Ti显卡上FPS达到99,与YOLOv3相比mAP提升了2.60%,模型大小为YOLOv3的1/10。
  • 图  1  YOLOv3网络结构

    Figure  1.  YOLOv3 network structure

    图  2  LMD-YOLOv3网络结构

    Figure  2.  LMD-YOLOv3 network structure

    图  3  SPP模块结构

    Figure  3.  The structure of SPP

    图  4  MTS-UAV数据集部分图片

    Figure  4.  Part of data set MTS-UAV

    图  5  损失函数曲线图

    Figure  5.  Loss function graph

    图  6  P-R曲线图

    Figure  6.  P-R graph

    图  7  检测结果可视化

    Figure  7.  Visualization of test results

    表  1  SimSPP模块在YOLOv3算法上实验结果对比

    Table  1.   Comparison of experimental results of SimSPP module on YOLOv3 algorithm

    Model Recall mAP FPS FLOPs Params
    YOLOV3 90.70% 87.90% 74 12.41G 61.52M
    YOLOV3+
    SPP
    90.90% 88.60% 71 12.57G 64.15M
    YOLOV3+
    SPPA
    90.20% 88.00% 73 12.51G 63.10M
    YOLOV3+
    SPPB
    90.60% 88.50% 72 12.51G 63.10M
    YOLOV3+
    SPPC
    90.90% 88.30% 72 12.51G 63.10M
    下载: 导出CSV

    表  2  SimSPP模块在YOLOX-s算法上实验结果对比

    Table  2.   Comparison of experimental results of SimSPP module on YOLOX-s algorithm

    YOLOX-S YOLOX-SA YOLOX-SB YOLOX-SC
    mAP 80.90% 80.70% 80.90% 80.60%
    FPS 84 90 91 91
    下载: 导出CSV

    表  3  LMD-YOLOv3消融实验结果对比

    Table  3.   LMD-YOLOv3 comparison of ablation experiment results

    Recall mAP FPS FLOPs Params
    Experiment 1 90.70% 87.90% 74 12.41G 61.52M
    Experiment 2 90.60% 88.50% 72 12.51G 63.10M
    Experiment 3 91.20% 90.40% 74 11.19G 46.20M
    Experiment 4 91.20% 90.50% 99 2.14G 6.25M
    下载: 导出CSV

    表  4  横向实验结果对比

    Table  4.   Comparison of horizontal experimental results

    Recall/% mAP/% FPS FLOPs/G Params/M
    YOLOv3 90.70 87.90 74 12.41 61.52
    YOLOv4 85.57 86.99 70 11.30 63.9
    YOLOX-s 88.20 80.90 84 2.13 8.94
    Faster-RCNN 81.60 81.30 49 26.24 41.12
    LMD-YOLOv3 (Ours) 91.20 90.50 99 2.14 6.25
    下载: 导出CSV
  • [1] 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.
    [2] Girshick R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision, 2015: 1440-1448.
    [3] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. Advances in neural Information Processing Systems, 2015, 28: 91-99.
    [4] LIU W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]//European conference on computer vision., 2016: 21-37.
    [5] Bochkovskiy A, WANG C Y, LIAO H Y M. Yolov4: Optimal speed and accuracy of object detection[J/OL]. arXiv preprint arXiv, 2020, https://arxiv.org/abs/2004.10934
    [6] WANG C Y, Yeh I H, LIAO H Y M. You Only Learn One Representation: Unified Network for Multiple Tasks[J/OL]. arXiv pre-print arXiv, 2021, https://arxiv.org/abs/2105.04206.
    [7] GE Z, LIU S, WANG F, et al. Yolox: Exceeding yolo series in 2021[J/OL]. arXiv preprint arXiv, 2021, https://arxiv.org/abs/2107.08430.
    [8] LIU M, DU H, ZHAO Y, et al. Image small target detection based on deep learning with SNR controlled sample generation[M]//Current Trends in Computer Science and Mechanical Automation, 2018: 211-220.
    [9] LIN Liangkui, WANG Shaoyou, TANG Zhongxing. Using deep learning to detect small targets in infrared oversampling images[J]. Journal of Systems Engineering and Electronics, 2018, 29(5): 947-952. doi:  10.21629/JSEE.2018.05.07
    [10] ZHAO D, ZHOU H, RANG S, et al. An adaptation of CNN for small target detection in the infrared[C]//IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium, 2018: 669-672.
    [11] 谢江荣. 基于深度学习的空中红外目标检测关键技术研究[D]. 上海: 中国科学院大学(中国科学院上海技术物理研究所), 2019.

    XIE Jiangrong. Research on Key Technologies of Air Infrared Target Detection Based on Deep Learning[D] Shanghai: University of Chinese Academy of Sciences (Shanghai Institute of Technical Physics, Chinese Academy of Sciences), 2019.
    [12] FAN M, TIAN S, LIU K, et al. Infrared small target detection based on region proposal and CNN classifier[J]. Signal, Image and Video Processing, 2021, 15: 1927-1936. doi:  10.1007/s11760-021-01936-z
    [13] 张凯, 刘昊, 杨曦, 等. 基于关键点检测网络的空中红外目标要害部位识别算法[J]. 西北工业大学学报, 2020, 38(6): 1154-1162. doi:  10.3969/j.issn.1000-2758.2020.06.003

    ZHANG K, LIU H, YANG X, et al. Key position recognition algorithm of aerial infrared target based on key point detection net-work [J]. Journal of Northwest University of Technology, 2020, 38(6): 1154-1162 doi:  10.3969/j.issn.1000-2758.2020.06.003
    [14] Redmon J, Farhadi A. Yolov3: An incremental improvement[J/OL]. arXiv preprint arXiv, 2018, https://arxiv.org/abs/1804.02767.
    [15] Howard A, Zhmoginov A, CHEN L C, et al. Inverted residuals and linear bottlenecks: Mobile networks for classification, detection and segmentation[J/OL]. Computer Science, 2018, https://arxiv.org/abs/1801.04381v2.
    [16] Howard A G, ZHU M, CHEN B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J/OL]. arXiv preprint arXiv, 2017, https://arxiv.org/abs/1704.04861.
    [17] HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9): 1904-1916. doi:  10.1109/TPAMI.2015.2389824
  • 加载中
图(7) / 表(4)
计量
  • 文章访问数:  132
  • HTML全文浏览量:  37
  • PDF下载量:  36
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-06-05
  • 修回日期:  2022-06-23
  • 刊出日期:  2023-05-20

目录

    /

    返回文章
    返回