高级搜索

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

基于动态视觉运动特征的脉冲神经网络识别方法

董峻妃 姜润皓 燕锐 唐华锦

董峻妃, 姜润皓, 燕锐, 唐华锦. 基于动态视觉运动特征的脉冲神经网络识别方法[J]. 电子与信息学报, 2023, 45(8): 2731-2738. doi: 10.11999/JEIT221478
引用本文: 董峻妃, 姜润皓, 燕锐, 唐华锦. 基于动态视觉运动特征的脉冲神经网络识别方法[J]. 电子与信息学报, 2023, 45(8): 2731-2738. doi: 10.11999/JEIT221478
DONG Junfei, JIANG Runhao, YAN Rui, TANG Huajin. Spiking Neural Network Recognition Method Based on Dynamic Visual Motion Features[J]. Journal of Electronics & Information Technology, 2023, 45(8): 2731-2738. doi: 10.11999/JEIT221478
Citation: DONG Junfei, JIANG Runhao, YAN Rui, TANG Huajin. Spiking Neural Network Recognition Method Based on Dynamic Visual Motion Features[J]. Journal of Electronics & Information Technology, 2023, 45(8): 2731-2738. doi: 10.11999/JEIT221478

基于动态视觉运动特征的脉冲神经网络识别方法

doi: 10.11999/JEIT221478
基金项目: 国家重点研发计划 (2020AAA0105900)
详细信息
    作者简介:

    董峻妃:女,博士生,研究方向为类脑计算、脉冲神经网络以及神经形态视觉

    姜润皓:男,博士生,研究方向为类脑计算、脉冲神经网络

    燕锐:女,教授,研究方向为类脑计算、神经计算与智能机器人

    唐华锦:男,教授,研究方向为类脑计算、计算神经科学以及智能机器人

    通讯作者:

    唐华锦 htang@zju.edu.cn

  • 中图分类号: TP391.4

Spiking Neural Network Recognition Method Based on Dynamic Visual Motion Features

Funds: The National Key Research and Development Program of China (2020AAA0105900)
  • 摘要: 针对现有脉冲神经网络(SNN)对动态视觉事件流识别精度低与实时性差等问题,该文提出一种基于动态视觉运动特征的脉冲神经网络识别方法。首先利用基于事件的运动历史信息表示与梯度方向计算提取事件流中的动态运动特征;然后引入时空池化操作来消除事件在时间和空间上的冗余,保留显著的运动特征;最后,将特征事件流输入脉冲神经网络进行学习与识别。在基准的动态视觉数据集上的实验结果表明,动态视觉运动特征可显著提升SNN对于事件流的识别精度与计算速度。
  • 图  1  基于动态视觉运动特征的SNN识别方法流程图

    图  2  动态视觉特征表示方法对比图

    图  3  不同特征表示方法下的识别混淆矩阵结果比较

     算法1 动态视觉运动特征算法
     1: 输出: $ ({e_i},{\theta _{{e_i}}}) $
     2: 输入: $ r,\tau $,初始化$ {\bf{IS}}(x,y,p) \leftarrow 0 $
     3: for 每个到来的事件 ${{\boldsymbol{e}}_i}$ do
     4:  $ {\bf{IS}}({x_i},{y_i},{p_i}) \leftarrow i $ // 更新$ {\bf{IS}} $
     5:  计算十字邻域$U({{\boldsymbol{e}}_i},r)$,式(3)—式(6)
     6:  利用式(8)、式(9)计算$ \Delta X $和$ \Delta Y $
     7:  计算当前事件的运动梯度方向$ {\theta _{{e_i}}} $,式(7)—式(10)
     8:  $ ({x_i},{y_i}) \leftarrow ({x_i}/4,{y_i}/4) $ // 时空池化
     9:  if ${{\boldsymbol{e}}_i}$不在不应期内 then
     10:  输出$({{\boldsymbol{e}}_i},{\theta _{ {e_i} } })$
     11:  end if
     12: end for
    下载: 导出CSV

    表  1  数据集信息与划分情况

    数据集类别数分辨率训练集测试集
    DVS128 Gesture11128×1281151191
    Action Recognition10346×26025041
    DailyAction-DVS12128×1281235205
    下载: 导出CSV

    表  2  基于事件的动作识别算法比较(%)

    方法结构DVS128 GestureAction RecognitionDailyAction-DVS
    SLAYER[12]深度SNN (8层)93.6
    STBP[13]深度SNN (6层)93.4
    DECOLLE[14]深度SNN (6层)95.579.691.7
    SCRNN[21]脉冲卷积循环网络(5层)92.0
    文献[22]基于卷积和储层计算的SNN65.0
    文献[9]HFirst特征+单层SNN61.555.068.3
    SPA [18]HFirst特征+单层SNN70.176.9
    文献[15]光流特征+单层SNN92.778.190.3
    本文运动特征+单层SNN94.779.596.1
    注: 加粗字体表示各列最优结果;“–”表示此处数据为空
    下载: 导出CSV

    表  3  不同特征表示方法的性能比较

    特征表示方法识别精度(%)处理速度(kev/s)
    HFirst特征61.5032.9
    HOTS特征85.5654.7
    光流运动特征91.533.67
    运动特征(本文)94.65913
    注: 加粗字体表示各列最优结果
    下载: 导出CSV

    表  4  不同运动历史信息表示方法的性能比较(%)

    运动历史信息时间平面计数平面索引平面
    DVS128
    Gesture
    93.5891.4494.65
    Action Recognition76.9274.3679.49
    DailyAction -DVS94.6393.6696.10
    注: 加粗字体表示各列最优结果
    下载: 导出CSV
  • [1] GALLEGO G, DELBRÜCK T, ORCHARD G, et al. Event-based vision: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(1): 154–180. doi: 10.1109/TPAMI.2020.3008413
    [2] 张宇豪, 袁孟雯, 陆宇婧, 等. 面向动态事件流的神经网络转换方法[J]. 计算机应用, 2022, 42(10): 3033–3039. doi: 10.11772/j.issn.1001-9081.2021091607

    ZHANG Yuhao, YUAN Mengwen, LU Yujing, et al. Neural network conversion method for dynamic event stream[J]. Journal of Computer Applications, 2022, 42(10): 3033–3039. doi: 10.11772/j.issn.1001-9081.2021091607
    [3] 李家宁, 田永鸿. 神经形态视觉传感器的研究进展及应用综述[J]. 计算机学报, 2021, 44(6): 1258–1286.

    LI Jianing and TIAN Yonghong, Recent advances in neuromorphic vision sensors: A survey[J]. Chinese Journal of Computers, 2021, 44(6): 1258–1286.
    [4] ROY K, JAISWAL A, and PANDA P. Towards spike-based machine intelligence with neuromorphic computing[J]. Nature, 2019, 575(7784): 607–617. doi: 10.1038/s41586-019-1677-2
    [5] 张铁林, 徐波. 脉冲神经网络研究现状及展望[J]. 计算机学报, 2021, 44(9): 1767–1785.

    ZHANG Tielin and XU Bo. Research advances and perspectives on spiking neural networks[J]. Chinese Journal of Computers, 2021, 44(9): 1767–1785.
    [6] 胡一凡, 李国齐, 吴郁杰, 等. 脉冲神经网络研究进展综述[J]. 控制与决策, 2021, 36(1): 1–26. doi: 10.13195/j.kzyjc.2020.1006

    HU Yifan, LI Guoqi, WU Yujie, et al. Spiking neural networks: A survey on recent advances and new directions[J]. Control and Decision, 2021, 36(1): 1–26. doi: 10.13195/j.kzyjc.2020.1006
    [7] ORCHARD G, MEYER C, ETIENNE-CUMMINGS R, et al. HFirst: A temporal approach to object recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(10): 2028–2040. doi: 10.1109/TPAMI.2015.2392947
    [8] ZHAO Bo, DING Ruoxi, CHEN Shoushun, et al. Feedforward categorization on AER motion events using cortex-like features in a spiking neural network[J]. IEEE Transactions on Neural Networks and Learning Systems, 2015, 26(9): 1963–1978. doi: 10.1109/TNNLS.2014.2362542
    [9] XIAO Rong, TANG Huajin, MA Yuhao, et al. An event-driven categorization model for AER image sensors using multispike encoding and learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 31(9): 3649–3657. doi: 10.1109/TNNLS.2019.2945630
    [10] LIU Qianhui, PAN Gang, RUAN Haibo, et al. Unsupervised AER object recognition based on multiscale spatio-temporal features and spiking neurons[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 31(12): 5300–5311. doi: 10.1109/TNNLS.2020.2966058
    [11] LIU Qianhui, RUAN Haibo, XING Dong, et al. Effective AER object classification using segmented probability-maximization learning in spiking neural networks[C]. The 34th AAAI Conference on Artificial Intelligence, New York, USA, 2020: 1308–1315.
    [12] HE Weihua, WU Yujie, DENG Lei, et al. Comparing SNNs and RNNs on neuromorphic vision datasets: Similarities and differences[J]. Neural Networks, 2020, 132: 108–120. doi: 10.1016/j.neunet.2020.08.001
    [13] SHRESTHA S B and ORCHARD G. SLAYER: Spike layer error reassignment in time[C]. The 32nd International Conference on Neural Information Processing Systems, Montreal, Canada, 2018: 1419–1428.
    [14] KAISER J, MOSTAFA H, and NEFTCI E. Synaptic plasticity dynamics for deep continuous local learning (DECOLLE)[J]. Frontiers in Neuroscience, 2020, 14: 424. doi: 10.3389/fnins.2020.00424
    [15] LIU Qianhui, XING Dong, TANG Huajin, et al. Event-based action recognition using motion information and spiking neural networks[C]. The Thirtieth International Joint Conference on Artificial Intelligence, Montreal, Canada, 2021: 1743–1749.
    [16] AFSHAR S, HAMILTON T J, TAPSON J, et al. Investigation of event-based surfaces for high-speed detection, unsupervised feature extraction, and object recognition[J]. Frontiers in Neuroscience, 2019, 12: 1047. doi: 10.3389/fnins.2018.01047
    [17] MANDERSCHEID J, SIRONI A, BOURDIS N, et al. Speed invariant time surface for learning to detect corner points with event-based cameras[C]. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 10237–10246.
    [18] ZHAO Bo, YU Qiang, DING Ruoxi, et al. Event-driven simulation of the tempotron spiking neuron[C]. 2014 IEEE Biomedical Circuits and Systems Conference (BioCAS) Proceedings, Lausanne, Switzerland, 2014: 667–670.
    [19] AMIR A, TABA B, BERG D, et al. A low power, fully event-based gesture recognition system[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 7388–7397.
    [20] MIAO Shu, CHEN Guang, NING Xiangyu, et al. Neuromorphic vision datasets for pedestrian detection, action recognition, and fall detection[J]. Frontiers in Neurorobotics, 2019, 13: 38. doi: 10.3389/fnbot.2019.00038
    [21] XING Yannan, DI CATERINA G, and SORAGHAN J. A new spiking convolutional recurrent neural network (SCRNN) with applications to event-based hand gesture recognition[J]. Frontiers in Neuroscience, 2020, 14: 590164. doi: 10.3389/fnins.2020.590164
    [22] GEORGE A M, BANERJEE D, DEY S, et al. A reservoir-based convolutional spiking neural network for gesture recognition from DVS input[C]. 2020 International Joint Conference on Neural Networks (IJCNN), Glasgow, United Kingdom, 2020: 1–9.
    [23] LU Junwei, DONG Junfei, YAN Rui, et al. An event-based categorization model using spatio-temporal features in a spiking neural network[C]. 2020 12th International Conference on Advanced Computational Intelligence (ICACI), Dali, China, 2020: 385–390.
    [24] LAGORCE X, ORCHARD G, GALLUPPI F, et al. HOTS: A hierarchy of event-based time-surfaces for pattern recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(7): 1346–1359. doi: 10.1109/TPAMI.2016.2574707
  • 加载中
图(3) / 表(5)
计量
  • 文章访问数:  732
  • HTML全文浏览量:  314
  • PDF下载量:  189
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-11-25
  • 修回日期:  2023-05-02
  • 网络出版日期:  2023-05-19
  • 刊出日期:  2023-08-21

目录

    /

    返回文章
    返回