Discrete-Time-Scheduling Spiking Neural Network for Image Self-Classification

doi:10.7652/xjtuxb202103007

您当前的位置：

首页 >

文章列表页 >

Discrete-Time-Scheduling Spiking Neural Network for Image Self-Classification

更新时间：2025-07-09

- Discrete-Time-Scheduling Spiking Neural Network for Image Self-Classification
- Vol. 55, Issue 3, Pages: 57-64(2021)
- 作者机构：
  
  西安交通大学微电子学院,西安,710049
- 作者简介：
- 基金信息：
- DOI：10.7652/xjtuxb202103007
  CLC： TN402
- Online First：10 March 2021，
  
  Published：2021
- 稿件说明：
移动端阅览
Discrete-Time-Scheduling Spiking Neural Network for Image Self-Classification[J]. 2021, 55(3): 57-64.
DOI：

Discrete-Time-Scheduling Spiking Neural Network for Image Self-Classification[J]. 2021, 55(3): 57-64. DOI： 10.7652/xjtuxb202103007.

摘要

针对现有脉冲神经网络(SNN)图像分类模型中存在的资源占用高和运算较复杂等实际约束问题

为寻求更加轻量高效的机器视觉解决方案

提出了一种新型的基于离散时间调度的SNN图像自分类模型。通过高斯差分归一化和首脉冲时间编码完成了从灰度图像到脉冲序列的转换; 结合经典的脉冲时间依赖可塑性算法与强化学习的奖惩机制

实现了网络自分类; 通过引入竞争性机制和双重约束条件

保证了脉冲传递的稀疏性和学习特征的特异性

有效抑制了过拟合的出现。在Face/Moto数据集上的实验结果表明:相较于传统SNN分类模型

权重更新算法复杂度由O(n

)降低为O(1)

脉冲编码模式简化了近90%

网络训练参数减少了60%以上; 模型开始迭代6次后权重近乎收敛

分类准确率由40%迅速上升至90%

在迭代20次后分类性能趋于稳定

最终准确率达到了93.4%; 当训练样本比例减少至原先的40%后

模型的分类准确率仍能稳定保持在80%左右。所提模型可为高效率低功耗的小型智能化硬件终端的边缘计算方案实现提供参考。

Abstract

Aiming at the actual constraints of existing spiking neural networks(SNN)for image classification such as high resource occupancy and complex operation

a new kind of discrete-time-scheduling based SNN model for image self-classification is proposed to find a kind of machine vision solution with weight reduction and high energy efficiency. The transformation from the gray image to pulse sequences is implemented with normalized difference of Gaussians and time-to-first-spike coding

and the self-classified network is realized via the combination of classical spike-time-dependent-plasticity algorithm and mechanism of reinforcement learning with feedbacks of reward and p

enalty. Simultaneously

the competitive mechanism and dual constraints are also introduced to ensure the sparseness of spike transmission and the specificity of learning features

which effectively inhibits the occurrence of overfitting. Compared with the traditional SNN classification models

the experimental results on the Face/Moto dataset show that the complexity of the weight updating algorithm is reduced form O(n

) to O(1) while the coding patterns of spike are simplified by nearly 90% and the trainable parameters of the network are decreased by over 60%; the weights nearly converge after 6 iterations

and the classification accuracy quickly rises from 40% to 90% and becomes stable after 20 iterations

finally reaches 93.4%; the accuracy can hold steady at 80% while the proportion of training samples drops to 40% of the original. The proposed model is beneficial for edge computing realization of high-efficiency and low-power minimalistic intelligent hardware terminals.

关键词

Keywords

references

徐学良. 人工神经网络的发展及现状 [J]. 微电子学, 2017, 47(2): 239-242.

XU Xueliang. The development and status of artificial neural network [J]. Microelectronics, 2017, 47(2): 239-242.

程龙, 刘洋. 脉冲神经网络: 模型、学习算法与应用 [J]. 控制与决策, 2018, 33(5): 923-937.

CHENG Long, LIU Yang. Spiking neural networks: model, learning algorithms and applications [J]. Control and Decision, 2018, 33(5): 923-937.

陶建华, 陈云霁. 类脑计算芯片与类脑智能机器人发展现状与思考 [J]. 中国科学院院刊, 2016, 31(7): 803-811.

TAO Jianhua, CHEN Yunji. Current status and consideration on brain-like computing chip and brain-like intelligent robot [J]. Bulletin of Chinese Academy of Sciences, 2016, 31(7): 803-811.

苏亚丽, 吴健行, 惠维, 等. 一种用于视觉颜色特征分类的脉冲神经网络 [J]. 西安交通大学学报, 2019, 53(10): 115-121.

SU Yali, WU Jianxing, HUI Wei, et al. A spiking neural network for classification of visual color features [J]. Journal of Xi'an Jiaotong University, 2019, 53(10): 115-121.

LI Xiumin, WANG Wei, XUE Fangzheng, et al. Computational modeling of spiking neural network with learning rules from STDP and intrinsic plasticity [J]. Physica: A Statistical Mechanics and Its Applications, 2018, 491: 716-728.

赵梓铭, 刘芳, 蔡志平, 等. 边缘计算: 平台、应用与挑战 [J]. 计算机研究与发展, 2018, 55(2): 327-337.

ZHAO Ziming, LIU Fang, CAI Zhiping, et al. Edge computing: platforms, applications and challenges [J]. Journal of Computer Research and Development, 2018, 55(2): 327-337.

DAVIES M, SRINIVASA N, LIN T H, et al. Loihi: a neuromorphic manycore processor with on-chip learning [J]. IEEE Micro, 2018, 38(1): 82-99.

柯成仁. 基于STDP的脉冲神经网络图像识别算法研究 [D]. 西安: 西安电子科技大学, 2018: 1-55.

LAMMIE C, HAMILTON T J, VAN SCHAIK A, et al. Efficient FPGA implementations of pair and triplet-based STDP for neuromorphic architectures [J]. IEEE Transactions on Circuits and Systems: I Regular Papers, 2019, 66(4): 1558-1570.

AMIRSHAHI A, HASHEMI M. ECG classification algorithm based on STDP and R-STDP neural networks for real-time monitoring on ultra low-power personal wearable devices [J]. IEEE Transactions on Biomedical Circuits and Systems, 2019, 13(6): 1483-1493.

LI F, FERGUS R, PERONA P. Learning generative visual models from few training examples: an incremental Bayesian approach tested on 101 object categories [J]. Computer Vision and Image Understanding, 2007, 106(1): 59-70.

KHERADPISHEH S R, GANJTABESH M, THORPE S J, et al. STDP-based spiking deep convolutional neural networks for object recognition [J]. Neural Networks, 2018, 99: 56-67.

LEE C, SRINIVASAN G, PANDA P, et al. Deep spiking convolutional neural network trained with unsupervised spike-timing-dependent plasticity [J]. IEEE Transactions on Cognitive and Developmental Systems, 2019, 11(3): 384-394.

LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition [J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.

ZHENG N, MAZUMDER P. Online supervised learning for hardware-based multilayer spiking neural networks through the modulation of weight-dependent spike-timing-dependent plasticity [J]. IEEE Transactions on Neural Networks and Learning Systems, 2018, 29(9): 4287-4302.

MOZAFARI M, KHERADPISHEH S R, MASQUELIER T, et al. First-spike-based visual categorization using reward-modulated STDP [J]. IEEE Transactions on Neural Networks and Learning Systems, 2018, 29(12): 6178-6190.

LECUN Y. Deep learning hardware: past, present, and future [C]∥Proceedings of the 2019 IEEE International Solid-State Circuits Conference(ISSCC). Piscataway, NJ, USA: IEEE, 2019: 12-19.

YOUSEFZADEH A, SERRANO-GOTARREDONA T, LINARES-BARRANCO B. Fast pipeline 128×128 pixel spiking convolution core for event-driven vision processing in FPGAs [C]∥Proceedings of the 2015 International Conference on Event-Based Control, Communication, and Signal Processing(EBCCSP). Piscataway, NJ, USA: IEEE, 2015: 1-8.

蔺想红, 张田文. 指数突触电导IF神经元模型及事件驱动模拟策略 [J]. 电子学报, 2008, 36(8): 1495-1501.

LIN Xianghong, ZHANG Tianwen. An integrate-and-fire neuron model with exponential synaptic conductances for event-driven simulation strategy [J]. Acta Electronica Sinica, 2008, 36(8): 1495-1501.

THORPE S, FIZE D, MARLOT C. Speed of processing in the human visual system [J]. Nature, 1996, 381(6582): 520.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Control Strategy for Battery Thermal Management System Using Deep Reinforcement Learning

Hierarchical Twin-Delayed Policy Gradient Reinforcement Learning for Intelligent Cooperative Control of Aircraft

Battery Thermal Management System Control Strategy Based on Deep Reinforcement Learning

Function Mapping and Scheduling Method of Virtual Network Combining Genetic Algorithm and Reinforcement Learning

Reinforcement Learning-Based Energy Management Strategy Considering Driving Style for Hybrid Electric Vehicle

Related Author

XIE Yi

HUANG Kunjie

ZOU Zihao

HUANG Chi

ZHAO Dongpeng

XI Chunfu

LIN Xixiang

MA Yu

Related Institution

School of Mechanical and Transportation Engineering, Chongqing University

School of Energy and Power Engineering, Chongqing University

Key Laboratory of Low-Grade Energy Utilization Technologies and Systems of Ministry of Education, Chongqing University

School of Mechanical Engineering, Tianjin University

China Automotive Engineering Research Institute Co., Ltd.

⁰