模拟神经反馈机制和工作记忆的图像分类网络模型
Image Classification Network Model of Simulated Neural Feedback Mechanism and Working Memory
童顺延 1刘海华1
作者信息
- 1. 中南民族大学 生物医学工程学院,湖北 武汉 430074
- 折叠
摘要
该项目在卷积神经网络(CNN)中引入神经反馈机制和工作记忆机制,提出层内深度反馈卷积神经网络模型(IDFNet).该网络以神经反馈机制构建深度反馈计算模块(DFS),并在模块中引入了工作记忆(WM),通过深度变化控制WM空间内容的更新,从而增强了信息获取能力.最后,利用DFS替代CNN的卷积层构建IDFNet网络.通过在Flower102 和CIFAR-10、CIFAR-100 数据集上的实验表明,相较于同类网络,在更少参数量和计算量的情况下,IDFNet仍取得显著的性能提升,其识别率分别达到了96.61%和95.87%、79.99%.
Abstract
This project introduces neural feedback and working memory mechanisms into Convolutional Neural Networks(CNN)and proposes an Intra-Layer Deep Feedback Convolutional Neural Network Model(IDFNet).The network constructs a Deep Feedback Structure(DFS)using neural feedback mechanisms,and introduces a Working Memory(WM)within this module.It controls the update of WM spatial content by depth variations,so as to enhance information retrieval capabilities.Finally,the IDFNet network is built by replacing CNN's convolutional layers with DFS.Experimental results on the Flower102,CIFAR-10,and CIFAR-100 datasets demonstrate that IDFNet achieves significant performance improvements compared to similar networks,with higher recognition rates of 96.61%,95.87%,and 79.99%,respectively,while requiring fewer parameters and computations.
关键词
反馈机制/工作记忆/循环计算/图像分类Key words
feedback mechanism/working memory/loop computation/image classification引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据