基于卷积神经网络的多噪声随机共振研究  

Multi-noise stochastic resonance study based on convolutional neural network

作　　者：刘赟 王友国[1] LIU Yun;WANG Youguo(College of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学理学院,南京210023

出　　处：《智能计算机与应用》2024年第10期201-205,共5页Intelligent Computer and Applications

基　　金：国家自然科学基金(62071248)。

摘　　要：针对噪声增强卷积神经网络缺少考虑噪声类型对模型性能影响的问题,提出了一种改进的噪声增强卷积神经网络算法,使用整流线性函数作为激活函数,向输出神经元分别加入均匀噪声、高斯噪声和拉普拉斯噪声,对比研究了不同噪声类型下卷积神经网络中存在的噪声增益。MNIST数据集上的仿真实验结果表明,改进的噪声增强卷积神经网络具有更高的分类准确率,同时发现拉普拉斯噪声为模型的最优噪声类型。Aiming at the problem that noise-enhanced convolutional neural networks do not consider the influence of noise types on model performance,an improved noise-enhancing convolutional neural network algorithm is proposed,which uses the rectified linear function as the activation function to add uniform noise,Gaussian noise and Laplace noise to the output neurons,and compares and studies the noise gain existing in convolutional neural networks under different noise types.The simulation results on the MNIST dataset show that the improved noise-enhancing convolutional neural network has higher classification accuracy,and it is demonstrated that Laplace noise is the optimal noise type of the model.

关 键 词：卷积神经网络 随机共振 加性噪声 交叉熵 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]