摘要
本文提出一种新的用于对卷积神经网络提取的特征进行分类的分类器即径向基函数神经网络(rbfnn)分类器。其思想是利用卷积神经网络作为特征提取器,使用rbfnn对提取的特征进行分类。同时在训练时采取softmax分类器与rbfnn分类器同步训练的方式,其中rbfnn分类器将MSE(均方误差)损失作为监督信息,softmax分类器用交叉熵损失作为监督信息。优化后的模型优于[1]中的72.9%的准确率。
A new classifier, the radial basis function neural network(rbfnn) classifier, for classifying the features extracted by convolutional neural networks is proposed. The idea is to use a convolutional neural network as a feature extractor to classify the extracted features using rbfnn. At the same time, the training method of softmax classifier and rbfnn classifier is used in training. The rbfnn classifier uses the MSE(mean square error) loss as the supervision information, and the softmax classifier uses the cross entropy loss as the supervision information. The optimized model is better than the 72.9% accuracy rate in [1].
作者
张凯凯
郭松林
毕晨琳
Zhang Kaikai;Guo Songlin;Bi Chenlin(School of Electrical&Control Engineering,Heilongjiang University of Science&Technology,Harbin Heilongjiang,150022;The First Senior Middle School in Xincai,Zhumadian Henan,463500)
出处
《电子测试》
2019年第22期66-67,76,共3页
Electronic Test
