基于卷积神经网络特征图聚类的人脸表情识别

针对卷积层存在的特征冗余问题,提出了一种基于卷积神经网络的特征图聚类方法。首先通过预训练网络参数提取网络最后一层卷积层的特征图,然后对特征图进行聚类操作,取聚类中心构成新的特征图集合,以聚类后的特征图集作为数据集训练分类器。将有监督的深度学习方法与传统的机器学习方法相结合,使用特征图聚类进行特征去冗余让网络学习到更有效的特征。去冗余后的特征使用神经网络分类器在fer2013测试集上达到了71.67%准确率,在CK+测试集上达到86.98%准确率,证明了该人脸表情识别方法的有效性。

Facial Expression Recognition Based on Convolutional Neural Network Feature Map Clustering

To solve the problem of feature redundancy in convolution layer,a feature graph clustering method based on convolution neural network is proposed.Firstly,the feature map of the last convolution layer of the network is extracted by pre-training network parameters,and then the feature map is clustered.The clustering center is taken to form a new set of feature graphs,and the clustered feature graph set is used as the training classifier of the data set.This paper combines supervised in-depth learning method with traditional machine learning method,and uses feature graph clustering to remove redundancy so that the network can learn more effective features.After redundancy removal,the neural network classifier achieves 71.67%accuracy on fer2013 test set and 86.98%accuracy on CK+test set,which proves the validity of the facial expression recognition method.