Facial Expression Recognition Based on Improved GoogLeNet Convolution Neural Network
In view of the shortcomings of the current use of CNN for facial expression recognition,this paper proposes a new facial expression recognition method based on improved GoogLeNet.This method reduces the dimension of convolution kernels in different layers,extracts facial features,and simplifies the Inception module that deepens the learning depth by improving the GoogLeNet network,which optimizes the network structure,reduces the amount of parameters,and further improves the operation efficiency.Finally,more comprehensive information about facial expression features is obtained,and the method proposed in this paper is verified in three data sets:JAFFE,CK+and FER2013.The experimental results show that the accuracy of the improved scheme reaches 97.53%on the premise of meeting the requirements of timeliness,the improved method have good generalization and robustness.
万方数据
维普
