Research Of Face Gender Classification Base On An Ensemble Of Convolutional Neural Networks

In recent years,the breakthrough in the study of deep learning raises a new round of upsurge in the field of artificial intelligence research.At the same time,intelligent hardware equipment replacement speed is also very fast.The extraction and excavation of face information has been a hot topic in both scholars and the Internet companies.The recognition of face related information has been applied in a variety of real scenarios,such as video surveillance,social networks,etc.In this paper,we mainly study face gender recognition and realize the video surveillance system about face gender recognition.For face gender recognition,this paper designs selective ensemble algorithm and convolution neural network,the algorithm in basic classifier is designed in such a way is selected benchmark database and let it to be mixed with other databases,and then the multiple classifiers are obtained after different convolution neural network model.After that,use selective ensemble algorithm and chose the basic classifier according to the sort of forecast accuracy.This paper uses Caffe deep learning framework,and Feret dataset and Adience dataset are used to train out different basic classifiers.And ensemble method is better than the basic classifier has highest accuracy.Comparison was made with selective ensemble method,all integrated method and integration method based on single CNNs.Experiment results show that in three methods,the effect of selective ensemble method is best and these methods are higher than basic classifier has highest accuracy.This paper constructs three kinds of different convolution neural network called G_CNNS₄,G_CNNS₅,G_CNNS₆ network,and using network to train on the Feret dataset and Adience dataset.The experimental results show that the effect of G_CNNS₅ and G_CNNS₆ network of training in above data set is the best to be used for the integration.This paper also analyzes the influencing factors of CNNs include convolutional layers,dropout parameter and classifier.Finally,a video monitoring system was realized for face gender recognition,it has done well in the real scene.