Research And Implementation Of Face Gender Recognition Based On Deep Learning

In recent years,with the improvement of the computational performance and the number of samples,deep learning has been able to give play to its advantages and has made outstanding achievements in the field of image processing.The research on face detection,feature extraction and face recognition has been relatively mature,but the research on face gender recognition is still relatively blank.Gender is one of the basic biological characteristics of human beings.This paper presents a method of face gender recognition based on Caffe deep learning framework.The experiment was divided into two groups of data sets,the first set contained 200000 images of upper body face,of which 50000 were for male and female training/testing,the second group of 200000 face data sets was obtained from the first group,and the same was done for each group of 50000 faces.Three deep learning models,including LeNet,AlexNet,VGG-16,are recognized in Linux system.Finally,the three depth models are compared and analyzed in terms of accuracy rate,error rate,training time,convergence speed,model file size generation and image recognition time.Experimental results show that the upper face data sets of the three deep learning models are significantly superior to the face data sets in terms of accuracy and error rate.Training time,single frame image recognition time and model size are related to the model itself,irrelevant to the data set.The accuracy rate of VGG-16 model based on upper body face data set was increased by 5.3%compared with the LeNet model,and the error rate was reduced by about 12%.Compared with VGG-16 model,the average image recognition speed is 7~8 times,the size of the model file generated by training is only 1/205,and the training time is saved by about 1/4.and the AlexNet model falls somewhere in between.Finally,taking the VGG-16 model as an example,experimental results were demonstrated through video file.