基于卷积神经网络的人脸性别识别研究与应用

Human faces contain important information such as sex, race, color, age, etc. If human faces can be identified by computer, a great deal of human biological characteristics can be obtained conveniently and contactlessly. Convolutional neural network is a kind of artificial neural network whose structure has three characteristics of local connection, weights sharing and pooling sampling. Such a network structure not only reduces the number of weights but also reduces the complexity of the network model, and zooming, rotation, translation, and other deformations have better robustness at the same time. In this paper, the principle of the classical Alexnet model and its performance in gender identification are studied first, then the Alexnet activation function and the full connectivity layer are improved according to the experimental results. Making the network can accept input pictures of any size and improve the recognition accuracy.