Facial Age Recognition Technology Based On Convolution Neural Network

With the improvement of technology,people put forward higher expectation for the security and ease of use of information systems.Biometric features have been widely used because of their good characteristics.As a biological feature,human face has achieved high accuracy in the field of identification,and a series of identification applications have been derived.In addition to identity information,face has other rich information connotation.Recognition based on age information of face image has broad application prospects in image retrieval,security monitoring and business intelligence,and has become a hot research field.This paper mainly studies the related technologies in face age recognition.In order to overcome the differences caused by uncertain position,illumination environment and different posture angles in the process of sample acquisition,it is necessary to pre-process the collected images,including human detection,image enhancement and structure normalization.In order to meet the needs of model training in machine learning,it is also necessary to expand the sample by random transformation to achieve compatibility processing.Convolutional neural network replaces the full-connected structure of traditional multi-layer perceptron by convolution operation,which reduces the parameters and has better local feature invariance,and has good feature extraction performance.It is more suitable for such applications as face age recognition.In the engineering field,convolutional neural network has many classical network structures and development platform tools to choose from.ChaLearn dataset provides age information which is measured by manual estimation in the open data set of face recognition application.Compared with other datasets,it is more consistent with the law of human age recognition.For embedded applications,a lightweight convolution network structure is proposed for the construction of face age recognition system.Through comparative experiments,the influence of training sample size,network complexity and other factors on the recognition accuracy is evaluated systematically,and the actual effect of BatchNorm and Dropout optimization technology is verified.A model is trained based on the optimal parameter combination obtained from the experiment,the CSj=5 index on the ChaLearn test set reaches 0.8178.This model is used to realize the automatic age recognition program of photographic faces.It is proved that the recognition effect is consistent with human perception and has certain practical value.