Research On Face Recognition Methods Based On Deep Learning

Face recognition is a kind of biometric recognition technology which bases on the collected facial features of people.It belongs to an important research direction of artificial intelligence and computer vision.At present,traditional face recognition algorithms need to go through a complicated calculation process,and the accuracy of face recognition finally achieved is low.In order to improve the performance of face recognition methods,this article researches and analyzes the problem from the perspective of convolutional neural networks.The specific research content is as follows:First,from the perspective of lightweight networks,this paper proposes a lightweight network face recognition method which bases on deep separable convolution and attention mechanisms.This method designs an inverse residual block that combines deep separable convolution and attention mechanisms to reduce the amount of network parameters while enhancing the ability to represent features.The attention mechanism can enable the network to autonomously learn the mapping from the input face images to high-quality feature representations,enabling the network to focus on learning the areas with a larger amount of information in the image.Lightweight residual module fused with attention mechanism to achieve the effect of taking into account accuracy,network parameters and model practicality.Secondly,in order to further improve performance,this paper proposes an improved face recognition method which bases on aggregative residual network.This method integrates a multi-branch topology network structure implemented by grouping convolution in the residual block to enhance it without increasing complexity.In order to express the features,the whole network is constructed by stacking residual blocks.In addition,a convolution operation is added to the skip connection in the residual block to reduce the feature dimension and increase the non-linearity in order to better extract the feature information of the image.Finally,in order to substantially improve the performance of face recognition algorithms based on deep learning,this paper proposes a face recognition method based on discriminative radial feature distribution.This method improves the shortcomings of the existing main loss functions of face recognition,enhances the separability of the embedded representations between different classes in the high-dimensional feature space from the perspective of radial features,and guides the high-dimensional network extracted by the loss function Features are distributed in a double-ring shape.This loss function can not only effectively improve the performance of face recognition tasks,but also improve the accuracy of general classification problems.In addition,the design of the loss function is not limited to a certain network,but to convolutional neural networks universally.