A Face Recognition Method Based on Deep Separable Convolution Model

Abstract

The present invention relates to a face recognition method based on a deep separable convolution model, comprising the following steps: the first step is to read a face image sample data set; the second step is to establish a deep separable convolution model. The above-mentioned deep separable convolution model cascades multiple residual bottleneck modules between two adjacent convolution modules; the third step is to use the gradient descent algorithm to update the deep separable convolution model parameters; the fourth step , Perform face recognition through the deep separable convolution model after updating the parameters. The present invention can improve the recognition speed on the basis of ensuring the accuracy of face recognition. Experiments show that the present invention can achieve a recognition speed of less than 300ms on the ARMv8 mobile terminal under the premise of ensuring that the recognition accuracy is higher than 99%. Therefore, the mobile terminal device can have an accurate and fast face recognition function.

Description

technical field [0001] The invention relates to a face recognition method based on a deep separable convolution model, belonging to the technical field of face recognition. Background technique [0002] In recent years, the demand for face recognition in daily life has been increasing. Some fields, such as face access control, face attendance, face ticket purchase, and face chasing, all have great room for development. These fields have high requirements for the speed and accuracy of face recognition. The convolutional neural network based on deep learning is the cornerstone of face recognition technology. The network continuously reduces the difference between the model output value and the real value through gradient descent feedback to approach the real result. [0003] At present, face recognition technology mainly focuses on the improvement of accuracy. However, in order to make face recognition technology more popular and user-friendly in daily life, the speed of reco...

AI Technical Summary

Problems solved by technology

Existing deep neural convolutional network models such as ResNet-50 have a prediction accuracy of over 97%, but there is still a lot of room for improvement before the model is put into actual products.

Some improved network models, such as Facenet, etc., have improved the accuracy of their models by using residual units, but they still face problems such as excessive memory usage and low calculation speed.

Images