Lightweight facial recognition method for edge computing

Abstract

The invention discloses a lightweight language face identification method for edge computing. The method comprises the following steps: S1, constructing a lightweight facial recognition network modelAntCNN oriented to edge computing equipment, the network structure of the AntCNN comprising a first convolution layer, a first pooling layer, a first dense block, a second pooling layer, a second dense block, a third pooling layer, a third dense block and a third pooling layer; S2, capturing a face image, compressing the face image into small-size pixels to serve as input of an AntCNN, and utilizing the AntCNN to perform feature extraction and classification, S3, enabling the acquired multi-dimensional feature map to pass through a full connection layer to obtain specific scores of all classes, and enabling the maximum score to represent the specific classification of the picture. According to the method, a dlib library of traditional machine learning is used for searching a face part, themethod is successfully operated on the edge computing device of the Raspberry Pi, the video for searching the face is very smooth, and the real-time requirement is completely met.

Description

technical field [0001] The invention belongs to the technical field of deep learning, and in particular relates to a lightweight face recognition method oriented to edge computing. Background technique [0002] Deep learning has better robustness to the diversity changes of the target. Therefore, directly running deep learning network models on edge computing devices is regarded as the most promising method and has been widely researched and applied. However, deep learning is computationally intensive. However, the calculation amount and storage space of edge computing devices are limited. This means that while designing a deep learning network model for edge computing, it is necessary to consider the accuracy rate and also pay attention to the amount of calculations and parameters required by the network. In order to be able to run deep learning network models on edge computing devices, lightweight network models such as MobileNet and ShuffleNet have been proposed. Howe...

AI Technical Summary

Problems solved by technology

If these networks are applied for object classification, this will result in waste of resources and poor performance

Images