基于深度可分离卷积的轻量化人脸识别方法设计

Design of Lightweight Face Recognition Method Based on Depth Separable Convolution

针对传统人脸识别方法识别精度较低的问题,提出基于深度可分离卷积的轻量化人脸识别方法。构建深度可分离卷积的轻量级卷积神经网络模型,采集人脸图像并进行预处理,从而增强数据集,采用多任务卷积神经网络提取人脸特征,完成人脸识别方法的设计。实验结果表明,该方法优于其他方法,人脸识别的准确率保持在90%以上,识别精度较高。

Aiming at the low recognition accuracy of traditional face recognition methods, a lightweight face recognition method based on depth separable convolution is proposed. Build a lightweight convolution neural network model with depth separable convolution, collect and preprocess face images, enhance data sets, extract face features using multitask convolution neural network, and complete the design of face recognition methods. The experimental results show that this method is superior to other methods, and the accuracy of face recognition remains above 90%, with high recognition accuracy.