基于改进MobileNetV2算法与频谱图的静默活体检测研究

打开文本图片集

【Abstract】Inorderto improve the accuracyof livebody detection，a silent livebody detection method basedonthe improved MobileNetV2 algorithmand spectrogram is proposed.This methoduses deplearning and image processng technologyto determinethefacebyanalyzing itsbiometricand spectral features.Inthisarticle，the MobileNetV2algorithm is enhanced by incorporating the Squeeze and Excitation （SE）attention mechanism and the Fourier spectrum-asisted learning method to improve the accuracy of silent living body detection.In addition，the MobileNetV2 + FFT + SE method is validated through experimental evaluationsonthePrint-Atack，OULU，and NUAA datasets.The experimentalresultsshowthatwhen RFP=10-2， RTP is 97.23% ，REE is 2.07% ，AUC is 96.27% ，and ACC is 92.31% .These metrics demonstrate that this method has excellent classificationperformance，high AUC value andlow errorrate.Atthesame time，this methodcan mprove the securityoffacerecognition，protectpersonalprivacyand informationecurityinpracticalapplications，andhas mpor tant Application prospects.

[Key Words】silent living detection; MobileNetV2; inverted residual structure; Fourier spectrum assisted learning

[中图分类号］TP391 [文献标识码］A [文章编号]1674-3229（2024）04-0101-09

0 引言

发展，个人身份认证和安全性问题日益突出[]。（剩余11391字）