基于深度学习的遮挡人脸识别方法研究OA
Research on Occluded Face Recognition Methods Based on Deep Learning
针对佩戴口罩导致人脸关键特征大面积缺失,传统人脸识别系统的性能大幅降低的问题,提出一种结合Inception-ResNet 结构与 CBAM 注意力机制的遮挡人脸识别方法.首先,利用 Inception 模块进行多感受野特征提取,并依靠ResNet残差机制保证深层网络的稳定;接着,通过CBAM注意力机制对特征进行通道与空间维度的提纯;最后,依靠 ArcFace 损失函数优化特征向量与分类权重的角度关系,让模型学习到更具区分性的人脸特征.实验结果表明,在口罩遮挡场景下,该方法具有较为优越的识别精度与遮挡鲁棒性.
To address the issue that wearing masks results in the substantial loss of key facial features,significantly degrading the performance of traditional face recognition systems,this study proposes an occluded face recognition method that combines the Inception-ResNet architecture with the CBAM attention mechanism.First,the Inception module is employed to extract features across multiple receptive fields,while the residual mechanism of ResNet ensures the stability of the deep network.Next,the CBAM attention mechanism refines the features along both channel and spatial dimensions.Finally,the ArcFace loss function is used to optimize the angular relationship between feature vectors and classification weights,enabling the model to learn more discriminative facial features.Experimental results demonstrate that the proposed method achieves superior recognition accuracy and occlusion robustness in masked face scenarios.
衣姿好;张烜赐;李涵若水;王雨霏;张豪
中国矿业大学(北京),北京 100083中国矿业大学(北京),北京 100083中国矿业大学(北京),北京 100083中国矿业大学(北京),北京 100083中国矿业大学(北京),北京 100083
信息技术与安全科学
遮挡人脸识别深度学习Inception-ResNet注意力机制ArcFace
occluded face recognitionDeep Learninginception-ResNetattentionArcFace
《现代信息科技》 2026 (17)
26-30,5
评论