A novel FaceNet method combined with wavelet transform and a multi-stage loss function for identity recognition

Key Points

• To develop a novel FaceNet method using wavelet transform features and multi-stage loss functions for better identity recognition performance.
• Implemented WT-FaceNet-MSL combining wavelet transform and multi-stage loss functions.
• Used the VGGFace2 dataset for training and validation.
• Monitored convergence speed and inference time during model evaluation.
• Achieved 97.00% validation accuracy after 18 epochs with the proposed method.
• Demonstrated faster convergence compared to traditional methods.
• Maintained short inference time during recognition tasks.