Research on Facial Recognition Based on Federated Learning and Improved IS-ResNet18
In facial recognition scenarios,there is a risk of privacy leakage between edge based facial data col-lection and cloud based data processing.In order to ensure accurate and efficient facial recognition,a facial recogni-tion method based on federated learning and improved IS-ResNet18 was proposed.This method trained the model through a federated learning framework without the need to obtain edge face data,optimizing the ResNet18 model,replacing ReLU activation function with Leaky ReLU activation function to reduce neuronal death.In the new model,Inception module was added with optimization of attention mechanism SE module.The results showed that the model's attention to important features was enhanced,and the model's expression ability and performance were improved,alleviating gradient vanishing and exploding problems,and enhancing model stability.It is proven that this method could not only protect user privacy but also maintain high recognition accuracy,demonstrating good fea-sibility and practicality.
NETL
NSTL
万方数据
