Face forgery detection model optimization method, device and system based on federated learning

The invention provides a federated learning-based face forgery detection model optimization method, device and system, and relates to the technical field of image processing, and the method comprises the steps: updating a last updated basic parameter set according to a basic parameter set after last iteration training in a plurality of clients, and obtaining a basic parameter set updated at the current time and sending the basic parameter set to the client, so that the client optimizes a face forgery detection model on a face forgery training data set in a local database according to the basic parameter set updated at the current time and the personalized layer parameter set after iteration training at the last time. Taking parameters of a basic network layer of the optimized face counterfeiting detection model as a basic parameter set after current iteration training; and iteratively executing the training process, and if a preset termination condition is satisfied, obtaining an optimal face counterfeiting d