Medical image classification method based on client clustering personalized federal learning

The invention discloses an image classification method based on client clustering personalized federated learning, and mainly solves the problems that the existing federated learning method is poor in local data privacy protection and high in communication traffic and calculation amount in an interactive learning process. According to the implementation scheme, a colon pathology data set is obtained from a public data set and classified; processing the data distribution degree according to Dirichlet distribution; selecting a Resnet18 classification network as a backbone network model of a server side and a client side; performing client group division by training the client model; the server side model is trained through adaptive weight aggregation; updating the client model by using the parameters of the trained server model; and classifying the local disease image data by using the updated client model. The method can guarantee the privacy security of the data, reduces the communication traffic and calculat