Research on Non-IID Cervical Cell Image Classification Algorithm Based on Federated Learning

Aiming at the performance degradation of existing federated learning algorithms in non-IID image classification tasks, a method of improving SRGAN and dynamically balancing the distribution of client data sets is proposed to enhance the data of non-IID images, and for depth The learning method ignores the privacy and security issues of cervical cell classification, and proposes a method combining federated learning, attention mechanism and convolutional neural network to improve the performance of non-IID cervical cell classification. The experimental results show that the proposed algorithm is 1.77% better than the original convolutional neural network on the liquid-based Pap smear cervical image dataset that is artificially divided into non-independent and identical distribution, and the correct rate can reach 91.75^%,which better improves the performance of the non IID distributed images on the federal learning algorithm.