Efficient Task Scheduling for Federated Learning-Driven Image Recognition in Cloud-Edge Collaborative Environments

Cloud-edge collaboration and artificial intelligence integration have advanced federated learning image recognition model training. However, in cloud-edge collaborative federated learning, imbalanced sample data often leads to low global model aggregation accuracy. To address this, a cloud-edge collaborative federated learning model aggregation method for sample imbalance is proposed, which considers local model accuracy, stability, and sample size as evaluation weights to address the issue of uneven sample distribution affecting global model aggregation. By increasing the contribution weight of high-quality local models, the impact of sample imbalance and low-quality models on the global model is reduced. This method is implemented through plugins and compared with FedAvg model aggregation method in multiple imbalanced sample scenarios. The experimental results demonstrate that this method mitigates the impact of sample diversity on global model aggregation and improves accuracy in imbalanced-sampling scenarios.