On Efficient Federated Learning for Aerial Remote Sensing Image Classification: A Filter Pruning Approach

To promote the application of federated learning in resource-constraint unmanned aerial vehicle swarm, we propose a novel efficient federated learning framework CALIM-FL, short for Cross-All-Layers Importance Measure pruning-based Federated Learning. In CALIM-FL, an efficient one-shot filter pruning mechanism is intertwined with the standard FL procedure. The model size is adapted during FL to reduce both communication and computation overhead at the cost of a slight accuracy loss. The novelties of this work come from the following two aspects: 1) a more accurate importance measure on filters from the perspective of the whole neural networks; and 2) a communication-efficient one-shot pruning mechanism without data transmission from the devices. Comprehensive experiment results show that CALIM-FL is effective in a variety of scenarios, with a resource overhead saving of 88.4% at the cost of \(1\%\) accuracy loss.