Abstract: Drones equipped with detectors can be used for applications such as air quality prediction. Federated learning empowers multi-drone systems to face challenges such as single point of failure, malicious attacks, and difficulty in achieving fair incentives. This article studies a reputation based blockchain enabled federated learning scheme for multi-drone systems. By automatically executing tasks through blockchain-based smart contracts, local model quality is evaluated based on reputation values, and malicious drones are identified and removed based on reputation thresholds. Sparse the global model based on reputation value to achieve fair distribution of model profits, and comprehensively consider reputation value and data volume to reward honest drones with corresponding costs. The simulation verified through the MNIST dataset that the algorithm proposed in this paper has higher accuracy than the FedAVG algorithm, and can identify and expel malicious drones under different proportions, achieving a fair distribution of model profits and total budget.