Telemetry Parameter Interpretation in Image Form Through Federated Learning

Abstract

The automatic interpretation of telemetry parameter during the flight can help monitor the status of the aircraft at any time. However, due to the lack of historical data, effective interpretation is very difficult. In this paper, we propose a method to interpretate telemetry parameter in image form through federated learning (FL). Firstly, to simulate the interpretation of human eye, the telemetry data is converted into an image form for feature extraction. Then, image-related dataset is utilized for model pre-training. Finally, FL is called to integrate data from multiple institutions and train together to obtain a higher-precision model. Experiments show that the method proposed in this paper can effectively improve the accuracy of interpretation and reduce the loss.