Multi-Unmanned Aerial Vehicle Agricultural Inspection System based on Federated Learning

Unmanned Aerial Vehicle (UAV) inspection systems have been widely used in precision agriculture and environmental monitoring in recent years. UAV can achieve coverage of a wider geographical area and meet the demands of large-scale monitoring and long-distance data communication. The use of deep learning (DL) technology to model and predict inspection data has made remarkable progress in meteorological monitoring, agricultural management, disease detection and other aspects. Current research attempts to improve the prediction performance of agricultural inspection prediction model based on deep learning by increasing the number of UAV devices to realize the collection of massive data. At present, the construction of multi-UAV agricultural inspection system often involves the use of UAV services from multiple manufacturers. However, the use of multiple UAVs as data sources may bring security issues of privacy for all parties involved. To solve this issue, this paper proposes a federated learning (FL) scheme based on Deep Bidirectional Long Short-Term Memory (Deep BiLSTM) algorithm to train the UAV agricultural inspection prediction model. Our proposed UAV agricultural inspection prediction model based on federated learning can improve the generalization ability and prediction performance of the model while protecting data privacy. In order to demonstrate the effectiveness of the proposed federated learning agricultural inspection prediction model based on Deep BiLSTM, we conduct data collection and performance evaluation of the prediction model on the built UAV agricultural inspection prototype system. Comprehensive experimental results demonstrate that our proposed prediction model for UAV agricultural inspection based on federated learning achieves superior prediction performance while protecting data privacy, obtaining 0.132 RMSE, 0.018 MSE, 0.107 MAE and 0.996 R2 - SCORE, respectively.