UAVs optimize data training and sensing for faster edge vision models
Joint Trajectory, Sensing, and Computation Co-Design for UAV-Assisted Federated Edge Vision
Networking and Internet Architecture
Summary
Training complex object detection models usually needs lots of data and computing power. This paper looks at using drones (UAVs) to collect visual data and handle computation on the spot. The authors link how drones move, what they sense, and how they compute together to reduce training time while respecting battery life. They also balance the data each drone handles to keep the model training efficient and accurate. Their tests show this combined approach works better than other methods.
What this means in practice
- •For edge computing engineers: Optimize drone movement and data processing to speed up training of vision models near the data source under energy limits.
- •For drone fleet operators: Coordinate drone sensing and computation schedules to improve object detection training efficiency while managing battery use.
Authors
Kaichen Liu, Yueling Che, Sheng Luo, Jianhua Tang
Abstract
This paper investigates UAV-assisted federated edge vision for training computationally heavy object detection models. Existing studies treat training data as static, ignoring the tight coupling between dynamic visual sensing and massive in place computational burdens. To minimize the expected training time under battery constraints, we propose a four-phase joint trajectory, sensing, and computation co-design. To mitigate data heterogeneity, we use a uniform sensing probability to translate federated convergence requirements into UAV mobility constraints. Furthermore, a two-layer alternating optimization (AO) algorithm is developed to adaptively determine the local data workload, optimally balancing the required training rounds against in-place computation delays. Simulations validate the superiority of our approach in model convergence and accuracy over baselines.