Vision-assisted Beam Prediction for UAV-enabled Millimeter-Wave Communications using SE-ResNet50

The high overhead associated with the beam training process poses a significant challenge for highly mobile applications such as UAV communication. To mitigate this issue, this study proposes Squeeze-and-Excitation (SE) network that leverage visual information for accurate beam prediction in mmWave UAV communication. The SE network can selectively emphasize informative features through channel-wise feature recalibration, which enables the network to adapt to changing conditions, and optimize its predictions for different scenarios.