Improved You Only Look Once Model for UAVs/Ships Relative Attitude Detection

The variability and complexity of maritime weather significantly affect unmanned aerial vehicle (UAV) landings, potentially destabilizing their attitude due to factors like wind and waves. Hence, precise attitude detection during autonomous UAV landings is crucial for enhancing landing safety and reliability. Addressing these challenges, this study proposes a deep learning-based method for relative UAV attitude detection. The algorithm integrates the Content-Aware ReAssembly of Features (CARAFE) upsampling operator to widen the receptive field and improve deep feature acquisition. Additionally, it introduces the Coordinate Attention (CA) mechanism to enhance the network’s feature extraction, thereby boosting UAV attitude detection accuracy and efficiency. Using a personally collected dataset of landing markers comprising 5145 images, experiments demonstrate an average precision of 94.0%, marking a 9.64% improvement over the baseline model, while maintaining a rapid detection speed of 42.92 frames per second.