Robust maritime search and rescue with UAVs on the edge

Maritime search and rescue (SAR) missions are of myriad importance to save lives at sea. The traditional SAR methods at sea such as visual search from aircraft or boats, using helicopters, leveraging radar and sonar technologies are not usually accurate and efficient enough due to being time-consuming, expensive and struggling with small targets. Deep-learning based techniques on the other hand have improved SAR operations in terms of detection, prediction and response capabilities. UAVs (unmanned aerial vehicles) are transforming maritime SAR by increasing response speed, accuracy, and safety for both rescuers and survivors. However, the obtained results still needs improvement in terms of accuracy and speed. Hence, this paper proposes a novel CNN-based model using UAV imagery to further improve maritime search and rescue operations success rates. Using the AFO (Aerial Data Set of Floating Objects) dataset, the proposed solution achieves an accuracy of 95.3% with an inference time of 33.3 FPS on the Jetson AGX Orin platform at 15W. The results indicate the potential of a cost-effective UAV SAR asset that could be used in real settings to increase the survival and success rates of search and rescue operations.