ResNet Based Parallel Intelligent System Design for Marine Search and Rescue

Abstract

Among numerous emergency responses to maritime emergencies, the most critical part is the search and rescue (SAR) for victims in distress on the sea. Maritime search and rescue missions are usually divided into two: search and rescue, and the most complex and difficult part is searching for targets in distress at sea. Traditional numerical models can provide accurate drift trajectories of targets in distress. However, owing to the mathematical mechanism, even with high-performance computing technology, the searching time is still far from meeting the real-time requirements of sea rescue missions. In view of this situation, in this paper, a parallel intelligent system is proposed for SAR targets in distress at sea based on a deep residual neural network. A large amount of natural system observation data is utilized to train the proposed ResNet network with the advantages of high-performance computing technique which can accelerate the training process. Furthermore, mathematical model simulation results are used to tune the parameters and optimize the results of the neural network. The precision, recall, and F1 scores of the predictions for drifting of the targets in distress can reach more than 90%. Moreover, the ResNet network-based parallel intelligent system combined with high-performance parallel computing technology reduces the time by about 1%. When compared with traditional numerical model systems in predicting target drifting results, the proposed system can enhance the timeliness of the maritime search and rescue process.