Research on Ship target recognition based on attention mechanism

Abstract: Marine ship target recognition can effectively identify the categories of sailing ships and realize effective management of ships. It is strategically important for both civil and military domains, but it is highly demanding in terms of accuracy. In this paper, a novel neural network ByCTE(Bayesian Classification Transformer-Encoder) is proposed to realize ship target recognition by using track information. First, the raw data is preprocessed to make the processed data more favorable for model learning. Secondly, four BayesianLinear Encoder(BLE) modules are used to learn the complex relationship between different spatial positions of the sequence, so as to capture the long-term dependence relationship between the input sequences, and further extract the deep features of the sequence. Finally, complete the recognition by attention layer and softmax function. We select the best performing model in the training and use open dataset Automatic Identification System (AIS) data from Europe for training and validating the validity of the proposed model. ByCTE can achieve better accuracy by comparison with other methods.