Exploration of marine ship anomaly real-time monitoring system based on deep learning

Abstract

Maritime transportation is the main traffic mode of globalization, accounting for about 90% of the proportion of global trade. Maritime safety is an important problem of marine transportation. Therefore, it is very important to set up the ship abnormal real-time monitoring system. In order to realize the real-time monitoring of abnormal data in the course of ship driving, a hybrid single classification framework based on depth learning is designed. DDBN-OCSVM framework uses the deep network to solve the problem that complex high-dimensional data are difficult to reduce and learn. The single classification algorithm is used to avoid the impact of data imbalance on the results of ship real-time monitoring anomaly detection. Finally, the experimental analysis and discussion of the data are carried out. The experimental results show that DDBN-OCSVM can effectively reduce the detection error under the accelerating effect of GPU and cuDNN. The DDBN-OCSVM algorithm proves that the unsupervised feature learning and hierarchical representation are effective and feasible. It is also proved that it is feasible to apply this deep learning mode to real-time monitoring of ship anomalies.