A Collision Prevention And Early Warning Method Based On Deep Learning With Multi-source Data Fusion

With the rapid development of economy along the Yangtze river,the waterway transport business in the basin has become more and more busy,and the number of waterway transport vessels going east and west has soared.Zhen-yang auto-ferry,as the highway vehicle ferry with the largest traffic flow along the Yangtze river,has received an unprecedented severe challenge to the driving safety of auto-ferry going from north to south.With such a large number of ships,the radar and the Marine automatic identification system(AIS)equipped by the ferry itself can no longer guarantee the safe navigation of the ferry.In this paper,a collision warning method based on deep learning and multi-source data fusion is proposed to improve the safety of steamboat by predicting the ship track of east-west heading.At present,there are still many shortcomings in ship track prediction.First,most of the existing prediction methods are based on the continuous data of the Marine automatic identification system(AIS)to analyze the ship trajectory.However,ionospheric delay and multipath interference exist in the GPS data of AIS system,which will lead to discrete,semantically missing or incomplete track data,thus affecting the accuracy of ship track prediction.Second,the existing prediction methods based on single target ship fail to consider the influence of the correlation among multiple targets on ship track prediction.Thirdly,the existing methods have not fully considered the influence of the weather and climate change of the channel and the traffic jam on the prediction of ship track.The main research contents of this paper include:(1)The field of multi-source data fusion is studied deeply.The multi-source data fusion in this paper refers to the fusion of Marine radar data and Marine automatic identification system(AIS)data.Through the data fusion of the two,the advantages of the two sides are complementary,and the prediction stability,accuracy and real-time will be greatly improved.(2)The fusion data are preprocessed to generate the navigation trajectory of the target ship,and then the track is converted into a discrete position sequence as input,and the spatial and temporal multidimensional characteristics of the ship are extracted by considering the changes in the actual environment.(3)Deep learning convolutional neural network and long and short-term memory(LSTM)network are studied deeply to construct collision warning method based on deep learning.By fully combining the advantages of convolutional neural network and deep bidirectional LSTM network,the local and global characteristic laws of ship's navigation trajectory are learned.(4)The method in this paper was tested and verified by using the AIS data of no.2036 Zhen-yang steam ferry and the vessel speed in the target area of 0 ? 20 kn.The experimental results show that the present method is better than the existing method.