| During sea target detection, the threat level of floating targets (iceberg, buoy and floating garbage) without artificial power is relatively low, while the potential threat of artificial power targets (ship and periscope) is high. Therefore, the track identification of floating targets can assist situational awareness. Modeling the movement characteristics and collecting the real track data of floating targets are difficult, thus the development of floating targets track identification is limited. In this paper, the movement characteristics such as velocity, acceleration, heading direction, track curvature and turning rate are used to describe the differences between floating target track and artificial power target track. Then, multi-layer convolution neural network (CNN) is designed for floating target track identification. To train CNN model and test algorithms, the AIS data are used to extract ship tracks, and the sea surface current and wind map measured by highfrequency radar are used to simulate the floating target track. Data analysis shows that the identification rate for floating target track is 98.31%, and the false alarm rate is 0.17%. |
