Akademik Veri Yönetim Sistemi
2017 International Conference on Radar Systems, Radar 2017, Belfast, Birleşik Krallık, 23 - 26 Ekim 2017, cilt.2017
© 2017 Institution of Engineering and Technology. All rights reserved.Many shore-based surveillance radars use preloaded maps to discriminate land and sea radar echoes, in order to accomplish critical surveillance or to aid navigational safety. In the case of absence or inaccuracy of maps, the classification is usually done by a human operator from the radar video and errors such as false alarms from land reflections or missed detections from sea targets are possible. In this paper, a reflectivity, amplitude statistics and temporal correlation based land-sea classification algorithm which uses only high resolution X-band radar data with no Doppler information and no a priori knowledge of land or sea presence is proposed. First, a neural networks approach is used to pose the problem as a pattern recognition problem and to find the forms of the decision boundaries and validate the feature selection. Secondly, a practical and fast algorithm is proposed to accomplish the task real time and to enable the user to adjust the algorithm parameters for changing environments. The algorithm computes both reflectivity, amplitude statistics and temporal correlation (or persistence value) of the radar range-azimuth bins and classifies the bin as land or sea. The application of the algorithm on real data shows that a high land/sea classification success rate is possible using the proposed algorithm.