The Research On Target Detection With Compact HF Surface Wave Radar Based On AIS Information

Since Crombie found HF radars can receive echoes backscattered by ocean current in 1955,they have been studied for more than half a century and now widely used in measuring ocean surface dynamic parameters.The HF radars can operate in difficult weathers,cover a huge area,and have a low cost.Therefore,they are important for fishery monitoring,ocean state forecasting,etc.Researchers also made great efforts for enabling HF radars to detect vessels at sea.However,the ability of target detection is still not practical,especially for compact radar system employing crossed-loop/monople receive antenna.Although the cost is reduced and the compact radar system show good performance on ocean state forecasting,many difficulties still exist on target detection,including:1.The interference and noise signal is very high in HF band.The vessel signals are easy to be contaminated by radio frequency interference,ionospheric interference,transient interference,etc.2.The point target hypothesis is sometimes invalid so the targets may not be detected by simple cell detection method.3.The range resolution of HF radars is several kilometers in general which is far longer than the vessels' size at sea.4.The antenna pattern of compact HF radar is easy to be disturbed by surrounding environment,therefore,brings difficulties for direction of arrival(DOA)estimation.5.The vessels information in radar coverage area cannot be collected completely,so the algorithms are hard to evaluate.Considering the above difficulties,this thesis studied on target detection with com-pact HF surface wave radar.On one hand,the traditional methods are fully inspected and some key factors which limit the target detection ability are found.One the other hand,the machine learning methods are introduced into this field with the help of AIS information.The results show many improvement on target detection and parameter estimation.This thesis includes the following:1.A new SVC-based target detection algorithm is proposed.This algorithm uses targets matched with AIS information as labeled samples,the signal to noise ratio in different dimensions are selected as sample features,and the random noise signals are used as negative samples.The field experimental results are used to compare the performance of the SVC detection algorithm and traditional CFAR detection algorithm.In the occasion of same number of targets detected by two algorithms,the SVC method obtains more matches with AIS information and more tracks on pepper plot,which demonstrates the new SVC method is better than the CFAR method.Besides,the SVC methods has good expansibility and generalization ability.2.An AIS-based crossed-loop/monopole antenna pattern estimation method is pro-posed.This method using plenty of targets from all directions which are matched with AIS as calibration sources.The antenna pattern is obtained by detection,match,classification,screening,interpolation and smooth.The field experimental data from three experiments,Shandong in 2013,Pingtan in 2014,Zhangzhou in 2015,are used to verified the new methods.The results show antenna pattern is easy to be distorted in field environment.In this situation,the DOA estimation errors are quite high using ideal antenna pattern.However,the estimated antenna pattern show good performance on DOA estimation.The shipborne antenna pat-tern measurement was conducted in Liuao,2015.The AIS antenna pattern shows high relevance with shipborne antenna pattern so the new method is proved effec-tive.3.A SVR-based target DOA estimation method is proposed.In order to settle some problems on DOA estimation by measured antenna pattern,the machine learning method is introduced to overcome the measurement errors and inherent flaws in previous algorithm.The labeled samples are targets matched with AIS information.The normalized amplitude and phase values in every channels are selected to be the sample features.The field experimental results demonstrate the new SVR method can obtain high DOA estimation accuracy than the MUSIC algorithm using measured antenna pattern.Besides,the SVR method can relieve the angle ambiguity,has good expansibility and small data volume requirement.4.A SVR-based target range estimation method is proposed.Consider the good per-formance on DOA estimation by using the SVR method,it can also be used on range estimation.The labeled samples are targets matched with AIS information.The normalized amplitude values of range cells before and after target cell are selected to be the sample features.The range estimation model is extracted by the SVR algorithm.The field experimental results show the SVR range estima-tion method can not only obtain more accurate and stable range estimation than traditional empirical formula methods,but also compensate the systematic range deviation.The SVR method shows good generalization ability when applied range estimation model to data collected from other radar sites.