Intelligent Clutter Suppression And Target Detection With Limited Sample Condition

With the increasing of maritime threats and challenges,the naval defense strategies of our country have changed from offshore defense to distant sea defense in recent years.Therefore,developing the detection and warning systems with high efficiency are one of the key development directions for naval defense.Thanks to the all day,all weather,and long distance,the radar techniques play an irreplaceable role in target detection and recognition.However,the performance of targets detection is weakened because the echoes are overwhelmed by the clutter.Consequently,the operation of clutter suppression is the prerequisite for efficient detection of the marine target.Due to the characteristics of sea clutter are related to the wavelength,the incident angle,and the polarization mode of radar,existing clutter suppression algorithms cannot obtain desired efficiency.With the development of artificial intelligence,by deeply exploiting the different features between the targets and background clutter,the performance of clutter suppression has remarkably improved.Nonetheless,the characteristic of sea clutter is highly relies on the parameters of radar platform and the sea domain,which limits the acquirement of sufficient clutter sample data and followed by the failure of deep learning training.Therefore,the approaches of clutter suppression based on artificial intelligence under small sample should be researched to improve the detection performance for marine targets.This thesis focuses on above problems and the main research contents and innovations are as follows:(1)Owing to the characteristics of sea clutter are highly rely on the parameters of radar platform and the environment of targets detection,which resulted in a limited sample data set,an novel approach based on deep convolutional generative adversarial network(DCGAN)is proposed to enhance the limited sea clutter data set.First,the clutter sample data set is generated by analyzing the statistical characteristics of the amplitude for sea clutter.Second,the sample data with high similarity are produced by utilizing the shallow and deep characteristics of the sea clutter based on the DCGAN.Finally,the sea clutter can be accurately classified based on the data set established above and followed by the obtained of the clutter samples with independent and identical distribution.In addition,the validity and accuracy of the model are verified by simulated and real-world measured data.(2)Aiming to the case of over-fitting and difficulty in accurately adjusting the parameters in clutter suppression based on deep learning network model under small samples,an adaptive clutter suppression and target detection method based on transfer learning is proposed in this work.First,the self-adaption deep transfer learning characteristics are presented by establishing the model of internal correlation representation for clutter data under different sea areas and the multi-characteristic joint matching method.Second,a transfer pre-training model data is applied to cope with the imitated characteristics of sea clutter using a single sea area,which helps to accurately classify the clutter data under small samples,and effectively improve the performance of clutter suppression and target detection.The method proposed in this work has the advantages of high classification precision and clutter suppression by exploiting the complementary characteristics of different sea clutter data sets.As a result,the validity and accuracy of the model are verified by simulated and real-world measured data.