Marine Oil Spill And Small Target Detection Based On Radar

Marine oil spill and target detection has important application value in the Marine safety,maritime search and rescue,monitoring illegal behavior,etc.The use of modern monitoring technologies and methods can improve the response ability of maritime incident emergency and the level of management and decision-making.Radar echo signal contains the surface scattering echo information.It can be used to achieve detection and identification of the goal.This paper studies the methods of oil spill identification based on SAR images and small target detection based on sea clutter signal.How to distinguish oil spills or lookalike objects quickly and accurately is key.In general,detection of oil spill from SAR imagery includes three steps:image pre-processing,image segmentation and oil spill identification.Firstly in image segmentation stage,to further improve the accuracy and efficiency of SAR image segmentation in the detection of marine oil spill,ontologies are employed to analysis the dark region,then some lookalikes images are excluded in this paper.Once the needed dark formations have been selected,KFCM algorithm is used for classification.Secondly,in oil spill identification stage,In the dissertation,the selected dark formations are first decomposed into several bidimensional intrinsic mode functions using BEMD method,subsequently a new feature set is extracted using the Hilbert spectral analysis and relief algorithm.It is used to distinguish oil spills or lookalike objects.Experimental results demonstrate that the novel method has the obvious improvement in overall accuracy.Small target in sea clutter has the characteristics of lower signal-to-clutter ratio.Marine small target detection effectively is challenging and hot topic in the field of radar detection.Firstly it is proved that the sea clutter after the EEMD decomposition has fractal characteristics,and then this paper uses the multifractal generalized Hurst exponent as features to detect small target.Secondly it was found by analysis that the correlation coefficient has obvious differences in presence of target before and after the decomposition.So this paper uses the correlation coefficient as features to detect small target.The results show that the two methods not only effectively realize the target detection in the sea clutter but also performs much better than the similar methods.Finally,in view of the lack of sea clutter data,the difficulty of collected and marked problems and the waste of historical data,this paper uses TrAdaBoost transfer methodand the SVM classifier to achieve target detection by the data migration between different domains.The experiments demonstrate that the algorithm is effective in historical data and different target data migration.In short,the dissertation constructs the ocean surface phenomena ontology library,attempts to select new feature vectors for classifcation and expands the application scope of transfer learning.This work lays the foundation for the future.