Study On Ships Detection Experiments With Spaceborne SAR Imagery

At present, ships monitoring have great significance in the area of shipping management and military application. SAR has strong advantages in surface ship monitoring, and becomes an effective means of ships monitoring. In recent years, researches of ship detection and classification using SAR images have received considerable attention in the area of marine remote sensing.Nowadays many ship target detection and classification algorithms have been developed based on SAR images. Due to run short of in situ measurement data, it is very difficult to give those algorithms objective evaluation. For evaluating typically ship detection algorithms and developing the study on vessel types'classification, seven simultaneously experiments of ships detection with SAR image were performed in coastal of Qingdao from 2005.10 to 2007.12. By these experiments, typically ship detection algorithms are evaluated; the precise of the length of ship extracted by SAR images and modification model are analyzed and developed, and the feasibility of ship classification by the configurations features of ship is discussed.As for evaluating typically ship detection algorithms, in this paper CFAR detection algorithm and wavelet transform detection algorithm are analyzed and evaluated by the experiment-obtained data set. And the application scope and detection precisions of each ship target detection algorithm are presented. The wavelet transform algorithm can extract more ship targets which are embedded in the surrounding sea clutter than CFAR method, and wavelet transform method can't distinguish ship with the noises which have high coherence. But those noises can't affect CFAR detection algorithm, and CFAR detection method can reserve the ship structure, which are very important for future vessel classification.Due to the limitation of resolution of SAR images and the azimuth distorting generated by wave, there are great differences between the length of ship extracted by SAR images and the measured length in real. Thus, in this paper, the effect on the length of ship extracted by SAR images caused by the circumstance condition is analyzed firstly. Then polynomial method and neural network are used to modify the length of ship extracted by SAR images. The result shows that the neural network holds more capability to modify the length of ship than mini-square method.In the fourth chapter of this paper, vessel types'scattering characters are analyzed with spaceborne SAR images based on in situ measurement data. By analyzing the relationship between scatter-distributing feature and different vessel types, several conclusions can be obtained. Bulk ship has several discrete peak values, oil ship has only on peak value, and container ship has several continuous peak values. Then the peak features of ship targets are extracted by a method of target's peak feature extraction. Finally, the feasibility of using the peak-feature-distributing to recognize ships types is discussed. And with the application of high resolution SAR images, the peak-feature-distributing can be used to the classification of vessel type.