Automatic Segmentation Of SAR Ice Water In Wide Observation Band With Incident Angle Perception

To obtain more land information in a wider range in one flight,many SAR(Synthetic Aperture Radar)systems have the ability to work on the ScanSAR(Scanning Synthetic Aperture Radar)mode.In ScanSAR mode,the scanning width in the range direction is more than 100 kilometers.The variation range of the side-looking angle(incidence angle is between 20°and 40°.This mode is widely used in automated sea ice charting.But the incidence angle effects have heavily influence on the ScanSAR sea ice image.In order to reduce the influence of incidence angle effects,correcting the incidence effects is necessary.Conventional correction methods(such as LLM,locally linear mapping)typically assume that each target class has a similarity distribution in the middle of the image.The objectives of this study are to extend the correction algorithm to full swath width without any assumptions.For a target class only distributed on one or both sides of the image,interpolation or extrapolation of the confidence interval is realized using the linear regression technique based on the exponential model.The position of the reference band is then determined and the correction is performed.The main idea of improved LLM methods using linear regression is “correct by classes”,which needs pre-classification/pre-segmentation.Considering incidence angle effect,at first this paper build the logarithmic linear model to adapt it.Subsequently,by dividing blocks according to histogram distribution,the effect of incident angle on each block is reduced.The blocks are pre-classified by the Gaussian Mixture Model(GMM).Fit two types of sea ice clustering centers obtained in each block by linear regression.The next segmentation is carried out by using the Constant Center Markov Random Field(CC-MRF)model with fixed clustering centers.The final result can be obtained by combining all block.Experiments show that the proposed algorithm can effectively improve the segmentation accuracy of ScanSAR sea ice image and reduce the label confusion caused by the incidence angle effect.