Coastline Detection Algorithms Based On Mean Shift For SAR Images

Due to the advantages of all-weather and all-day,Synthetic Aperture Radar(SAR)images have been widely applied in the fields,such as in automatic navigation and sea area management etc.Coastline detection is one way to monitor coastal changes in sea area management.However,the speckles and the complex echo characteristics degrade the performance of SAR images.In coastline detection methods,mean shift algorithm is applied because of its unsupervised,no prior knowledge and fast convergence speed.However,in its iterative shift process,due to oversmoothed,resulting the wrong segmentation in the weak edge and the sharp corner.Moreover,the bandwidth parameter needs to be set manually.In the merging process,the traditional region merging algorithm of mean shift fails to fall land areas into one class due to complex land conditions.Therefore,from the perspective of the mean shift algorithm,this paper mainly focuses on the coastline detection algorithm in case of the weak edge with low contrast between sea and land as well as sharp corners using the restartting factor,coefficient of variation and piecewise Sigmoid function.The main work of this paper is as follows:1)A modified mean shift coastline detection algorithm based on restartting factor and coefficient of variation is proposed.The mean shift coastline detection algorithm has a weak edge and a sharp corner region with low contrast between sea and land in SAR image,which is too smooth due to imbalance of weight distribution,resulting in inaccurate detection of weak edges or sharp corners.Also,in the merging algorithm,some areas with similar seas in the land are not effectively merged.In order to solve the above problems,this paper presents an improved mean shift filtering algorithm.By introducing the restartting factor function,the edge with the corresponding restartting probability stays at the current point without filtering,thus keeping the edge from being smoothed.At the same time,an improved region merging criterion is given by integrating edge information and variation coefficient.The experimental data sets were Envisat-ASAR,Radarsat-2,and Sentinel-IA.For false positive ratio FPR,the proposed algorithm was well below0.0057,which verified that the proposed algorithm is more effective than the mean shift and Gamma level set algorithms.2)A modified mean shift coastline detection algorithm based on piecewise Sigmoid bandwidth and harmonic average is proposed.The traditional sean shift algorithm spectral bandwidth value is a single fixed value that needs to be mannully given,resulting in the over-segmentation at the ambiguous edge in the mean shift merging algorithm.In order to solve the above problems,this paper presents a modified mean shift filtering algorithm based on segmented Sigmoid bandwidth is presented.,constructing a segmented Sigmoid function using the region difference for the Sigmoid function,so that the bandwidth based on the segmented Sigmoid function is different due to the difference in pixel size.Then a modified merging criterion is proposed.By harmonic average,the criterion further merges the pixels belonging to the same type while maintaining the edge.Experimental data sets were Envisat-ASAR and Sentinel-1 A,for the false positive ratio FPR,the accuracies of proposed approach could be 0.0029,which verified that the proposed method is more effective than the mean shift algorithm,Gamma level set algorithm and object-based region merging algorithm on the problems of weak boundary and regional angle.