Abstract: The inherent speckle noise in Synthetic Aperture Radar (SAR) images brings difficulty of SAR image segmentation. Especially in the case of high resolution, the real data information of many details is submerged by the speckle noise. A SAR image thresholdsegmentation method based on grey number and improved flower pollination algorithm (SDEFPA) is proposed in this paper.Segmentation threshold can be regarded as an interval grey number, whitening process can be finished via improved flower pollination algorithm (SDEFPA). In SDEFPA, Sobol sequence strategy is used to initialize the population to improve the population uniformity, and differential evolution mutation strategy is introduced to enhance the ability of flower pollination algorithm (FPA) to escape from local optima. The experiment results on the CEC 2017 benchmark functions show that SDEFPA has better optimization accuracy and convergence speed than 5 comparison algorithms. Application of SDEFPA to SAR image segmentation, grey entropy is selected as the fitness function of SDEFPA. Between global pollination and local pollination randomly switching of multiple cyclesby switch probability 0.8, the best flower is gradually close to the optimal threshold. The optimal segmentation is searched quickly by SDEFPA and meanwhile the grey numbers are whitened. Then achieve the final segmented image by whitened threshold. The proposed segmentation method obtains the best effect using the least time compared with contrast segmentation methods. The results demonstrate effectiveness of the proposed method.

Key words: SAR image segmentation, Improved flower pollination algorithm, Grey number, Grey entropy