基于增强学习机制的SAR图像水域分割方法

SAR IMAGE WATER AREA SEGMENTATION METHOD BASED ON ENHANCED LEARNING MECHANISM

合成孔径雷达(Synthetic Aperture Radar,SAR)图像的水域分割任务在航运监视、违法船只捕捉等遥感信息处理领域具有较高的研究价值。当前主流分割算法对于SAR图像信息的利用不充分,造成分割性能欠佳。针对SAR图像样本数据不足、目标特征信息不明显的问题,从这两方面入手,利用数据增强和特征增强的学习机制,在扩充样本数据集的同时增强目标特征的显著性。通过定性和定量的实验结果分析,在相关数据集上,该方法可在不增加过多计算成本的基础上提升分割准确率。

The task of water area segmentation in synthetic aperture radar(SAR)image has high research value in the field of remote sensing information processing,such as shipping surveillance and illegal ship capture.At present,the mainstream segmentation algorithms do not make full use of SAR image information,resulting in poor segmentation performance.In view of the problem that the sample data of SAR image is insufficient and the feature information of target is not obvious,this paper started from these two aspects,and used the learning mechanism of data enhancement and feature enhancement to expand the sample data set and enhance the significance of target features.Through the analysis of qualitative and quantitative experimental results on relevant data sets,this method can improve the segmentation accuracy while not increasing too much calculation cost.