目的：海水浮筏养殖是海域使用动态监测中的重要类型,合成孔径雷达卫星遥感影像可以克服海洋气象环境的不确定性,有效体现浮筏养殖目标信息。为了降低乘性相干斑噪声的敏感性,改进得到广义局部二值模式(Generalized Local Binary Pattern, GLBP),进而将其用于改进广义统计区域合并算法(Generalized Statistical Region Merging, GSRM),构建以GLBP_GSRM为核心的多特征集成模型,得到更具纹理一致性的超像素,实现浮筏养殖信息精确提取。方法：根据合成孔径雷达数据的乘性噪声特性改进局部二值模式算子得到GLBP算子,将其加入GSRM的合并准则中,结合纹理信息的超像素分割能得到更具纹理一致性的超像素,有效抑制相干斑噪声。进而利用非下采样轮廓波变换得到轮廓信息丰富数据特征,使用FCS(Fuzzy Compactness and Separation)算法聚类实现浮筏养殖信息的无监督提取。结果：实验选取辽宁省长海县邻近海域作为研究区域,针对C波段的Radarsat-2和X波段的TerraSAR的合成孔径雷达图像,与实地现场调查结果比较,所提模型对不同波段的合成孔径雷达图像均能精确无监督地提取浮筏养殖信息,定量定性的分析结果验证模型的有效性。结论：所提模型充分集成纹理特征、空间特征和轮廓特征,有效解决相干斑噪声和数据特征单一的问题,针对不同波段的合成孔径雷达遥感图像,均能在复杂的海洋背景中实现有效地无监督浮筏养殖信息提取,利于海域使用动态监测。
Objective: Floating raft culture is the main type in the dynamic monitoring of sea area. The satellite synthetic aperture radar image can overcome the uncertainties of marine meteorological environment and show the location of floating raft. In order to reduce the sensitivity of multiplicative speckle noise, generalized local binary pattern (GLBP) is improved and used to improve generalized statistical region merging (GSRM). Based on GLBP_GSRM as the core, the multi-feature integration model is constructed to get the super-pixels with more consistent texture, and to achieve accurate extraction of raft-breeding information. Method: Based on the multiplicative noise characteristics of synthetic aperture radar data, the improved local binary pattern operator is used to obtain the generalized local binary pattern operator, which is added into the merging criterion of generalized statistical region merging. The super-pixels from segmentation with texture information are more texture consistency, and they effectively suppress speckle noise. Then, the nonsubsamples contourlet transform is used to obtain the contour feature to enrich the data feature. Lastly, the fuzzy compactness and separation algorithm is used to cluster and achieve the goal that unsupervised extraction of the floating raft. Result: In the experiment, the sea area near Changhai County in Liaoning Province was chosen as the research area. The images of Radarsat-2 with C-band and of TerraSAR with X-band were used in the experiments and compared with the real results of field survey. The proposed model could precisely extract the information of floating raft. The results from quantitative and qualitative analysis show that the model is effective. Conclusion: The proposed model can take advantage of texture feature, spatial feature, and contour feature to solve the problems of speckle noise and lack of features. The synthetic aperture radar images with different bands can be used to extract effectively the floating raft in complex oceanic backgrounds, and it’s conducive to the use of dynamic monitoring of the sea.