视觉皮层振荡神经网络在遥感影像分割中的应用  

Remote Sensing Image Segmentation Using Neural Oscillator Network of Visual Cortex

作　　者：施蓓琦[1,2] 刘春[1] 姚连璧[1] 陈能[2] 

机构地区：[1]同济大学测量与国土信息工程系,上海市四平路1239号200092 [2]上海师范大学城市信息研究中心,上海市桂林路100号200234

出　　处：《武汉大学学报（信息科学版）》2011年第12期1442-1446,共5页Geomatics and Information Science of Wuhan University

基　　金：上海市教育委员会科研创新资助项目(10ZZ25;11YZ290);国家教育部留学回国人员科研启动基金资助项目;上海师范大学校级科研资助项目(SK200921);光华同济土木学院基金资助项目

摘　　要：对于具有复杂场景的遥感影像,提出了基于各向异性扩散特征保持平滑的简化LEGION算法,弥补传统LEGION图像分割算法对噪声敏感的不足。对于多(高)光谱影像,则采用马氏距离计算光谱向量间的连接权得到扩展LEGION算法进行分割。实验结果表明,LEGION算法较传统的图像分割方法对单波段灰度遥感影像目标及其重要细节部分的分割效果更显著;扩展LEGION算法适用于多光谱遥感影像分割。Locally excitatory globally inhibitory oscillator network（LEGION）,a neural oscillator network based on biologically framework,is applied to remote sensing image segmentation.For remote sensing images with complicate environment,a simplified LEGION algorithm with improved anisotropic diffusion method is presented to overcome sensitiveness.In multi-spectral images all the pixel values of the same location is considered as a vector.The connection weight between two oscillators is determined by a Mahalanobis distance metric between the two corresponding vectors.Experiments on real panchromatic image show that the proposed method can effectively produce more accurate segmentation than traditional algorithms.And the extending LEGION image segmentation algorithm can also be applied to process multi-spectral images segmentation.

关 键 词：遥感影像分割 振荡关联 局部兴奋全局抑制振荡网络 各向异性扩散 

分 类 号：P237.4[天文地球—摄影测量与遥感]