摘要
当前高空间分辨率遥感图像分割与目标检测仍然面临着精度与效率的两难问题。文章提出了一种基于图特征的随机行走分割方法来提高分割效果。该方法有3个步骤。首先,通过奇Gabor滤波和分水岭变换将图像从像素转换为图,并进行基于图的特征表达。其次,利用图像的光谱特征、纹理特征、形状特征和位置特征,构建加权函数。为了有效地表达特征,采用KNN(K-nearest neighbors)高分标签交互选择种子点。最后,通过建立拉普拉斯函数并求解Dirichlet边界问题,对最大可能性进行标记,完成随机行走分割。文章讨论了加权函数中的参数设置,并通过对比实验对基于高分辨率遥感图像的分割结果进行了评价。研究表明,文章所采用的结合图特征的随机行走分割算法只需要小样本便可以实现更精确的分割。
The image segmentation and object detection from high spatial resolution(HSR)remote sensing images still face the dilemma between accuracy and efficiency.Here we propose a method of graph-based random walk(GRW)segmentation to improve the effect.Three steps are taken for our GRW segmentation.Firstly,images are transitioned to graphs by odd Gabor filtering and watershed transforming for feature representation.Secondly,the graph features,including the spectral,texture,shape,and location features,are considered to build a weighting function.To express the features effectively,interactively selected seeds with KNN(K-nearest neighbors)high score labels are suggested.Finally,by setting up the Laplace function and solving the Dirichlet boundary problem,the max possibilities are labeled to complete GRW segmentation.The parameters in the weighting function are discussed in the article,and segmentation based on GRW of QuickBird images is evaluated by comparing tests.Our results show that the GRW method can achieve more accurate segmentation,while needing small samples.
作者
赵好好
管海燕
ZHAO Haohao;GUAN Haiyan(School of Remote Sensing and Geomatics Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China)
出处
《遥感信息》
CSCD
北大核心
2023年第6期95-102,共8页
Remote Sensing Information
基金
国家自然科学基金(41801239)。
关键词
随机行走
高分辨率遥感图像
图像分割
图特征
特征表达
random walk
high resolution remote sensing image
image segmentation
graph feature
feature expression
