基于非规则标识点过程的遥感图像零星地物目标几何特征提取

Geometric feature extraction of scattered targets from remote sensing images based on irregular mark point process

针对遥感图像中零星目标几何特征的非规则性,提出基于非规则标识点过程的遥感图像零星目标几何特征提取方法.首先,将图像分为目标类和背景类,利用非规则标识点过程建立目标的分布和几何特征,其中非规则标识采用一系列由节点连接起来的多边形定义而成,用以拟合目标几何形状,标识点用以确定目标位置;然后分别利用多值高斯分布和KL(Kullback Leibler)散度定义特征场能量函数和异质性能量函数,形成全局目标提取能量函数,应用非约束吉布斯表达式将全局能量函数转换为概率分布函数,并在最大化概率分布函数准则下设计合适的M-H(Metropolis-Hastings)采样算法,获得最优目标提取结果;最后,采用所提出方法对遥感图像零星目标进行提取,由实验结果可以看出,所提出方法不仅能准确定位各零星目标,而且可以对其几何形状进行精确拟合.

In view of the irregularity of the geometric features of the scattered targets in remote sensing images, a geometric feature extraction method for remote sensing images based on irregular mark point process is proposed. Firstly,the image is divided into target and background, and the distribution and geometric features of the target are established by using the irregular marking point process. The random point are used to determine the target position, and the marks associating individual points are defined with a set of nodes to depict the geometric shape of the target. Then, on this basis, the multi-Gaussian distribution and Kullback-Leibler divergence are used to define the energy function and the heterogeneous energy function, and form the global energy function. The global energy function is converted to the probability distribution function by using the unconstrained Gibbs function, and the appropriate Metropolis-Hastings sampling algorithm is designed under the maximization probability distribution function criterion. The optimal target extraction result is obtained. Finally, the proposed method is used to extract the scattered targets from remote sensing images, it can be seen from the experimental results that the proposed method achieves better experimental results. The traditional mark point process method is extended to the irregular mark point process. According to the experimental results of remote sensing image extraction, it can be seen that the method can be more accurate to fit the scattered targets.