论文部分内容阅读
为了实现影像的自动化分割,提出一种利用非监督方式将观测数据采样化的遥感影像分割方法。该方法利用欧氏空间的概率分布建模采样数据和观测数据,并将其映射到黎曼空间,通过不断将观测数据转换为采样数据的方式实现影像的自动采样化。每次采样过程只需计算观测数据点到采样点的测地线距离,将距采样点测地线距离最小的观测数据转化为采样数据,以保证采样数据不断趋于该类数据的真实分割结果,同时使算法能够有效分割具有不同像素数的类别。将算法应用于模拟影像和真实遥感影像分割,对其分割结果以及传统基于统计、基于模糊的非监督算法和基于神经网络的监督算法相应分割结果定性定量的对比分析验证了该算法的有效性及可行性。
In order to realize the automatic segmentation of the image, a remote sensing image segmentation method using the unsupervised method to sample the observed data is proposed. This method uses the probability distribution of Euclidean space to model the sampling data and observation data and maps it into the Riemannian space, and automatically images the image by continuously converting the observation data into the sampling data. Each sampling process only needs to calculate the geodesic distance from the observation data point to the sampling point and the observation data with the smallest distance from the geodesic of the sampling point into the sampling data so as to ensure that the sampling data tends to be the true segmentation result of the data , While allowing the algorithm to efficiently partition classes with different pixel numbers. The algorithm is applied to the segmentation of the simulated images and the real remote sensing images. The qualitative and quantitative comparison of the segmentation results and the corresponding results of the traditional statistical-based, fuzzy-based unsupervised and supervised algorithms based on the neural network verifies the effectiveness of the algorithm. feasibility.