摘要:
针对模糊C均值算法对初始聚类中心比较敏感的问题,提出了一种改进的空间加权模糊C均值算法。该算法首先通过BP神经网络训练已知样本,得出初始隶属矩阵,从而提高了初始聚类中心的可靠性。鉴于空间数据具有自相关的特点,在算法中考虑邻域像元对中心像元的影响,并赋予不同的权重,使算法具有更强的抗噪能力。为验证方法的有效性,将改进的方法用于SPOT 2.5m遥感影像数据的分割,并与FCM、SFCM等算法的分割结果进行了对比分析,结果表明,本文方法取得了较好的效果。
Abstract:
In this paper,we propose an improved spatial-weighed fuzzy C-means algorithm since the traditional fuzzy C-means algorithm is more sensitive to the initial cluster centers than other commonly deployed algorithms,such as the FCM,SFCM algorithms.BP neural network algorithms are used to train samples and obtain an initial membership matrix,thus increasing the reliability of the initial cluster centers.Since spatial data have pattens of spatial auto-correlation,the neighboring pixels will contribute to the center pixel with different weights to robustly handle noises.Segmentation experiments were conducted using SPOT 2.5 meters remote sensing images to verify the effectiveness of our algorithm.In comparison with FCM,SFCM algorithms,the experimental results show that the proposed method obtains better results.
下载:
