基于K-means聚类的遥感影像条状地物半自动提取方法

A Strips Features Semi-automatic Extraction Method of Remote Sensing Image Based on K-Means Clustering

遥感图像地物目标提取是遥感图像分析的关键步骤,通过分析遥感图像的频谱特性,提出一种基于K-means聚类的地物目标提取方法。首先通过时域和频域相结合的方法对原始图像进行增强,再利用K-means聚类算法对图像各个分量进行聚类,聚类结果分为目标类和背景类,然后分别计算每一类的特征值均值和方差,迭代两类像素的灰度值,同时结合数学形态学和阈值方法进行地物目标提取,得到最终的目标提取结果。实验对多幅遥感图像进行不同地物目标的提取,实验结果表明:该算法具有很好的抗噪能力,目标提取结果精确,有一定的现实意义。

Through analyzing the characteristics of remote sensing image, this paper proposed a segmentation method based on K-means. Firstly we did image enhancement with time domain and frequency domain combination. Then, clustering segmentation algorithm was employed to all components for poly class, which was divided into target class and background class. After that, each characteristic value of the mean and variance value was calculated to do iteration on any two class pixel. Finally, image segmentation was finished and results were achieved using mathematics morphology and threshold value. We tried to do extraction on multiple images with different target. The results showed that the algorithm has good resistance to noise with fast operation, and it also has certain significance.