基于改进遗传算法遥感图像非监督分类研究

Research of unsupervised classification of remote sensing image based on modified genetic algorithm

传统的非监督分类方法通过人为预先设定的类别数把像素划分到相应的类别中,但类别数事先不能精确得到,因此会增大误分率,降低分类精度。提出一种新的可变聚类数目的染色体、采用Davies-Bouldin系数作为适应度,通过对传统遗传算法的一系列改进自动进化出高分辨率遥感图像的类别数和聚类中心。同时,采用整型数据来进行染色体编码,不仅降低了计算复杂度,同时也节省了存储空间。算法已用VC实现程序设计,程序结果证明该改进算法的正确性并获得令人满意的实验结果。

The traditional unsupervised classification divides all pixels into a corresponding class pixel by pixel.But the clustering number is not obtained accurately in advance,which would result in increasing the error classification rate and decreasing the classification accuracy.A new modified genetic algorithm is brought forward in order to obtain the clustering number and clustering centers of high resolution remote sensing image automatically properly.First,a new chromosome string is encoded by integer data in order to satisfy the variable clustering number,which not only decreases the computing complexity but also saves the memory space.Second,the Davies-Bouldin index is used as a measure of the validity of the clusters.Finally,a series of measures are used to modify the traditional genetic algorithm in order to enhance the adaptability and robusticity.All these are programmed with VC.The results prove the validity of the modified genetic algorithm and an optimization result of the remote sensing image classification is obtained.