High Resolution Remote Sensing Image Classification Based On Different Features Fusion On Various Optimal Scales

In the object-oriented high-resolution remote sensing image classification method, the image is segmented into a few homogeneous objects firstly;various features of the target objects are extracted secondly;image classification is completed finally. An integrated object-oriented technology solution of high resolution remote sensing image classification is proposed in this paper,whose main content are as following:(1) Multi-scale segmentation model is constructed to segment high resolution remote sensing image.Firstly, the index of homogeneity and heterogeneity are constructed,which is used to establish multi-scale segmentation model.The optimal segmentation scales of different objects through the model are available to extract features of a variety of objects and the experiment verifies the segmentation results.(2)There are different segmentation results corresponding to various objects.In order to fully extract the characteristic of the object,features of objects will extracted in their respective optimal scale,that is bound to characteristics of duplication and redundancy.For purpose of eliminating redundant features,improved SFFS algorithm is proposed firstly.Then,mutual information is applied to remove redundant features.Finally the initial features set is generated.(3)Multiple SVM classifiers are combined in order to realize the classification of multiple classes.For making full use of the features,each classifier training a feature set and combination of multiple kernel function is utilized to improve the classification accuracy.On the basis of previous researchers’ work,the determination of parameter selection of kernel function is made by means of genetic algorithm which dynamic changes in the process of training parameters until the global optimal solution be found.