Research And Realization Of Object Oriented Nearest Neighbour Classification Algorithm

Classification is an important link in the process of getting the information of the land use / cover from remote sensing images, but traditional classification methods for remote sensing images cannot meet the progressive development needs of the high-resolution remote sensing image. The traditional classification methods fail to make full use of images provide the information of the spatial structure, the phenomenon of misclassifications and omissions are arose. Limitations of the basic classification and pixel units is broke to, and more semantic information to a number of adjacent pixels of the object (including Super object and the object) is used to be the processing units, Objects are composed by the same characteristics pixels and classified base on the characteristics of each object . Then remote sensing image classification and objectives of the feature extraction are achieved a higher level . The characteristics of the spectrum, more of structure geometric structure information, such as shape, color and texture are depend by Object-oriented classification.In this paper, this object-oriented classification technology to high-resolution remote sensing image has been studied systematically, and on this basis of the high-resolution remote sensing image classification system has been developed. The results of main research are as follows:1. The description method of image object features has been studied. The segmentation technology of high-resolution remote sensing image and some of definition who may be touched have been introduced and realized. The simulation and description of objects features based on image objects has been solved mainly, and the quantitative expression model of the spectrum, texture, shape and other features of image objects has been researched and established. The structural,save and visual expression of each description feature of the image object has been studied. So the sufficient a foundation of classification features have been provided for the object classification.2. The features optimization algorithm based on samples are researched. Feature optimization is central to classification; substantive characteristics cannot be described by few of feature, however many dimension of features may be reduce the arithmetic speed.In this paper, separable distance is as features optimization standard and to found optimization feature who in favor of classification, So the advantageous foundation have been provided for the object classification.3. The object-oriented nearest neighbor image classification algorithm is researched. Several commonly used distance and two kinds of near neighbour classifiers are introduced, pixel-based and object-oriented nearest neighbour classification methods are presented, and object-oriented method are primary studied. Feature normalization in computational process is resolved. The basic principle and realization steps of nearest neighbour classification are explained, and several methods of accuracy assess of based on samples classification are introduced simply.4. Test regional and datasource are presented. Based on the object-oriented high-resolution remote sensing data classification system, the classification case of ChangPing test is provided,and then to contrast the classification results of pixel-based, to choose a suitable solution for the high-resolution remote sensing images classification.Finally, the research work was summed up and the further research direction and some problems were pointed out.