Research On High Resolution Remote Sensing Image Classification Of Object Oriented Technique

With the continuous development of space science and technology, the growing emergence of new sensors, the quality of the remote sensing image data is more and more high. The use of high resolution remote sensing image data such as in the digital city construction, mineral resources exploration, the land use classification and change detection of surveying and mapping, national defense, aviation, military and precision agriculture are more and more widely in the field of production and so on multiple disciplines, remote sensing information technology has got unprecedented development and concern, fully processing analysis of remote sensing image data to obtain useful information for human is the hot topic of research, and the important step of the research is the remote sensing image classification. The main research content of this article is in view of the high resolution remote sensing image using object-oriented technology for image classification.At present more mature remote sensing image classification is based on the pixel spectrum statistical classification method, common divided into two categories: supervised classification and unsupervised classification method, this paper discusses the commonly used the mean K- Means clustering algorithm, minimum distance method and B P neural network and support vector machine(SVM) that several methods of remote sensing image classification. Classification method based on pixel only uses the spectral characteristics of information, the precision of containing less spectral information of high resolution remote sensing image classification is lower.In order to solve the above problems according to the characteristics of the high grade image data, this article take the object-oriented classification technology, its core steps include image segmentation and image classification, the first step in image segmentation directly determines the accuracy of the image classification. The result of segmentation is based on the rule of regional heterogeneity and homogeneity, using the multi-scale segmentation technology based on region growing image segmentation into homogeneous image object units, the second step is conducted on the basis of image segmentation object feature selection and extraction, and then through the fuzzy classification and KNN classification method to complete the image classification, nearby the most object-oriented classification technology to make full use of the high grade image data abundant spectral information, shape and texture characteristics, such as spatial information, to overcome the phenomenon of "salt and pepper", effectively reduced the "homogenous different spectrum" and "heterogeneous with spectrum" phenomenon, and improved the classification precision of remote sensing image.