Multi-Scale Segmentation Method Study Based On Object-Oriented Classification

As a data source, remote sensing image especially the high resolution remote sensing image, is more and more important in territorial resources surveys. The Object-Oriented Classification is widely applied due to the features of high resolution. The basic part of Object-Oriented Classification is image segmentation which create data source of classification afterwards. Beijing Aoyuan Soprt Club is the study area contenting building, road, sand, grass and water. Remote sensing image is a part of WorldView-2 product. The review and summarize previous research results and reference based on the conclusion of research in related fields, of high resolution remote sensing image in image segmentation of whether it is necessary to appropriate to deal with the noise are discussed, the auxiliary remote sensing image optimal segmentation scale selection of segmentation quality evaluation index is modified and validated by the experiment.The results of this paper are as follows:1、 Based on local standard deviation of image noise evaluation method of image noise levels were assessed and use based on compressed sensing image to noise technology realized in the retention of the main structure of the image based on the image denoising, image quality improved significantly, effectively avoid the influence of noise information on the following segmentation operation;2、 This paper in high resolution remote sensing image processing process is optimized, the image using the first principal component as the image segmentation, from a certain degree to go up in addition to the redundant information of the image, highlighting the main features of the image and eliminates the multidimensional image information to the image segmentation of uncertainty.3、In the image segmentation, the segmentation quality evaluation index of this paper is used to evaluate the results of each segmentation, and based on the evaluation results, to guide the selection of the optimal segmentation scale. Revised index the ratio form, the numerator and denominator respectively, indicating segmentation image of internal homogeneity and external heterogeneity, molecule is weighted by area correlation coefficient and the denominator is aple. Compared with the original evaluation index(the standard deviation of the area weighted and the weighted sum of Morans’ I), the new evaluation index has been strengthened in objectivity and sensitivity, and good results have been achieved.