Object-Oriented Classification Technique Of Remote Sensing Image Used In Classification Of Land Use

With the development of remote sensing technology, remote sensing image has a higher and higher spatial resolution and makes us obtain very abundant information of nature. But the high spatial resolution remote sensing data application efficiency is not high at present.One important reason is that the traditional image classification method is based on spectral of pixels, can not effectively extract the spatial information. Pixel-based classification method also has the phenomenon of salt and pepper classification problem, resulting in the appearance of a number of broken polygons, ultimately leading to the low classification accuracy. A wide range of land use / cover data in real-time updates on the application of high resolution remote sensing images as well as challenges. High resolution remote sensing images are often costly, and the huge amounts of data of it need to spend a lot of time and manpower to complete the extraction of information.High resolution remote sensing images in real-time and large-scale land use / cover information extraction does not have practical significance.In order to solve the problem, moderate resolution remote sensing image data and high resolution image data were integrated based on object-oriented remote sensing image classification method to realize large scale real-time land use / cover information extraction. In this paper Songpan County is the study area, extract the data of representative region in the form of high-resolution remote sensing data as information source, and guide a larger scope of moderate resolution remote sensing image data classification by some typical parts of high-resolution images classification. In this paper, object-oriented classification method was used instead of traditional pixel-based approach, which was an image-object-based approach. The proposed object-oriented remote sensing image classification method had some advantages compare to traditional classification method: image object derived from homogeneous pixel by image segmentation, whose interior information could be ignored for little variance of spectral value, so that salt and pepper phenomenon could be avoided. The proposed method comprehensively combines two factors of spectrum and shape, which provides more features for classification, and effectively overcomes the limitations of pixel-based classification as a result. multi-scale spatial analysis can meet requirements of information extraction for different scale features. It can also fully take advantage of all features of images by mimicing human brain's way of thinking, so as to achieve the purpose of extracting information as high accurate as possible. Last but not least, it can integrate use of multi-source data, and make fully and quick use of multi-source remote sensing data and existing thematic data.In this paper, Songpan County, Sichuan Province is the study area. The research data are county-wide TM images of medium-resolution and Songpan County regional SPOT5 images of high-resolution. study the advantage and differences of pixel-based classification and object-oriented classification for SPOT5 image and TM image of Songpan County, then apply the result to object-oriented classification of county-wide TM images. The result is that object-oriented classification method is far superior to pixel-based classification method in high-resolution image, and is a little better than pixel-based classification method in medium-resolution image. As a result the county-wide land use / cover information extraction is well done by object-oriented classification for TM images of the whole county under the guide of typical regional SPOT5 image classification results.