Research On Object-oriented Segmentation And Classification Methods Of High Resolution Remote Sensing Image

With the rapid development of remote sensing technology and the growing demand of high-resolution image, high resolution remote sensing images are becoming more and more popular. Compared with middle and low resolution remote sensing images, the high resolution remote sensing images has more details information. Therefore, the information retrieval from the high resolution remote sensing image require to use spectral information and spatial structure information. As a result, the object-oriented remote sensing image segmentation and classification method has become an advanced research hotspot.The object-oriented segmentation and classification method regard the object as a basic processing unit, because of the object has more category information than the single pixel and can represent category attribute in a multi-dimensional circumstances. Through the way of calculation and combination of characteristics, we can accomplish the information retrieval. We use image segmentation algorithm cut the images into regions of similar characteristics. Making full use of spectrum, shape, texture information, the segmentation results of image polygon is more closer to the real situation. The subsequent calculation and classification are based on object segmentation. Therefore, the image segmentation is foundation, which determines the accuracy of classification.The multi-scale segmentation method of fractal net evolution approach(FNEA) is a classic object-oriented segmentation method. The splitting process of multi-scale segmentation combines object spectrum with geometry information. The hierachical network structure of multi-scale segmentation provides different features suitable parameters and lower noise influences. FNEA method of segmentation parameters, on the other hand, is more, and each level has different segmentation parameters, operation cumbersome and parameter setting no strict mathematical theory of constraints, hierarchy, the more the more serious the man’s subjective influence. On the other hand, FNEA contains a large number of layers with many segmentation parameter, which influences people’s subjective greatly. FNEA also has the defect of tedious operation and parameter settings with poor mathematics theory foundation. In this paper, considering the above situations we experimented on a single scale segmentation of FNEA. We used fuzzy classification classify objects and compared the results with multi-scale segmentation. The multi-scale segmentation of FNEA and improved FNEA both had the overall classification accuracy of more than 85 percentage and Kappa coefficient over 0.8.The two methods of segmentation had almost same classification accuracy in the city’s classification and made a little differences in Suburban areas.The research showed that compared to FNEA multi-scale segmentation method, improved FNEA segmentation method had a more simple operation, a higher time-efficiency and an equivalent precision. Also, the improved FNEA segmentation method had a higher classification accuracy in images that feature dimension vary little.