Research And Application Of Multi-scale Segmentation For High-resolution Remote Sensing Image

With the rapid development of satellite remote sensing technology, the spatial resolution of remote sensing image is continuously improved, remote sensing application field is being more and more widely. Object information extraction from high resolution remote sensing image is one of the main contents of the application of remote sensing image. Due to the high resolution remote sensing images with complex spatial structure, geometry, rich texture information, large quantities and so on, object-oriented information extraction method arises and has been developed rapidly. The smallest element of processing is an image object with practical significances, it colligates multidimensional features of image, which can accurately and stably describe the actual surface features.The key technology of object-oriented information extraction is image segmentation. For the high-resolution remote sensing image, surface features are complex and diverse, different objects need to be expressed in different scales. Therefore, the study of multi-scale segmentation algorithm can meet the needs of high resolution remote sensing image information extraction, and improve the application level of remote sensing.In this paper, according to the characteristics and requirements of high-resolution remote sensing image, the theory of multi-scale segmentation algorithm is studied and the following aspects of work is carried out.(1) Analyzing existing segmentation algorithms, we choose the effective clustering and high efficiency of the mean shift segmentation algorithm as the basis of the segmentation algorithm, and improve mean shift combined with color-texture pattern. Analyzed the key technology of multi-scale segmentation, and based on the improved mean shift, we introduced and implemented a novel multi-scale segmentation.(2) In order to solve the segmentation of large quantity remote sensing data, we adopt the strategy of block parallel processing, and construct the "buffer" to eliminate the stitching line between data blocks, and thus to ensure the integrity of image segmentation. A series of experiments are carried out to verify the feasibility of this algorithm.(3) In order to verify the effectiveness of the multi-scale segmentation algorithm and its application in information extraction, ZY3 of Tongzhou District, Beijing, is selected as the experimental data, and then surface features elements information extraction experiments are carried out.