Study On Multi-scale Segment Method Of High Resolution Remote Sensing Image

In recent years, with the development of remote sensing technology, remote-sensing images improve towards the "three high" direction that high spatial resolution, high spectral resolution and high time resolution. More high resolution remote sensing satellite has been applied, they provide more information than low resolution remote sensing images which includes spatial structure, geometry and texture information, topological relationship. The higher space resolution improve, the more it's data get, traditional Interpretation method with eyes can not meet the needs of users, so the only way to solve this problem is computer intelligent interpret, and the core technique is remote-sensing image segmentation.Object-oriented image segmentation shows a significant advantage in many application fields. And now, we don't have a lot of object-oriented image interpretation systems for high-resolution remote-sensing image, and it's seriously restricts the application of high-resolution remote-sensing images in all walks of life. So we need to research the high-resolution remote-sensing image segmentation and develop intelligent Interpretation system for remote-sensing image. This paper do some research based on this method, and the result is:(l)This paper present a multi-scale segmentation algorithm based on improved KL dispersion principle with summarizes a variety of segmentation theory, analyses strengths of them and combine with the traditional multi-scale segmentation algorithm. Aim at The under segmentation problem exist in fractal network segmentation methods for high resolution remote sensing image, this paper present a region merging standard with KL divergence as the core, improved a new fractal network segmentation method based on the principle of KL divergence. And design experiments to verify the effectiveness of the method.(2) To improve the efficiency of segmentation, this paper realizes a parallel segmentation method with MPI method and presents a method to resolve the segmentation problem of parallel lines. In this paper, the parallel technology using master-slave parallel segmentation strategy, master processes read image data and divide it into blocks, and then, slave processes receive data and run image segmentation data independently. Finally, master process receives the segment result and put it together which is the image that we need. For the parallel segmentation line's problem, this paper presents a method that operate vector object directly, to combine similar pattern which have same features around the segmentation lines, and clear the segmentation lines.(3)This paper present an intelligent remote sensing image interpret framework based on Geographical condition monitoring project, to use it in image segmentation and classification or typical Automatic extraction of elements. This system provides a series functions from image display to segmentation.System's process is very efficient which contains image segmentation, feature extraction, objects classify and so on. It make interpret process can use vector polygons as operator objects and consider raster image as reference layer to execute the man-machine interactive visual interpretation.