Research On Remote Sensing Classification Based On Super Pixel Initial Segmentation And Region Merging

Image segmentation is indispensable in image processing.In the exploration and application of high-resolution remote sensing images,sometimes it is necessary to extract information about some parts of the image(some areas).Since the 1970 s,it has been highly concerned by people.In previous studies,although scholars have proposed a variety of different segmentation techniques,but due to the specific problems of various images,a variety of sizes and image differences in specific regions,there is still a lack of a universal.A common segmentation method that can be used for all images.Similarly,there is a lack of scientific and general criteria to determine the success of image segmentation.Using high resolution GF-1 remote sensing data as data source,the image processing technology of high resolution remote sensing image suitable for large area is studied from two aspects of image segmentation and image classification.The work done is summarized as follows:The main results are as follows:(1)based on GF-1 satellite image,the initial image segmentation is completed by watershed transform,and the suitable image enhancement technology and noise removal technology are selected to reduce image over-segmentation.Combined with the method based on graph theory,by establishing a certain measurement criterion of regional heterogeneity,the initial segmentation regions are merged to form the super-pixel segmentation results of the original image.(2)based on GF-1 satellite image,the result of super-pixel segmentation is used as the feature extraction sample of support vector machine image classification,and the land cover type of highresolution remote sensing image is extracted by selecting the appropriate feature set.(3)complete the image classification experiment and corresponding evaluation of the high resolution image in the study area.Finally,the accuracy analysis and comprehensive analysis of the classification results are carried out.The results show that the algorithm proposed in this paper has high extraction accuracy and well realizes the image classification of remote sensing images in large areas.