Hierarchical Muti-scale Segmentation Of Remote Sensing Image Based On Spectral Histogram

In remote sensing image texture segmentation,selection of the scale remains a constraint segmentation accuracy difficulties.First,for single level multi-classification method can't make full use of the scale of the texture of different vegetation target so as to achieve more accurate multi-classification problem,the authors proposed a hierarchical multi-scale remote sensing image vegetation segmentation method based on spectral histogram.Further promote this method,the authors proposed a riverine wetland remote sensing image hierarchical multi-scale segmentation method,make the hierarchical multi-scale segmentation method promote to segmentation processing of the main feature targets of remote sensing image.The study of vegetation classification,vegetation fine-grained segmentation generally have three targets,according to the scale divided into the arbor,shrub,grass and moss.First,the vegetation areas in remote sensing images were extracted with the normalized difference vegetation index(NDVI),and then the multiple binary classification algorithm was implemented in the region to achieve multi-classification operation.In each classification level,the advantage of the prior knowledge and texture scale were taken to selected texture filtering parameters,the spectral histogram of each sub-block image was extracted from the filtering result to express texture features,so as to achieve the segmentation of a level.The experimental results show that the proposed method well used the prior knowledge and texture scale of vegetation target at all levels,so that made the texture filter to enhance treatment more targeted,the spectral histogram feature has much more degree of differentiation,and the accuracy of the vegetation fine-grained segmentation has been improved significantly.Make the segmentation feature target from vegetation to expand to the relevant features,taking an example of Riverine wetland study hierarchical multi-scale texture segmentation methods.First,completed the hierarchical segmentation of grass and moss,shrub,arbor in the vegetation areas based on the normalized difference vegetation index(NDVI)method.The next,completed the two levels of segmentation of water and shoals in non-vegetation area.Every level segmentation is implemented by a specific texture scale binary classification algorithm,texture scale of vegetation target acquired by the blue noise detection method,and the texture scale of other target is determined by the prior knowledge and the resolution of the image.In every level,according to texture scale to set the texturefiltering parameters,using the spectral histogram to express texture features.The experimental results show that the proposed method make full use of the texture at every level target,significantly enhance the effect of texture filtering,and the accuracy of the every feature target segmentation has been improved significantly.