Research On Automated Determination Of Optimal Segmentation Results For High-Resolution Remote Sensing Data Based On Unsupervised Method

Object-based image analysis is the mainstream method for extracting information from high-resolution remote sensing data,and image segmentation is an important part of this method.The segmentation result provides basic data for image information extraction,which affects the accuracy of information extraction.Therefore,it is significant to obtain the optimal image segmentation results from different image segmentation results for extracting image information.Currently,there is no efficient and reliable research scheme and method in this field at home and abroad.In this paper,we adopt the idea of unsupervised evaluation to explore an objective and automatic method to obtain the optimal segmentation result from high-resolution remote sensing image.The research contents of this paper are reflected in the following three aspects:(1)The overall optimal segmentation method based on remote sensing imageAiming at the problem that the existing methods can not take into account the multi-spectral information of high-resolution remote sensing data,this paper constructs spectral information dispersion(SID)index based on the information entropy principle,which is used to respectively express the uniformity ((?)_mean) of the segmentation object and the difference ((?)_mean) between adjacent objects.Segmentation parameter range(SPR)is constructed based on (?)_mean and (?)_mean.By analyzing the change of SPR index value with the segmentation parameters,the parameters of serious under-segmentation results are filtered,and parameters range are automatically obtained.This method effectively avoids the problem of setting segmentation parameters subjectively in the existing methods.Beside,the segmentation entity index(SEI)is constructed based on (?)_mean and (?)_mean to express the quality of image segmentation results.According to the SEI value,the optimal segmentation parameters range(L,R)is obtained.In the range(L,R),the segmentation result of the maximum SEI point is considered as the optimal segmentation result based on remote sensing image.In order to obtain the optimal segmentation result based on segmented object for the next step,the constructed E value method is used to obtain the better segmentation results in the range(L,R)based on remote sensing image(ie: some type of surface feature objects on the image are optimal in the segmentation result).The parameters of the better segmentation results based on remote sensing image are taken as the constraints to obtain the optimal segmentation result based on segmented object for the next step.(2)The global optimal segmentation method based on segmented objectThe optimal segmentation result based on remote sensing image does not guarantee that each type of surface object in remote sensing image can achieve its optimal segmentation.Some objects in the image are over-segmented or under-segmented.Optimizing these segmentation objects can improve the quality of segmentation.Therefore,this paper obtains the under-segmented objects with the segmentation object index(SOI)constructed,based on the overall optimal segmentation result of image.These objects are segmented with parameters of the better segmentation results based on remote sensing image,and the global better segmentation result based on segmented objects is obtained.the over-segmented objects are obtained with the SOI index constructed based on the global better segmentation result.Similarly,the over-segmented objects are dealt with parameters of the better segmentation results based on remote sensing image,and the global optimal segmentation result based on segmented objects is obtained.(3)Validation and application of the methodComparing the overall better segmentation result,the overall optimal segmentation result,the global better segmentation result,the global optimal segmentation result and the reference segmentation result,it proves the validity of method proposed in this paper that automatically gets optimal segmentation result for highresolution remote sensing image.The method is compared with other mainstream method to prove its superiority.At the same time,the method is applied to other experimental areas,which further proves that it has certain universality.