Research On Uncertainty Modeling And Fuzzy Classification Of Multi-source Remote Sensing Images

Remote sensing technology has been widely used in various fields such as disaster monitoring,land use planning and vegetation protection monitoring,and has become an important research tool for solving geospatial information problems.Remote sensing image classification is a key technology in remote sensing applications,and the inherent uncertainty of remote sensing images and the limitations of existing image classification algorithms seriously affect the accuracy of remote sensing image classification,thus restricting its commercial development and practical application.Accurate and reliable remote sensing image classification results are crucial to the application of remote sensing images.Therefore,this paper addresses the phenomenon that the existing uncertainty assessment models often focus on the uncertainty in the classification process and the classification results,while ignoring the inherent uncertainty of remote sensing images themselves.It proposes to analyse the sources of uncertainty in remote sensing image data and its essence from the characteristics of remote sensing image data,and construct a remote sensing image uncertainty assessment model to calculate the uncertainty degree of pixels in remote sensing image classification.Then,based on the uncertainty measure results combined with the neighbourhood information,the remote sensing image fuzzy clustering algorithm is improved to avoid the impact of data uncertainty on remote sensing image classification as far as possible and to improve the classification accuracy and reliability.The main research work is summarised as follows:(1)Construction of uncertainty assessment model for remote sensing image classification.The uncertainty assessment model of remote sensing image classification is constructed based on two aspects,namely classification error-prone points and feature differences.The uncertainty assessment model is applied to noisy composite images and multi-source remote sensing images,and it is demonstrated that the proposed uncertainty assessment model can reasonably reflect the uncertainty level of each pixel in remote sensing images.(2)Improvement of fuzzy clustering algorithm based on uncertainty assessment and neighbourhood information.The uncertainty of remote sensing image data can affect its classification results,thus reducing the reliability of remote sensing image applications.Therefore,the uncertainty assessment results of remote sensing images are introduced into the classification process,and then fused with pixel neighbourhood information for filtering,to design a fuzzy C-mean clustering algorithm improved based on the neighbourhood trust factor,in order to effectively circumvent and reduce the influence of image uncertainty on the classification results and improve the classification accuracy.Synthetic images and multi-source remote sensing images are selected for experiments and compared with the fuzzy C-mean(FCM)clustering algorithm and its series of improved algorithms(FCM＿S,ENFCM,FGFCM and FLICM),and the performance of each algorithm is quantitatively evaluated by partition coefficients,partition entropy and overall classification accuracy OA,The results show that the proposed algorithm has stronger noise immunity and robustness,and improves the classification accuracy of remote sensing images.