Fuzzy Classification And Application Of Remote Sensing Image Based On Interval Modeling

The inherent uncertainty of remote sensing image data and its ubiquitous phenomenon of the “same object with different spectra” often lead to the ambiguity and uncertainty of attributes in the computer classification results.The complexity and interference of the ground information,which bought by the high-resolution images,is a new challenge for the robustness of the algorithm.How to characterize the uncertainty of remote sensing image data,enhance the robustness of the algorithm,and establish a high-precision classification model are the keys to remote sensing image classification.It has important theoretical significance and application value for the land cover classification using remote sensing technology.Interval-valued data model,used to describe the variability and ambiguity of observed data,may be able to characterize the intra-class heterogeneity in remote sensing imagery data for better classification performance from the source.Fuzzy set mathematical theory is an effective way to express uncertainty.Fuzzy c-means algorithms have been widely studied in the field of remote sensing image classification,but they are not ideal for the classification of remote sensing image data with density differences and large uncertainty.In contrast to type-1 fuzzy sets,type-2 fuzzy sets are better able to describe multiple types of fuzzy or uncertain information and are more suitable for dealing with the inter-class multiple uncertainties of in remote sensing image classification.Therefore,the research contents of this paper include four parts: uncertain data description of remote sensing image in the classification process,robustness analysis of fuzzy algorithms,applicability of fuzzy algorithms,and application of classification model for the land cover classification in Yantai.The main research works are summarized as follows:1.We create the interval-valued data model to characterise the uncertainty of multi-band remote sensing imagery data and propose the adaptive factor definition and the corresponding of control strategy according to the particularity of different data samples.We theoretically prove that the preferential interval-valued data model has greater separability than the traditional model,and then a preferential adaptive interval-valued fuzzy clustering algorithm is proposed.Compare with other state-of-the-art fuzzy classification methods and interval-valued fuzzy classification methods,our algorithm obtains the better classification performance from the multi-spectral remote sensing image classification results which markedly improved the ground-object separability and suppresses the density difference2.According to the proximity principle of the geographical first theorem,a spatial gravity model based on reliability is proposed to characterize the nonlinear relationship between pixels in local space to improve the robustness to "noise" interference and intra-class heterogeneity during the classification process,where the influence of the neighbouring pixel on the central pixel in local space is inversely proportional to its spatial distance.And then we propose an adaptive interval spatial fuzzy c-means algorithm.Multiple sets of artificial experiments and high-resolution multispectral remote sensing images show that,compared with the other state-of-the-art fuzzy c-means algorithm,our algorithm balances the contradiction between preserving classification details and suppressing intra-class heterogeneous information,and obtains the classification results with more compact classes and obvious boundaries.3.For remote sensing image data with higher density differences and high-order uncertainty,we propose a novel classification method based the semi-supervised adaptive interval type-2 fuzzy c-means algorithm.The fuzzy metric method is used to construct the interval type-2 fuzzy sets with stronger uncertainty characterization ability that reduces subjectivity of fuzzy index selection.By using the semi-supervised approach,an evolutional fuzzy weight index m by is proposed.The strategy of adaptive control is introduced to find the method of reducing the equivalent fuzzy set to improve the fuzzy control ability and reduce the computational complexity.In addition,soft constraint supervision is performed using a small number of labeled samples,which optimizes the iterative process of the algorithm and determines the optimal set of features for the data and enhances the applicability and robustness of the algorithm.Experimental results show that the proposed algorithm has better classification performance and better applicability for remote sensing image data with large coverage area and abundant ground objects.4.For the high resolution remote sensing images in Yantai,the proposed algorithm combined with object-based method: adopt ENVI 4.6.1 software for GF-2 image feature extraction,to extract the segmentation unit as an object,respectively,using the proposed algorithm to classify and compare the classification performance of the algorithm,at the same time in the meet the needs of actual project application to verify the applicability of the proposed algorithm for different data.The experimental results show that the proposed algorithms all obtain high quality classification results.