High-Resolution Remote Sensing Image Classification Based On Feature Database And Information Fusion

In recent years,image classification is a hot topic in the field of remote sensing images.With the improvements of the resolution in remote sensing images,single feature is insufficient to describe the image contents and details.Information fusion,which could integrate different perspectives of feature descriptions,helps to obtain more comprehensive image representations and improve classification performances significantly.Current fusion methods mostly focus on a single level of information fusion,which can't take full advantage of the complementary information between different fusion levels.In this paper,a remote sensing image classification method combined feature-level and decision-level fusion was proposed.Firstly,several kinds of complementary feature was extracted.Then,the classification results of every single feature were fused in the way of Dezert-Smarandache Theory(DSmT)to obtain the decision-level fusion result;at the same time,several features were serial connected to generate a new feature,and the new feature was used to classify to obtain the feature-level fusion result.Finally,the feature-level and decision-level fusion results were adaptive fused to finish classification.To solve the Basic Belief Assignment(BBA)problem of DSmT,a posterior probability matrix constructed from training samples was proposed.The experiments on 21-classes of remote sensing images showed that the proposed method can both take advantage of the complementary information between different fusion levels and different features to obtain a better classification performance.In addition to the above work,a new soft platform called ELUOFDB,which is supported by the 973 program(“The theory and method of high-resolution remote sensing data processing and spatial information transformation”)is developed.ELUOFDB that can effectively manage the original remote sensing images and feature data,is an object feature database of high-resolution remote sensing images.It also provide services for the fundamental surveying and mapping,disaster emergency response and other major applications.Experiments on 5-class land use objects in ELUOFDB demonstrate that the proposed method,which based on the combination of feature-level and decision-level fusion,helps to improve the use of feature information as well as to boost the classification performances.