Discrete Graphical Models for Alphabet-Based Multisensory Data Fusion and Classification

The way of multisensory data integration is a crucial step of any data fusion method. Different physical types of sensors (optic, thermal, acoustic, radar, etc.), different resolution, and different types of GIS digital data (elevation, vector maps, etc.) require a proper method for data integration. Incommensurability of the data may not allow to use conventional statistical methods for fusion and processing of the data. Correct and established way of multisensory data integration is required to deal with such incommensurable data, while employment of an inappropriate methodology may lead to errors in the fusion. To perform a proper multisensory data fusion several methods were developed (weighted Bayesian, linear (log linear) opinion pool [1], [2], neural networks [1]-[3], fuzzy logic approaches [4], etc.). Employment of these approaches is motivated by weighted consensus theory, leading the fusion of incommensurable data to be performed in a correct way. In this paper data fusion is proposed to perform using a finite predefined domain alphabet. Feature extraction (data fission) is performed separately on different sources of data. Extracted features are processed to be represented on the predefined domain (alphabet). Alternative method such as factor graph (discrete graphical model) is employed for data and feature aggregation. The nature of factor graphs in application on data coded on a finite domain allows us to obtain an improvement in accuracy of real data fusion and classification for multispectral high resolution WorldView-2, TerraSAR-X SpotLight, and elevation model.