Fusing Terrain and Sensor Data During Spectral Feature Extraction

Terrain data needs on the battlefield are generating greater military requirements for new, rapid feature mapping capabilities. Unfortunately, current methods are woefully inadequate to meet the needs of the modern digitized battlefield in which thecommander must have the ability to rapidly obtain accurate terrain information and topographic products. Terrain Reasoning methods will allow more rapid and accurate labeling of terrain features by making use of multisensor/multisource data, bothnumerical (e.g. spectral data) and non-numerical (e.g. land-use maps). These methods can improve automated feature extraction by fusing disparate data sources while reducing operational costs. Building on existing Terrain Reasoning approaches, we proposeto combine methods from two areas, the Theory of Evidence and the Knowledge-Based Processing/Decision-Trees, to allow fusion and rapid processing of disparate data sources. The proposed Trade Study will focus on three measures of success: improvedaccuracy, increased speed, and flexibility of the model to accept new data sources. We will develop a classification methodology constituting the most cost-effective approach for improving current supervised spectral feature extraction techniques. Theresulting prototype software will be completely compatible and scalable for use within the Army's Combat Terrain Information Systems (CTIS) program.The anticipated results are a prototype software module that provides rapid fusion of disparate data sources(e.g. DTM, radar data, land use/land cover maps, etc.) for processing and extracting spectral feature information. This software will be developed entirely from COTS/GOTS systems and will be fully compatible with the Army's Combat Terrain InformationSystems (CTIS) program and well as other commercial image processing software packages (e.g. ENVI and ERDAS). In general, the final product will provide an enhanced spectral feature extraction tool for DoD and commercial image processing packages.