2 Mar 2022

Modernizing Environmental Signature Physics for Target Detection—Phase 3

Abstract: The present effort (Phase 3) builds on our previously published prior efforts (Phases 1 and 2), which examined methods of determining the probability of detection and false alarm rates using thermal infrared for buried object detection. Environmental phenomenological effects are often represented in weather forecasts in a relatively coarse, hourly resolution, which introduces concerns such as exclusion or misrepresentation of ephemera or lags in timing when using this data as an input for the Army’s Tactical Assault Kit software system. Additionally, the direct application of observed temperature data with weather model data may not be the best approach because metadata associated with the observations are not included. As a result, there is a need to explore mathematical methods such as Bayesian statistics to incorporate observations into models. To better address this concern, the initial analysis in Phase 2 data is expanded in this report to include (1) multivariate analyses for detecting objects in soil, (2) a moving box analysis of object visibility with alternative methods for converting FLIR radiance values to thermal temperature values, (3) a calibrated thermal model of soil temperature using thermal IR imagery, and (4) a simple classifier method for automating buried object detection.

16 Sep 2021

Natural Language Indexing for Pedoinformatics

Abstract: The multiple schema for the classification of soils rely on differing criteria but the major soil science systems, including the United States Department of Agriculture (USDA) and the international harmonized World Reference Base for Soil Resources soil classification systems, are primarily based on inferred pedogenesis. Largely these classifications are compiled from individual observations of soil characteristics within soil profiles, and the vast majority of this pedologic information is contained in non-quantitative text descriptions. We present initial text mining analyses of parsed text in the digitally available USDA soil taxonomy documentation and the Soil Survey Geographic database. Previous research has shown that latent information structure can be extracted from scientific literature using Natural Language Processing techniques, and we show that this latent information can be used to expedite query performance by using syntactic elements and part-of-speech tags as indices. Technical vocabulary often poses a text mining challenge due to the rarity of its diction in the broader context. We introduce an extension to the common English vocabulary that allows for nearly-complete indexing of USDA Soil Series Descriptions.