Methods of using self-organising maps for terrain classification, using an example of developing a military passability map

Abstract

The classification of terrain by its passability plays a significant role in the process of Intelligence Preparation of the Battlefield (IPB). In the process of developing passability maps, the classification of terrain to a specific class (GO, SLOW-GO, NO-GO). In this paper the problem of terrain classification to the respective category of passability was solved by the application of Self Organizing Maps by generating a continuous Index of Passability (IOP), which characterizes the terrain in a range from 0 (the impassable area) to 1 (the area of high manoeuvrability). The article describes the methodology of using this type of network to develop a terrain passability map. As a “case of use”, three voivodeships located in the north-eastern part of Poland were selected. To prepare a training set, topographic vector data from VMap L2 and SRTM (Shuttle Radar Topography Mission) digital terrain model were used. Research was conducted on a primary grid field with dimensions 1 km 1 km. As a result of the research conducted, normalised parameters associated with terrain cover were introduce into the neural network. As a result of the network learning, the analysed area was divided into classes, to which the index of passability (IOP) was arbitrarily subordinated. In the research results, the influence of the method of organisation of the input data on the generated maps of passability was defined. The tests were conducted on two sizes of a Kohonen map: 10 10 and 5 5 neurons. The described experiments proved that a properly taught artificial neural network is very well suited to the analysis of an area in terms of passability. The presented methodology is universal in nature and after the modification of parameters may be used to solve tasks of terrain classification associated with various subjects (division of soils, marking out areas for development, etc.).