Classification of high-mountain plant communities using artificial neural nets and hyperspectral data

摘要

The paper presents results of plant communities mapping of an alpine and subalpine zones of the Tatra National Park (southern part of the Polish Carpathian Mts.), located within a range of altitudes of 1500–2549 m a. s. l. Classification algorithm based on the hyperspectral DAIS 7915 imagery and the fuzzy ARTMAP (FAM) neural networks simulator of 2 key polygons (Biesnik and Uchrocie Kasprowe) using training sets of 40 original bands (after geometric and atmospheric correction) and 20 MNF bands (derived from 60 preselected DAIS 7915 channels). The results of 37 plant communities were compared with the reference sets acquired from ground validation. The best overall accuracy (87%) for the test set was achieved using 40 original bands bands.