SELECTION OF A SYSTEM OF INFORMATIVE FEATURES FOR CROP CLASSIFICATION USING HYPERSPECTRAL DATA
S. M. Borzov, O. I. Potaturkin
Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, Novosibirsk, Russia
Keywords: дистанционное зондирование Земли, гиперспектральные изображения, классификация сельскохозяйственных культур, выбор информативных признаков, remote sensing, hyperspectral images, crop classification, selection of informative features
Abstract
Methods based on video data processing have demonstrated their effectiveness in many areas of agriculture and forestry. However, they lack classification accuracy for objects and vegetation that can only be achieved by using hyperspectral sensors, but such devices have been quite expensive and difficult to operate until recently; they have been mainly used on satellites and manned aircraft. In recent years, technologies have been proposed for creating more compact and lightweight sensors based on selection of a limited number of spectral intervals and their positions at the design stage. These sensors can be used for scientific or commercial purposes in field conditions; in particular, they can be mounted on drones. In this paper, by an example of 220-channel hyperspectral image processing, we experimentally investigate the possibility of significantly reducing the volume of recorded data by selecting the position and width of a limited number of the most informative spectral channels in solving the problem of crop classification. It is shown that the method used to form the system of informative features has a significant advantage over decimation and is close in efficiency to methods based on the principal component analysis, having a significantly lower required computational complexity.
|
