LAUSR

The methods based on processing video data have shown their efficiency in many areas of agriculture and forestry. Nevertheless, they are not enough accurate for classification of barely discernible objects and types of plants, which can be provided only using hyperspectral sensors. However, such devices by now were expensive and difficult in operation and were used mostly on satellites and manned aircraft. In recent years, the technologies for creating smaller and lighter sensors have been proposed; these technologies are based on the choice of restricted number of spectral intervals and their position at the design stage. They may be used for scientific or commercial goals under field conditions and, in addition, installed on unmanned aerial vehicles. Exemplified by a 220-channel hyperspectral image, this paper investigates the capability of significant reduction in the amount of registered data due to choice of position and width of restricted number of most informative spectral channels in solving the classification problem of agricultural plants. It is shown that the applied method for generating feature systems has a considerable advantage against the regular decimation and is close in its efficiency to the methods based on the analysis of principal components, but has a significantly less amount of needed computations.