Texture classification applied on aerial imagery in forestry
Feature extraction from aerial images is an important research topic with a wide area of applications, like traffic or agriculture monitoring, natural disaster early warning system, etc. Because, to deal successfully with, information gained by remote sensing is by several factors much more cost effective compared to manually accomplished measurements. Certain Information is actually only available by remote sensing since, as the area under investigation is as huge, that any other means would be infeasible. The extraction of certain objects from aerial images has proven to be a very difficult problem especially if the investigated objects do not have sharply bounded lines. This is common the regular case when dense forests are concerned during image analysis. Here, trees usually occlude each other and are hard to differentiate from epiginous vegetation, that make geometrical approaches of object identification, as well as direct representations, hard to apply. On the other hand, counting's performed in the frequency domain offer the advantage of transformation invariance and suffer lesser from diffuse object boundaries. The determination of a clear signature is difficult, if the objects in question are quite similar, though. Hence the paper suggests an approach derived from texture classification that achieves better results in the forestry. © ECMS, 2005.