Detecting Forest Damage in Cir Aerial Photographs Using a Neural Network

Abstract

Forest dieback is taking on increasing proportions in many parts of Croatia. To improve the situation, it is of primary importance to acquire timely, accurate and inexpensive information on the scale of forest damage. Such information can be collected for large forest areas with remote sensing techniques. This paper explores the possibility of applying segmentations of color infrared aerial photographs (CIR). Self-organizing artificial neural networks are used to detect damage in beech-fir forests and determine its spatial distribution. The results of the research confirm the benefits of applying neural networks to forest damage detection, since there are no statistically significant differences between damage in the field and damage detected with a neural network.