Wilt oak trees detection using hyperspectral images with neural networks and GTI method

In this work I proposed methods for detecting wilt oak trees. There were used images regarding Japanese Oak forests captured during summer and autumn of years 2007, 2008 and 2009, from airborne and from low-altitude. Artificial Neural Network approach was the first adopted and obtained results were not reliable as expected from other past studies. On the other hand, GTI method permitted of splitting into 2 subsets (wilt and healthy), the pixels of an image taking care also about autumnal characteristics of leaves, using reflectance graphs of each pixel in function of wavelength. Two different strategies were followed: finding a static threshold with many observations or running an algorithm (MCC) to find a dynamic threshold. The second approach was better for both dark and subject to accentuated atmospheric effects images. Moreover it resulted effective also for other goals, like detecting wilt rice fields and classifying different grass species in a certain area.