On the Potentiality of UAV Multispectral Imagery to Detect Oak Wilt Disease Using Convolutional Neural Network

초록

Oak wilt disease, a fungal disease caused by Raffaelea quercus-mongolicae, has been a serious threat to oak trees which is the second dominant species in Korea. Despite the huge efforts to manage oak wilt disease, damaged area by the disease have been continuously reported since 2004. As the early symptom, the leaves begin to wilt and the leaves quickly discolor into the reddish brown as the disease progresses. Thus, it is critical to detect the locations of damaged trees as early as possible to prevent further spreading into healthy trees. Unmanned aerial vehicle (UAV) is becoming an effective alternative for monitoring forest disease. UAV provides very high spatial resolution imagery to detect individual damaged trees since it could be operated at very low altitudes. With the development of sensor technology, various types of UAV image have become available, ranging from natural color to multispectral or hyperspectral imagery. Convolutional neural network (CNN) is a deep learning algorithm optimized for image processing because it uses multiple convolution filters that can generate various spatial features. CNN has shown outstanding performance in object detection and classification of image. In remote sensing area, CNN also has been effectively used for detecting tree damaged tree and tree species (ex: oil palm, shrub, and weed) and have shown improved performance than machine learning.