Pine trees, which account for 21.9 percent of Korea’s forest area as of 2013, began to lose their occupancy area significantly as of the following year. PWD(Pine Wilt Disease) is blamed for the first outbreak in Busan in 1988. it is so deadly that i...
Pine trees, which account for 21.9 percent of Korea’s forest area as of 2013, began to lose their occupancy area significantly as of the following year. PWD(Pine Wilt Disease) is blamed for the first outbreak in Busan in 1988. it is so deadly that it is called pine tree AIDS, and after the infection, it die at 100 percent within 20 days or as long as three months.
The PWD were damaged not only in Gyeongsangnam-do and Gyeongsangbuk-do but also in Gyeonggi-do and Gangwon-do, therefore Feeling serious, the government enacted the SPECIAL Activation on the Exterme on PINE WILT DISEASE in 2005 by banning pine trees from moving to affected areas and implementing massive controls.
However, although the basic methods of control currently in place involve a large amount of budget and manpower, problems exist in detecting PWD in a wide range of areas and in places such as hills, steep slopes and cliffs that are inaccessible to people. In addition, it is difficult to observe the damaged area quickly and accurately because it occurs as a dot in the early stages of the damage.
Recently, the government has been making efforts to detect pine trees affected by pine tree reoccurrence by using satellite images or remote exploration technology, but various problems have arisen in terms of smooth cooperation from related agencies, irregularity in data collection cycles and uniformity of data depending on weather conditions. Therefore, research is under way on how to make up for these problems, but it has yet to suggest effective ways to control PWD.
Therefore, in this study, the detection method for PWD was proposed using UAV(Unmanned Aerial Vehicle) and Deep-Learning to solve the problems of the existing prevention.
First, the target site selected Mt. Mabong in Pohang where area with pine trees and capable of field survey and aerial shooting.
The UAV used for aerial shooting used a fixed-wing UAV equipped with Multi-spectral cameras and high-resolution cameras, and produced NDVI, RGB, and high-resolution RGB ortho-mosaics, respectively, using the collected data, for analysis.
According to the distinct of the PWD affected by data, 520 NDVI and RGB respectively and 198 high-resolution RGB were observed.
In order to determine the existence of pine trees affected by the disease, an on-site survey was conducted except for places such as steep slopes and cliffs that were inaccessible, and as a result, the NDVI, 249 RGB, and 84 high-resolution RGB were identified.
Analysis was performed using SegNet, the method of Semantic Segmentation, and YOLOv2, the method of Object Detection, to propose algorithms suitable for automatic detection of PWD.
The analysis that the performance of SegNet was 0.47 NDVI, 0.49 RGB and 0.57 high resolution RGB, while the detection performance of YOLOv2 was 0.49 NDVI, 0.53 RGB, and 0.77 high resolution RGB, making YOLOv2 better overall, and particularly effective in detecting high resolution RGB.
Through this study, PWD were detected. However, aerial shooting was difficult to analyze, as the analysis results showed that the flightability and the analysis results were different depending on the season and weather condition. Also, in the analysis of deep-learning, land, buildings, dry grass, and maple trees that look similar to PWD were recognized as damaged trees, which is one of the important points of lowering detection rates.
It is expected that if various types of research are added to conduct a deep-learning analysis that can utilize unrestricted data for weather condition in the future, it will reduce the manpower, cost, and time required to prevent damage from the existing PWD and obtain a quick detection effect.