Early Detection of Pine Needle Diseases in Southeast US Forests: A Deep Learning Approach Using UAV Imagery

Fungal diseases pose a significant threat to the health of forests in the Southeast United States, and brown spot needle blight (Mycosphaeralla dearnessii) is emerging as a substantial threat to the pine species of the region. This paper explores the application of deep learning techniques, i.e., Single Shot Detector (SSD) and RetinaNet algorithms, to detect diseased and dead pine trees in the Knight property, which is managed by Kingwood Forest Services. The combination of unmanned aerial vehicles (UAV) with advanced image processing methods, such as SSD and RetinaNet algorithm, helped accurately identify tree stands that were either diseased or dead. Drone images were acquired for two different years i.e., 2022 and 2023. For each image, both diseased and dead trees were identified and used in training the algorithm to detect both conditions. For the year 2022, the SSD model achieved an accuracy of 67% for diseased trees and 75% for dead trees, while RetinaNet only reached 59% and 30% accuracy. For the year 2023, the SSD algorithm again maintained higher accuracy with 74% for both diseased and dead trees. Conversely, the RetinaNet algorithm continued to provide a lower accuracy rate at 35% for diseased trees and 53% for dead trees. The study also looked at the trend in diseased and dead trees from both years and the result shows that there was an increase in the number of diseased trees in the year 2022, from 192 to 737 in the year 2023. Furthermore, there was an increase in the number of dead trees in 2022, from 172 to 551 in 2023. Utilizing the SSD model based on deep neural architectures has shown promising results in accurately detecting diseased and dead pine trees within the forest observed in this experiment. Overall, the results of the study highlight the effectiveness of deep learning algorithms in monitoring and managing pine disease outbreaks in forest ecosystems; this capability would help in providing valuable information, such as the number of diseased and dead trees with their precise location, to enhance disease detection and forest management strategies in the Southeast United States.