2024 Volume 106 Issue 2 Pages 31-36
Mapping and monitoring of tree seedlings are crucial tasks in silvicultural management. Unmanned aerial vehicles (UAVs), in recent years, have been used as a cost-effective technique to capture high-resolution images. However, the method is still labor-intensive and time-consuming if you use with manual monitoring techniques. For continuous utilization of UAV images in silvicultural management, novel automated monitoring methods are required. This study investigated the potential application of automatic tree seedling detection method based on deep-learning (DL) by comparing the accuracy between images acquired in different seasons. The results showed that high seedlings detection accuracy of about 90% can be achieved with the DL-based method. Moreover, as automated methods are easily influenced by weed growth, a decrease in the number of detected tree seedlings and an increase in the number of misdetections were observed during seasons of weed growth, such as summer and autumn. These results suggested that aerial images should be obtained during unfavorable weedy growing conditions to enhance the accuracy of automated methods.