Drone Aerial Image Identification of Tropical Forest Tree Species using the Mask R-CNN

Abstract

Tropical forests have a wide variety of species and support environmental activities. The drone's image resolution is 90% more accurate than satellite data. It boosted productivity, safety, and the capacity to make better decisions by comparing archived and prospective images. Labeling tree species in heavily forested locations is labor-intensive, time-consuming, and expensive. This research seeks to design a new model for classifying tree species based on drone imagery, then test and assess its effectiveness. This study shows that drone technology can diminish productivity per hectare compared to conventional ground approaches. The study shows drones are more productive than ground approaches. The approach is feasible since it targets commercial timber species in the forest's higher stratum. Drones are cheaper than satellite data, therefore they're being used more in forest management and deep learning. Drones allow flexible, high-resolution data collection. This research uses Mask R-CNN to recognize and segment trees. This study uses high-resolution RGB images of tropical forests. The mAP, recall, and precision all performed well. Our suggested method yields a solid prediction model for detecting tree species, validated by 75% of ground truth data. This strategy can help plan and execute forest inventory, as shown. This initiative's success may lead to the first phase of a forest inventory, affecting the region's logging and forest management.