Version 1
: Received: 2 October 2024 / Approved: 3 October 2024 / Online: 3 October 2024 (11:23:39 CEST)
How to cite:
Luo, W.; Scott, B.; Cadogan, J.; Combs, T. P.; Swann, D. E.; Didan, K.; Li, H. Saguaro Recognition from Drone Imagery Using Mask R-CNN in Detectron2. Preprints2024, 2024100230. https://doi.org/10.20944/preprints202410.0230.v1
Luo, W.; Scott, B.; Cadogan, J.; Combs, T. P.; Swann, D. E.; Didan, K.; Li, H. Saguaro Recognition from Drone Imagery Using Mask R-CNN in Detectron2. Preprints 2024, 2024100230. https://doi.org/10.20944/preprints202410.0230.v1
Luo, W.; Scott, B.; Cadogan, J.; Combs, T. P.; Swann, D. E.; Didan, K.; Li, H. Saguaro Recognition from Drone Imagery Using Mask R-CNN in Detectron2. Preprints2024, 2024100230. https://doi.org/10.20944/preprints202410.0230.v1
APA Style
Luo, W., Scott, B., Cadogan, J., Combs, T. P., Swann, D. E., Didan, K., & Li, H. (2024). Saguaro Recognition from Drone Imagery Using Mask R-CNN in Detectron2. Preprints. https://doi.org/10.20944/preprints202410.0230.v1
Chicago/Turabian Style
Luo, W., Kamel Didan and Haiquan Li. 2024 "Saguaro Recognition from Drone Imagery Using Mask R-CNN in Detectron2" Preprints. https://doi.org/10.20944/preprints202410.0230.v1
Abstract
The saguaro cactus (Carnegiea gigantea) plays a pivotal role in desert ecosystems, making its population monitoring essential. Traditional census methods used by the United States Forestry Service are resource-intensive, prompting a need for more cost-effective alternatives. Automated detection methods using advanced object detection models applied to drone imagery present a promising solution. In a proof-of-concept study, 244 drone images of saguaros were captured from a top-down perspective over an undeveloped hill adjacent to Sun Ray Park, Phoenix, Arizona (33.3188° N, 111.9980° W), from altitudes of 486, 507, and 519 meters above mean sea level (AMSL). We employed the Mask R-CNN model from the Detectron2 framework for model training. The images were divided into training, validation, and test sets in an approximate 8:1:1 ratio, with the sets separated by their physical locations within the park: training data was centralized, validation data was positioned to the east, and test data was located in the west. The Mask R-CNN achieved an average precision of 89.8% and an average F1 score of 90.3% in identifying saguaros across 27 test images from 486/507 m AMSL, demonstrating the model's effectiveness in accurately identifying saguaro cacti. Despite the limited sample size, the model's adaptability to diverse scenarios underscores its potential for practical applications in ecological conservation. This research contributes to the field of automated monitoring by offering a viable alternative to labor-intensive methods, thus supporting the sustainability of saguaro in their native habitat.
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
