YOLO: A FUSION APPROACH TO GEOLOCATING NATURAL RESOURCES
Advancing near real-time AI and drone technology to improve wildlife monitoring and conservation in hard-to-reach landscapes.
Processing Drone Data
Data collected by UAS surveys has become an incredible resource for wildlife researchers and managers, however downloading, processing, and storing data can be time-consuming and expensive . Additionally, delays between data collection and processing can hinder the ability to make time-sensitive decisions regarding management actions. In addition to the need to develop more reliable and cost-effective monitoring methods for difficult to survey wildlife species, there is also a need to be able to process this data quickly and accurately.
Artificial intelligence (AI) has recently emerged as a viable option to process images collected during UAS flights in near real-time. Specifically, the computer vision model “You Only Look Once”, or YOLO, has demonstrated the ability to quickly process tens of thousands of images to detect and classify wildlife . In particular, it has shown promise in its ability to detect nests and determine nest status of avian species . YOLO, however, is still an emerging technology and much work is needed to refine and improve its object detection abilities.
Workflow of real-time classification process using imagery collected from UAS flights using YOLO computer vision model.
In this project we will be advancing this technology by incorporating not only visual images, but also thermal images and LiDAR data to optimize YOLO detection of nest locations and occupancy in near-real time. The incorporation of additional data sources aims to increase detection accuracy and allow precise geographic 3D locations of nests . The end result would be the ability to identify Golden Eagle nests, determine their occupancy, and document their exact location all in near real-time so they can be validated by personnel in the field, all while reducing time in the field.There is also ample potential for this technology, once refined, to be used for other target species or natural resource challenges that have traditionally been difficult to monitor. This includes other cliff nesting species, such as prairie falcons, wetland nesting birds, arctic species, or species that occur in other difficult to reach areas.
Platforms used to collect imagery, transmit data, and run detection model.