Pig-finding Thermal Drones

The Problem

The Galapagos Islands are a unique place which hosts creatures such as giant tortoises and is considered the birthplace of the theory of evolution. However, this UNESCO World Heritage Site has been declared as under threat due to increasing invasive species and tourism. Galapagos most iconic inhabitants and name givers of the Galapagos, Galapagos giant tortoises, have seen a 90% reduction in population over the last 180 years. Two species of giant tortoises from Santa Cruz Island—Chelonoidis porteri and C.donfaustoi—were downgraded to critically endangered by IUCN in 2018. Thus, giant tortoises are projected to go extinct in the wild if no actions are taken. This is because feral pigs find and dig out tortoise nests destroying entire generations of Galapagos tortoises. A recent pilot study performed by Diego and colleagues found that 50% of tortoise nests were destroyed by pigs (Half of the future population!). During the nesting season, park rangers go through challenging terrain to find and kill pigs and identify tortoise nests, which are protected with wire mesh. This is an arduous task that involves thousands of man hours annually and makes protecting nests a race against time (or against pigs). If we want to ensure the survival of these iconic species (which are critically important for Galapagos due to their ecosystem services and for tourism), we must [1] know where and when feral pigs are close to tortoise nesting sites and [2] find tortoise nests before the pigs do.

Our Proposal

We plan to attach a thermal camera to a drone (DJI phantom/matrice) so we can find feral pigs and tortoise nesting sites from the air. Pigs, normally shy and hard to find by foot, will be found from above by their thermal signature. Specifically, we will use computer vision and machine learning techniques to recognize, count, and identify pigs and tortoise nests by thermal signature. Our team member, Jonathan, has extensive experience using deep learning for scene segmentation and object recognition. It would take a team of five park rangers several days to identify tortoise nests, but by combining our algorithm with drone imaging this could be done in hours. We will share locations of pigs and nests with park rangers daily during the tortoise nesting seasons. Rangers will then go to marked nesting coordinates and protect nests. With technology, luck, and a caring society, tortoise hatchlings can grow up and hopefully remain alive in 2200! Local collaborators will be trained in drone piloting and our workflow after a successful pilot project by Diego in 2019. We will create a dataset of validated thermal images of nests, pigs and humans and upload it into Imagenet (a database for classified images for machine learning). Our algorithm will be open source and made publicly available. This dataset would be useful for conservationists to identify invasive species or even poachers.

We Assume that...

1) A close dialogue between the Galapagos National Park (GNP) monitoring authorities remains with Diego

2) Demand for monitoring tortoise nests remains extremely high and park rangers are available to go into the field and protect nests

3) GNP remains interested in adopting new technologies assisting with eradicating invasive and aiding conservation. GNP recently adopted drones to monitor the marine reserve with great success

4) Computer vision algorithm can distinguish thermal signature of pigs from signatures of nests and other. We will start training and refining our algorithm with local farm animals in USA

5) GNP will allow Diego to fly drones in the nesting sites of Galapagos tortoises in Santa Cruz in the future (Permits have been ongoing to fly drones in Santa Cruz since 2017)

6) The GNP has technical capacity to take over our methodology. GNP has >300 employers, many highly trained and GNP invested heavily in capacity building in GIS

7) Weather conditions are favorable for drones (Limited heavy rain/wind). We are confident that weather forecast will help us decide when to fly

8) The larger conservation arena interested in drones will be interested in our work combining computer vision, drones and thermal imaging using low cost technologies

9) Our drone based method does not affect the biological community in a negative way and even curbs the need to destroying endemic plants

Constraints to Overcome

Our project will help find invasive pigs and stop nest predation of endemic giant tortoises. Our breakthrough is due to a combination of things; attributed to the right timing and maturation of technology and a due to our technical knowledge and combined expertise. 1) Drones are slowly going mainstream in conservation and thermal cameras and drones are accessible to many practitioners at affordable prices 2) Regulatory constraints on drones in Galapagos are strict and we are one of few projects using drones and the only one monitoring tortoise nests 4) Our combined expertise Diego has on the ground experience in Galapagos, knows the study site, local animals, pressing conservation problem and experience piloting drones while Jonathan handles computer vision and will make our tools open source and public available. Finally, because our project is tailored towards such a specific conservation problem, this problem-despite pressings-has not been addressed to date.

Current Work

Primary activities in the next 6 months are: a) Fly a drone with a thermal camera above local farms in Connecticut to obtain thermal signatures of pigs, goats, humans, cats, dogs (asking farm owners “May I fly my thermal drone above these pigs”). This will reduce error while in the Galapagos b) Jon trains the algorithm with obtain images and refines it c) Submit additional grant proposals to scale up this project d) Approach conservation groups working with Drones and the makerspace to exchange ideas and learn from the “tribe” e) Diego stays in close contact with park authorities and rangers (email, whatsApp) f) Make youtube videos of our setup and describing our methods, to share with the community g) In August, Diego flies to Galapagos (peak tortoise nesting season). He will perform a pilot project, fly over nesting sites and evaluate success rate by estimating accuracy rates of drones identifying pigs and nests and evaluating how much time is saved through our method

Current Needs

We will label our data with the assistance of undergraduate research assistant or perhaps even a service such as Amazon Mechanical Turk or Zooniverse. This will provide us with the required annotations for further analysis of the thermal images Regarding resources, funding will allow us to purchase a drone (DJI phantom for instance) and a thermal camera (FLIR duo or FLIR Vue Pro - ideal for DJI drones), and accesories to maximize flight time. Diego already has funding to fly and housing in Galapagos, so this will not be needed. In terms of knowledge, we are always looking forward to discussing, learning and refining our ideas from the digital makerspace community and the rest of the tribe Regarding skills, we are always happy to hear suggestions as we are sure that a more diverse team will have a wider range of ideas and solutions to our proposed problem. Especially someone with expertise on photogrammetry would be be very useful for this project.