Fluviator

Inspiration

In the US, the price of eggs has doubled since last year, after an outbreak of avian influenza wiped out 50 million chickens, racking up some $1.6 billion in ever-increasing damages. A previous outbreak in 2014/2015 also had death tolls of 55 million chickens, with direct economic damages of $3.3 billion.

This multibillion-dollar figure doesn't even factor downstream impacts of the outbreaks, which include widespread factory shutdowns, increased unemployment, and heightened food security in lower-income communities. The health risks also aren’t limited to just farms— human-to-human infection of the avian flu would cause economic losses of USD $2 trillion. The poultry industry also contributes 790 million tons of CO2 emissions each year; widespread avian flu and massive cullings cost GHG emissions on the order of 400 million tons of CO2-equivalent emissions.

What it does

Fluviator empowers farmers to monitor bird flu and potential zoonotic outbreaks in agriculture, building a more healthy and sustainable food system. Most outbreaks start with foreign vectors: specifically, non-flock birds (e.g. ducks, geese, seagulls) that wander near the flock and shed the virus. FluFinder uses autonomous drone technology to patrol the region where chickens are allowed to roam, identifying red flags/infection threats from anomalous birds. (This active detection can be coupled with effective sanitation of contaminated areas and pre-emptive quarantine of chickens.) After our specialized computer vision models have detected any potential infection threats on a farm, we send an alert notification to the farmer through our mobile app, in addition to a picture of the suspicious scene. This is especially critical for ethical and environmentally-friendly husbandry practices that allow chickens space to live healthily, such as cage-free, free-range, and pasture-raised egg production. Investigations show that several million birds have been restricted from outdoor access in order to curb the spread of bird flu from wild bird vectors.

Making the Business Use Case

The global poultry market itself is a $350 billion market with an estimated 10% compounded annual growth rate. (Expected value of the market is $487 billion in just 5 years.) Within that, livestock insurance is a market of about $6 billion, and the global animal healthcare market is $177 billion. Farmer willingness-to-pay for a product like this should match the amount of damages that could be prevented; considering that recent outbreaks have cost upwards of $3 billion to farmers, and several billions more to government agencies, this puts the value of anti-influenza initiatives at some $5 billion in annual spending.

Due to the extreme virulence and rapid spread of modern strains of bird flu, most solutions developed in the industry are reactive rather than proactive. This is what has allowed outbreaks on the order of several hundred million chickens globally. In the US, the common response is that outbreaks are quite slow to be detected; only after a large number of birds have died can farmers notice disease, at which point their options are to try containing the cases through quarantine or mass culling of entire farms. In addition, huge additional costs are incurred down the economic chain, with shutdowns of processing plants and much higher prices for consumers. Even the process of health checks may spread disease because the virus spreads easily through contaminated equipment and clothing. A sanitary, remote monitoring system for preventative measures and early detection is critical to solving the flu crisis-- our technology needs to improve as viral strains become more pathogenic.

How we built it

On the hardware side, we made use of Parrot’s ANAFI AI, the first 4G connected robotic UAV Drone. ANAFI AI comes with 4G internet connectivity, autonomous photogrammetry designed for large-scale Mapping, and very-high resolution imagery (48MP). The drone’s live video capture is uploaded to a remote file directory at a rate of 1 FPS, where we then take each still image frame/snapshot as input to an object detection algorithm. We fine-tuned the pre-trained YOLOv5 object detection algorithm, for the purpose of identifying wild (non-chicken) birds in farm contexts. We trained the model with PyTorch and deploy the model for inference by hosting the endpoint on Modal. When a wild bird is detected in any of the still-frames we read in from our drone, then we pass this image to an Android app along with a notification for the farmer.

To keep the UI simple and key information easily accessible, we built the front-end using the Flutter framework with streamlined versions between iOS and Android. We kept the number of screens minimal, showing 1) an overview of flu risk events and 2) details about a specific logged incident for a farmer’s flock (time, farm location, number of birds impacted). Data for the application is passed directly from the Modal instance through webhooks, to minimize unnecessary API calls compared to a conventional REST API.

What's next for Fluviator

Fluviator isn’t limited to just prevention– we also would like to help farmers with early detection of cases. Top officials in the USDA and CDC have cited that early detection of avian flu is the most critical step for containing outbreaks. Next up for Fluviator, we will incorporate thermal imaging on our drones: thermal imaging has been shown in the peer-reviewed literature to be able to detect bird flu infections before any other modes of testing. Fluviator drones will monitor the flock and flag infected, febrile chickens for immediate quarantine, thus allowing the earliest, and most effective, strategy for the containment of isolated cases before they can spread. This will further help to reduce the occurrence of mandated mass-euthanasia incidents on large poultry farms.