Future Environments: Naira Hovakimyan Discusses How Imaging and AI Can Aid Farmers

Naira Hovakimyan, a professor of mechanical science and engineering, discusses how technology such as drones and artificial intelligence can help farmers grow more food.

Naira Hovakimyan, a professor of mechanical science and engineering and a member of the Cognition, Lifespan Engagement, Aging, and Resilience Group, discusses how drones, planes, satellites, and artificial intelligence will help farmers meet the food needs of a growing world population.

What is "crop loss" and why is it a problem?

Crop loss happens due to not taking timely action. For example, in April when farmers start planting, we have a lot of rain that leads to loss of vegetation. (Some seeds get lost or blown away.) Early imagery within the first two weeks can show areas where one can replant and reverse the crop loss that would have occurred because of missing seeds. Crop loss happens also because of weeds, as those compete for the same water and nutrients under the soil. If weeds are not extracted in timely manner, then they consume the nutrients and spread widely causing more damage. These are just two examples.

Why is it important to modernize farming and bring technology into the fields?

The crop loss is an estimated $28 billion in the U.S. and $110 billion in the world for just corn and soybeans in 2017. In 2050, we will have 9 billion people to feed that with today’s farming technology would fall short. Taking timely actions to prevent weed spread, help with missing vegetation, strategic timing of the harvest, and many other important parameters, can ensure more food is produced from every acre.

How does taking pictures of farms from the air help farmers on the ground?

Aerial photos provide early alerts about problems in the field that if addressed in a timely manner can ensure higher yields per acre. The alerts provide information to farmers on problems such as missing vegetation, weeds, double planting, nutrient deficiency, watersheds, and many other anomalies.

You first started imaging farms with drones. Why did you switch to airplanes?

We differentiate between high volume crops and high value crops. When we deal with high volume crops in the Midwest such as corn and soybeans, it is not optimal to image these farms with drones, as drones do not fly for a very long length of time. (They need to be charged every 15-20 minutes.) Also, we need more sensors for a sophisticated analysis that the payload of drones can't provide (more payload consumes more charge). Satellites do not have thermal cameras, which are critical in farming. With airplanes, we are able to identify the right altitude, where the planes can fly, and collect the data with all the sensors that we need to provide the specific analytics needed by farmers.

Why is it difficult to use artificial intelligence to identify and categorize images of plants?

It is difficult because the anomalies are not similar. When we categorize people, cats, or dogs, then all people are like each other (they all have the same body parts, same pair of eyes, ears, etc.). The same is true for cats and dogs. The anomalies in the farms are not alike. The weeds spread differently. Some of the double planting equipment errors cannot be distinguished from weeds in aerial images. The ground truthing is very much needed by agronomists to understand and differentiate between images that are very much alike. The corresponding statistical analysis has to be very sophisticated.

In addition to reducing crop loss, what other advances do you think will be necessary to meet the world’s food needs in the future?

I strongly believe that robotic technologies will need to be more developed in order to take over the farm work in many instances to give the farmers time to think over other important issues for productivity. I also believe that biotech is essential to enable development of new crops, seeds, and other grains as well. The biochemical engineering can help with better fertilizers and nutrients to support the crop growth.